2-train_model.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train model\n",
    "\n",
    "Dataset \n",
    "16840 entries of business news\n",
    "\n",
    "in 5 categories by how negative/positive for given entity\n",
    "\n",
    "Entity is a company name, with given location within text. Can be several entities in single article.\n",
    "\n",
    "headline is also given."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import ast\n",
    "import pandas as pd\n",
    "import html\n",
    "import numpy as np\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "import sys\n",
    "sys.path.append(\"../fastaiDL/fastai/courses/dl2\")\n",
    "from fastai.text import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.6\n"
     ]
    }
   ],
   "source": [
    "import fastai\n",
    "print(getattr(fastai, '__version__', 0.6))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings('ignore')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA = 'data/'\n",
    "file = 'polarity_data.js' # business news\n",
    "datafile = DATA+file\n",
    "\n",
    "LM_PATH=Path('work/lm/')\n",
    "LM_PATH.mkdir(exist_ok=True)\n",
    "CLAS_PATH=Path('work/clas/')\n",
    "CLAS_PATH.mkdir(exist_ok=True)\n",
    "PATH=Path('.')\n",
    "CLASSES = ['veryneg', 'neg', 'neut', 'pos', 'very pos']\n",
    "\n",
    "#tags\n",
    "XBOB = ' xbob ' # beginning-of-body tag\n",
    "#ENTBEG = 'xentb '  # beginning-of-entity in text\n",
    "#ENTEND = ' xente'  # end of entity in text\n",
    "ENT =  'ENTITY '   # replace target entity in the text\n",
    "\n",
    "# '_entity_' ended up as 3 tokens '_', 'entity', '_'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "with open(datafile, encoding='utf-8') as data_file:\n",
    "    data = ast.literal_eval(data_file.read())\n",
    "df = pd.DataFrame.from_dict(data, orient='columns')"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "# extract headline start and end\n",
    "df = pd.concat( [df, df['headline'].apply(pd.Series)], axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Randomize"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "np.random.seed(1)\n",
    "texts_idx = np.random.permutation(len(df))\n",
    "df = df.iloc[texts_idx]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LM tokenize"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "clas_file = 15718 # by maxsize, see filename in PATH CLAS_PATH if need\n",
    "lm_file = 14840\n",
    "\n",
    "trainfile =   'train'+str(lm_file)+'.csv'\n",
    "testfile  = 'test_of'+str(lm_file)+'.csv'\n",
    "\n",
    "chunksize=24000\n",
    "BOS = 'xbos'  # beginning-of-sentence tag\n",
    "FLD = 'xfld'  # data field tag"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train = pd.read_csv(LM_PATH/trainfile, header=None, chunksize=chunksize)\n",
    "df_validate = pd.read_csv(LM_PATH/testfile, header=None, chunksize=chunksize)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "import re, html\n",
    "re1 = re.compile(r'  +')\n",
    "\n",
    "def fixup(x):\n",
    "    x = x.replace('#39;', \"'\").replace('amp;', '&').replace('#146;', \"'\").replace(\n",
    "        #'nbsp;', ' ').replace('#36;', '$').replace('\\\\n', \"\\n\").replace('quot;', \"'\").replace(\n",
    "        # '\\n' replaced with ' '\n",
    "        'nbsp;', ' ').replace('#36;', '$').replace('\\\\n', ' ').replace('quot;', \"'\").replace(\n",
    "        '<br />', \"\\n\").replace('\\\\\"', '\"').replace('<unk>','u_n').replace(' @.@ ','.').replace(\n",
    "        ' @-@ ','-').replace('\\\\', ' \\\\ ')\n",
    "    return re1.sub(' ', html.unescape(x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_texts(df, n_lbls=1):\n",
    "    labels = df.iloc[:,range(n_lbls)].values.astype(np.int64)\n",
    "    texts = f'\\n{BOS} {FLD} 1 ' + df[n_lbls].astype(str)\n",
    "    for i in range(n_lbls+1, len(df.columns)): texts += f' {FLD} {i-n_lbls} ' + df[i].astype(str)\n",
    "    texts = list(texts.apply(fixup).values)\n",
    "\n",
    "    tok = Tokenizer().proc_all_mp(partition_by_cores(texts))\n",
    "    return tok, list(labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_all(df, n_lbls):\n",
    "    tok, labels = [], []\n",
    "    for i, r in enumerate(df):\n",
    "        print(i)\n",
    "        tok_, labels_ = get_texts(r, n_lbls)\n",
    "        tok += tok_;\n",
    "        labels += labels_\n",
    "    return tok, labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n",
      "0\n"
     ]
    }
   ],
   "source": [
    "tok_train, trn_labels = get_all(df_train, 1)\n",
    "tok_validate, val_labels = get_all(df_validate, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.save(LM_PATH/'tmp'/'tok_train.npy', tok_train)\n",
    "np.save(LM_PATH/'tmp'/'tok_validate.npy', tok_validate)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "tok_train = np.load(LM_PATH/'tmp'/'tok_train.npy')\n",
    "tok_validate = np.load(LM_PATH/'tmp'/'tok_validate.npy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('the', 349664),\n",
       " (',', 294419),\n",
       " ('.', 287418),\n",
       " ('\\n', 182449),\n",
       " ('to', 181482),\n",
       " ('of', 153664),\n",
       " ('and', 144620),\n",
       " ('in', 143840),\n",
       " ('a', 134955),\n",
       " ('-', 79642),\n",
       " ('t_up', 77125),\n",
       " ('for', 66610),\n",
       " ('that', 60775),\n",
       " ('on', 59742),\n",
       " ('$', 53449),\n",
       " ('is', 53366),\n",
       " ('it', 48670),\n",
       " ('\"', 46549),\n",
       " ('its', 46457),\n",
       " ('with', 43788),\n",
       " (\"'s\", 41296),\n",
       " ('said', 41261),\n",
       " ('company', 39228),\n",
       " ('as', 37325),\n",
       " ('by', 36956)]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "freq = Counter(p for o in tok_train for p in o)\n",
    "freq.most_common(25)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_vocab = 60000\n",
    "min_freq = 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "itos = [o for o,c in freq.most_common(max_vocab) if c>min_freq]\n",
    "itos.insert(0, '_pad_')\n",
    "itos.insert(0, '_unk_')\n",
    "\n",
    "#itos.insert(0, '_pad_')\n",
    "itos.insert(0, '<unk>')\n",
    "\n",
    "# WAS it <unk> in the data, not _unk_?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# insert for ENTITY (target entity)\n",
    "itos.insert(0, 'ENTITY')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "34915"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "stoi = collections.defaultdict(lambda:0, {v:k for k,v in enumerate(itos)})\n",
    "len(itos)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_lm = np.array([[stoi[o] for o in p] for p in tok_train])\n",
    "validate_lm = np.array([[stoi[o] for o in p] for p in tok_validate])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.save(LM_PATH/'tmp'/'trn_ids.npy', train_lm)\n",
    "np.save(LM_PATH/'tmp'/'val_ids.npy', validate_lm)\n",
    "pickle.dump(itos, open(LM_PATH/'tmp'/'itos.pkl', 'wb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# EXTRA\n",
    "# STORE THIS! Needed for prediction\n",
    "pickle.dump(itos, open(LM_PATH/'tmp'/'itosMax.pkl', 'wb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_lm = np.load(LM_PATH/'tmp'/'trn_ids.npy')\n",
    "validate_lm = np.load(LM_PATH/'tmp'/'val_ids.npy')\n",
    "itos = pickle.load(open(LM_PATH/'tmp'/'itos.pkl', 'rb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(34915, 14840)"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vocsize=len(itos)\n",
    "vocsize,len(train_lm)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "14840"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(tok_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Wikitext-103 conversion"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Our source LM is the wikitext103 LM created by Stephen Merity @ Salesforce research. Link to dataset The language model for wikitext103 (AWD LSTM) has been pre-trained and the weights can be downloaded here: Link. Our target LM is the IMDB LM. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "# ! wget -nH -r -np -P {PATH} http://files.fast.ai/models/wt103/"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The pre-trained LM weights have an embedding size of 400, 1150 hidden units and just 3 layers. We need to match these values with the target IMDB LM so that the weights can be loaded up."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "emb_size,nhiddenu,nlayers = 400,1150,3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "PosixPath('.')"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "PATH=Path('.')\n",
    "PATH"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "PRE_PATH = PATH/'models'/'wt103'\n",
    "PRE_LM_PATH = PRE_PATH/'fwd_wt103.h5'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "#PRE_PATH = 'models/wt103'\n",
    "#PRE_LM_PATH = 'models/wt103/fwd_wt103.h5'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "PosixPath('models/wt103/fwd_wt103.h5')"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "PRE_LM_PATH"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "wgts = torch.load(PRE_LM_PATH, map_location=lambda storage, loc: storage)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We calculate the mean of the layer0 encoder weights. This can be used to assign weights to unknown tokens when we transfer to target IMDB LM."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "enc_wgts = to_np(wgts['0.encoder.weight'])\n",
    "row_m = enc_wgts.mean(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "itos2 = pickle.load((PRE_PATH/'itos_wt103.pkl').open('rb'))\n",
    "stoi2 = collections.defaultdict(lambda:-1, {v:k for k,v in enumerate(itos2)})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Before we try to transfer the knowledge from wikitext to the IMDB LM, we match up the vocab words and their indexes. We use the defaultdict container once again, to assign mean weights to unknown IMDB tokens that do not exist in wikitext103."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "new_w = np.zeros((vocsize, emb_size), dtype=np.float32)\n",
    "for i,w in enumerate(itos):\n",
    "    r = stoi2[w]\n",
    "    new_w[i] = enc_wgts[r] if r>=0 else row_m"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now overwrite the weights into the wgts odict. The decoder module, which we will explore in detail is also loaded with the same weights due to an idea called weight tying."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "wgts['0.encoder.weight'] = T(new_w)\n",
    "wgts['0.encoder_with_dropout.embed.weight'] = T(np.copy(new_w))\n",
    "wgts['1.decoder.weight'] = T(np.copy(new_w))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.cuda.FloatTensor"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# only extra print info\n",
    "# Signature: T(a, half=False, cuda=True)\n",
    "# Docstring: Convert numpy array into a pytorch tensor\n",
    "\n",
    "a = wgts['0.encoder.weight']\n",
    "type(a)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Language model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "wd=1e-7 #weight decay\n",
    "bptt=70 # back propagation through time\n",
    "# was 52 but out of GPU mem\n",
    "# with 32 seems to use 7,36 GB \n",
    "# bs=32\n",
    "#bs = 25\n",
    "#bs = 52 # batch size\n",
    "bs=72\n",
    "bs=100\n",
    "bs=32\n",
    "optim_fn = partial(optim.Adam, betas=(0.8, 0.99))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The goal of the LM is to learn to predict a word/token given a preceeding set of words(tokens). We take all the movie reviews in both the 90k training set and 10k validation set and concatenate them to form long strings of tokens. In fastai, we use the LanguageModelLoader to create a data loader which makes it easy to create and use bptt sized mini batches. The LanguageModelLoader takes a concatenated string of tokens and returns a loader.\n",
    "\n",
    "We have a special modeldata object class for LMs called LanguageModelData to which we can pass the training and validation loaders and get in return the model itself.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_dataloader = LanguageModelLoader(np.concatenate(train_lm), bs, bptt)\n",
    "validate_dataloader = LanguageModelLoader(np.concatenate(validate_lm), bs, bptt)\n",
    "model = LanguageModelData(PATH, 1, vocsize, train_dataloader, validate_dataloader, bs=bs, bptt=bptt)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We setup the dropouts for the model - these values have been chosen after experimentation. If you need to update them for custom LMs, you can change the weighting factor (0.7 here) based on the amount of data you have. For more data, you can reduce dropout factor and for small datasets, you can reduce overfitting by choosing a higher dropout factor. No other dropout value requires tuning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "# dropouts\n",
    "#drops = np.array([0.25, 0.1, 0.2, 0.02, 0.15])*0.7\n",
    "drops = np.array([0.25, 0.1, 0.2, 0.02, 0.15])*0.8"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We first tune the last embedding layer so that the missing tokens initialized with mean weights get tuned properly. So we freeze everything except the last layer.\n",
    "\n",
    "We also keep track of the *accuracy* metric."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "learner= model.get_model(optim_fn, emb_size, nhiddenu, nlayers, \n",
    "    dropouti=drops[0], dropout=drops[1], wdrop=drops[2], dropoute=drops[3], dropouth=drops[4])\n",
    "\n",
    "learner.metrics = [accuracy]\n",
    "learner.freeze_to(-1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "learner.model.load_state_dict(wgts)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We set learning rates and fit our OWN LM. We first run one epoch to tune the last layer which contains the embedding weights. This should help the missing tokens in the wikitext103 learn better weights."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr=1e-3\n",
    "lrs = lr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "bs=140\n",
    "# LM 140 is ok, 200 is oom (full length LM-articles)\n",
    "# bs 140, 8k truncated 6659 MB -> up to 6965 when proceeding\n",
    "\n",
    "# expected\n",
    "# learner.fit  0.317314\n",
    "# lr_find 0.355"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "c17d1d3718974ec782792ecaa7fe2ff9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                     \n",
      "    0      4.453638   4.120457   0.300652  \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([4.12046]), 0.3006517675098957]"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learner.fit(lrs/2, 1, wds=wd, use_clr=(32,2), cycle_len=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note that we print out accuracy and keep track of how often we end up predicting the target word correctly. While this is a good metric to check, it is not part of our loss function as it can get quite bumpy. We only minimize cross-entropy loss in the LM.\n",
    "\n",
    "The exponent of the cross-entropy loss is called the perplexity of the LM. (low perplexity is better).\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "learner.save('lm_last-ent_ft')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [],
   "source": [
    "learner.load('lm_last-ent_ft')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "learner.unfreeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "#bs=140"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5ab07eea092242e49eda821e3cc2e4a8",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                     \n",
      "    0      4.095493   3.788278   0.338212  \n",
      "\n"
     ]
    }
   ],
   "source": [
    "learner.lr_find(start_lr=lrs/10, end_lr=lrs*10, linear=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learner.sched.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.001"
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lrs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9403bc36731446749bb525e696aca9bb",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=20), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                     \n",
      "    0      3.635672   3.312772   0.395224  \n",
      "    1      3.220798   2.92392    0.449953                     \n",
      "    2      2.997155   2.690247   0.488311                     \n",
      "    3      2.810141   2.549778   0.513016                     \n",
      "    4      2.687036   2.458257   0.530001                     \n",
      "    5      2.626963   2.389082   0.544225                     \n",
      "    6      2.553965   2.338485   0.554421                     \n",
      "    7      2.611864   2.292894   0.561937                     \n",
      "    8      2.440274   2.256401   0.569852                     \n",
      "    9      2.43019    2.224718   0.576048                     \n",
      "    10     2.392038   2.202989   0.580578                     \n",
      "    11     2.383107   2.172906   0.58659                      \n",
      "    12     2.276034   2.15677    0.591451                     \n",
      "    13     2.305274   2.136978   0.594777                     \n",
      "    14     2.228293   2.121025   0.598528                     \n",
      "    15     2.219782   2.103678   0.602048                     \n",
      "    16     2.157073   2.091512   0.604977                     \n",
      "    17     2.217391   2.084229   0.606461                     \n",
      "    18     2.161222   2.07643    0.609006                     \n",
      "    19     2.193371   2.069207   0.610115                     \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([2.06921]), 0.6101147709810227]"
      ]
     },
     "execution_count": 68,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bs=140\n",
    "lrs =1e-3\n",
    "learner.fit(lrs, n_cycle=1, wds=wd, use_clr=(20,10), cycle_len=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "c75f2ec6a4d14d3e95e562ced563e218",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=20), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                     \n",
      "    0      3.451486   3.219849   0.406877  \n",
      "    1      3.476988   3.22243    0.406604                     \n",
      "    2      3.365643   3.12433    0.421181                     \n",
      "    3      3.204469   3.02028    0.436237                     \n",
      "    4      3.136014   2.944028   0.448716                     \n",
      "    5      3.074043   2.873266   0.461102                     \n",
      "    6      2.972078   2.798478   0.473559                     \n",
      "    7      2.865905   2.739218   0.483731                     \n",
      "    8      2.770412   2.683193   0.493587                     \n",
      "    9      2.79299    2.627315   0.503411                     \n",
      "    10     2.651105   2.577286   0.51316                      \n",
      "    11     2.57571    2.531726   0.521849                     \n",
      "    12     2.517409   2.488049   0.529755                     \n",
      "    13     2.456295   2.446648   0.538016                     \n",
      "    14     2.480263   2.408098   0.545173                     \n",
      "    15     2.385134   2.367422   0.553632                     \n",
      "    16     2.368415   2.336312   0.559832                     \n",
      "    17     2.238819   2.305802   0.565981                     \n",
      "    18     2.313814   2.280693   0.570566                     \n",
      "    19     2.218202   2.262762   0.574868                     \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([2.26276]), 0.5748678171737294]"
      ]
     },
     "execution_count": 165,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# after splitting entities, first main try, .59+ is goal, data randomized, entities not marked in LM\n",
    "# 20 reaches 0.57 and still increasing fast 0.57 to 0.574\n",
    "# batchsize 200 worked for 4k truncated\n",
    "bs=140 \n",
    "lrs =5e-3\n",
    "learner.fit(lrs, n_cycle=1, wds=wd, use_clr=(20,10), cycle_len=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# after splitting entities, first main try, .59+ is goal\n",
    "lrs =5e-3\n",
    "learner.fit(lrs, n_cycle=1, wds=wd, use_clr=(20,10), cycle_len=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "402e6f179b7440a885f527f5988a0eb2",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=15), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                     \n",
      "    0      3.574274   3.2758     0.399256  \n",
      "    1      3.500602   3.235741   0.404575                     \n",
      "    2      3.353847   3.093312   0.425605                     \n",
      "    3      3.221653   2.985482   0.442458                     \n",
      "    4      3.124829   2.893923   0.456953                     \n",
      "    5      3.026518   2.815494   0.469902                     \n",
      "    6      2.926473   2.736441   0.483996                     \n",
      "    7      2.852153   2.670463   0.496062                     \n",
      "    8      2.799434   2.600488   0.507748                     \n",
      "    9      2.706199   2.541887   0.518628                     \n",
      "    10     2.633005   2.487245   0.528666                     \n",
      "    11     2.568277   2.439431   0.538178                     \n",
      "    12     2.492155   2.39227    0.547058                     \n",
      "    13     2.427679   2.358953   0.554006                     \n",
      "    14     2.39076    2.329921   0.560096                     \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([2.32992]), 0.5600963007129652]"
      ]
     },
     "execution_count": 82,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# trunc 4k, maybe not good, repeating same data instead of new\n",
    ".learner.fit(lrs, n_cycle=1, wds=wd, use_clr=(20,10), cycle_len=15)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "learner.save('lm1-full-noent-061')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "learner.save_encoder('lm1-full-noent061_enc')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We save the trained model weights and separately save the encoder part of the LM model as well. This will serve as our backbone in the classification task model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [],
   "source": [
    "#learner.save('lm1-trunc4k12k-ent')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [],
   "source": [
    "#learner.save_encoder('lm1-trunc4k12k-ent_enc')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0,0.5,'Accuracy')"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x864 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "epochs = len(learner.sched.val_losses)\n",
    "fig,ax = plt.subplots(2,1,figsize=(8,12))\n",
    "ax[0].plot(list(range(epochs)),learner.sched.val_losses, label='Validation loss')\n",
    "ax[0].plot(list(range(epochs)),[learner.sched.losses[i] for i in range(epochs)], label='Training loss')\n",
    "ax[0].set_xlabel('Epoch')\n",
    "ax[0].set_ylabel('Loss')\n",
    "ax[0].legend(loc='upper right')\n",
    "\n",
    "ax[1].plot(list(range(epochs)),learner.sched.rec_metrics)\n",
    "ax[1].set_xlabel('Epoch')\n",
    "ax[1].set_ylabel('Accuracy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learner.sched.plot_loss()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Classifier Tokens"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The classifier model is basically a linear layer custom head on top of the LM backbone. Setting up the classifier data is similar to the LM data setup except that we cannot use the unsup movie reviews this time.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "trainsize = 12000\n",
    "devsize = 3000\n",
    "trainfile = 'train'+str(trainsize)+'.csv'\n",
    "devfile = 'test'+str(devsize)+'.csv'\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# extra if start here\n",
    "chunksize=24000\n",
    "BOS = 'xbos'  # beginning-of-sentence tag\n",
    "FLD = 'xfld'  # data field tag"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# fix for filename to save\n",
    "trainsize = 15718"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# clasfile len = 15718\n",
    "trainfile = 'train15718entity.csv'\n",
    "devfile = 'test_of15718entity.csv'\n",
    "\n",
    "df_train = pd.read_csv(CLAS_PATH/trainfile, header=None, chunksize=chunksize)\n",
    "df_validate = pd.read_csv(CLAS_PATH/devfile, header=None, chunksize=chunksize)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<pandas.io.parsers.TextFileReader at 0x7f412b4300b8>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n",
      "0\n"
     ]
    }
   ],
   "source": [
    "tok_train, train_labels = get_all(df_train, 1)\n",
    "tok_validate, validate_labels = get_all(df_validate, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "(CLAS_PATH/'tmp').mkdir(exist_ok=True)\n",
    "\n",
    "np.save(CLAS_PATH/'tmp'/'tok_trn.npy', tok_train)\n",
    "np.save(CLAS_PATH/'tmp'/'tok_val.npy', tok_validate)\n",
    "\n",
    "np.save(CLAS_PATH/'tmp'/'trn_labels.npy', train_labels)\n",
    "np.save(CLAS_PATH/'tmp'/'val_labels.npy', validate_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "tok_train = np.load(CLAS_PATH/'tmp'/'tok_trn.npy')\n",
    "tok_validate = np.load(CLAS_PATH/'tmp'/'tok_val.npy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "34915"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "itos = pickle.load((LM_PATH/'tmp'/'itos.pkl').open('rb'))\n",
    "stoi = collections.defaultdict(lambda:0, {v:k for k,v in enumerate(itos)})\n",
    "len(itos)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_clas = np.array([[stoi[o] for o in p] for p in tok_train])\n",
    "validate_clas = np.array([[stoi[o] for o in p] for p in tok_validate])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.save(CLAS_PATH/'tmp'/'trn_ids.npy', train_clas)\n",
    "np.save(CLAS_PATH/'tmp'/'val_ids.npy', validate_clas)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "list"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(train_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[array([3]), array([0]), array([0]), array([1])]"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_labels[0:4]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[<matplotlib.axes._subplots.AxesSubplot object at 0x7f41025a6fd0>]], dtype=object)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#train_labels\n",
    "#hist(train_labels\n",
    "\n",
    "histog = pd.DataFrame({'labels': train_labels})\n",
    "# unpack\n",
    "histog['labels'] = histog.apply(lambda row: row['labels'][0], axis=1)\n",
    "histog.hist(column='labels')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_clas = np.load(CLAS_PATH/'tmp'/'trn_ids.npy')\n",
    "validate_clas = np.load(CLAS_PATH/'tmp'/'val_ids.npy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_labels = np.squeeze(np.load(CLAS_PATH/'tmp'/'trn_labels.npy'))\n",
    "validate_labels = np.squeeze(np.load(CLAS_PATH/'tmp'/'val_labels.npy'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### optional - turn into BINARY CLASSIFICATION"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "# negative vs positive + neutra  \n",
    "# 0 and 1 -> 0\n",
    "# 2,3 and 4 -> 1\n",
    "\n",
    "binary_train_labels = [0 if (x==0 or x==1) else 1 for x in train_labels]\n",
    "binary_validate_labels = [0 if (x==0 or x==1) else 1 for x in validate_labels]\n",
    "\n",
    "# remember to do same for test set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "binary_train_labels"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Optional - truncate documents"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "trun_len = 80  #words\n",
    "trun_len = 200\n",
    "\n",
    "for i in range(len(train_clas)):\n",
    "    train_clas[i] = train_clas[i][0:trun_len]\n",
    "    \n",
    "for i in range(len(validate_clas)):\n",
    "    validate_clas[i] = validate_clas[i][0:trun_len]\n",
    "    \n",
    "assert len(train_clas[0]) == trun_len\n",
    "assert len(validate_clas[0]) == trun_len"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "bptt,emb_size,nhiddenu,nlayers = 70,400,1150,3\n",
    "vocsize = len(itos)\n",
    "optim_fn = partial(optim.Adam, betas=(0.8, 0.99))\n",
    "#bs = 26\n",
    "#bs = 32  # 7175 MB memory usage on business news data at final training, this with 13k length articles\n",
    "# 32 too high\n",
    "bs = 14"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "bs=12\n",
    "bs=25"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "min_lbl = train_labels.min()\n",
    "train_labels -= min_lbl\n",
    "validate_labels -= min_lbl\n",
    "c=int(train_labels.max())+1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the classifier, unlike LM, we need to read a movie review at a time and learn to predict the it's sentiment as pos/neg. We do not deal with equal bptt size batches, so we have to pad the sequences to the same length in each batch. To create batches of similar sized movie reviews, we use a sortish sampler method invented by @Smerity and @jekbradbury\n",
    "\n",
    "The sortishSampler cuts down the overall number of padding tokens the classifier ends up seeing.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_ds = TextDataset(train_clas, train_labels)\n",
    "validate_ds = TextDataset(validate_clas, validate_labels)\n",
    "train_samp = SortishSampler(train_clas, key=lambda x: len(train_clas[x]), bs=bs//2)\n",
    "validate_samp = SortSampler(validate_clas, key=lambda x: len(validate_clas[x]))\n",
    "train_dl = DataLoader(train_ds, bs//2, transpose=True, num_workers=1, pad_idx=1, sampler=train_samp)\n",
    "validate_dl = DataLoader(validate_ds, bs, transpose=True, num_workers=1, pad_idx=1, sampler=validate_samp)\n",
    "model = ModelData(PATH, train_dl, validate_dl)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([3, 0, 0, 1, 4, 3, 3, 1, 3, 0])"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_labels[0:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "dps = np.array([0.4, 0.5, 0.05, 0.3, 0.1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "m = get_rnn_classifier(bptt, 20*70, c, vocsize, emb_sz=emb_size, n_hid=nhiddenu, n_layers=nlayers, pad_token=1,\n",
    "          layers=[emb_size*3, 50, c], drops=[dps[4], 0.1],\n",
    "          dropouti=dps[0], wdrop=dps[1], dropoute=dps[2], dropouth=dps[3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "optim_fn = partial(optim.Adam, betas=(0.7, 0.99))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = RNN_Learner(model, TextModel(to_gpu(m)), opt_fn=optim_fn)\n",
    "learn.reg_fn = partial(seq2seq_reg, alpha=2, beta=1)\n",
    "learn.clip=25.\n",
    "learn.metrics = [accuracy]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr=3e-3\n",
    "lrm = 2.6\n",
    "lrs = np.array([lr/(lrm**4), lr/(lrm**3), lr/(lrm**2), lr/lrm, lr])\n",
    "\n",
    "# THESE NOT USED, NEXT OVERWRITES"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "wd = 1e-7\n",
    "wd = 0\n",
    "# WEIGHT DECAY stays ALWAYS 0 this way!\n",
    "\n",
    "#learn.load_encoder('lm1-12k-ent_enc')\n",
    "#learn.load_encoder('lm1-full-noent061_enc')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.freeze_to(-1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "70f8b9002598431188dd55f88abb61dc",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 78%|███████▊  | 1020/1310 [02:18<00:39,  7.35it/s, loss=5.6] "
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.lr_find(lrs/1000)\n",
    "learn.sched.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.0001, 0.0001, 0.0001, 0.001 , 0.01  ])"
      ]
     },
     "execution_count": 85,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lrs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "1fec50982fd04b6298b9b8bb9ebf9815",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                     \n",
      "    0      1.431822   1.436869   0.315     \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([1.43687]), 0.31499999957159164]"
      ]
     },
     "execution_count": 86,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lrs, 1, wds=wd, cycle_len=1, use_clr=(8,3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('clas_0'+'_trainsize_'+str(trainsize))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('clas_0'+'_trainsize_'+str(trainsize))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.freeze_to(-2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "33560bf650da4f2082d393b7e6a51c3f",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                     \n",
      "    0      1.415951   1.438869   0.315     \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([1.43887]), 0.31499999957159164]"
      ]
     },
     "execution_count": 90,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lrs, 1, wds=wd, cycle_len=1, use_clr=(8,3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('clas_1'+'_trainsize_'+str(trainsize))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('clas_1'+'_trainsize_'+str(trainsize))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.unfreeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "bd33b012d8ea4f4cb2f5cbf93bba31eb",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 79%|███████▉  | 1032/1310 [04:31<01:13,  3.80it/s, loss=5.04]"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.lr_find(lrs/1000)\n",
    "learn.sched.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "51013540fc9e43f6b555d1ad7ae270e3",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=20), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                     \n",
      "    0      1.025954   1.017071   0.5575    \n",
      "    1      0.888767   0.772047   0.6595                        \n",
      "    2      0.721386   0.673798   0.7315                        \n",
      "    3      0.589075   0.677841   0.748                         \n",
      "    4      0.483754   0.478499   0.8225                        \n",
      "    5      0.476519   0.428138   0.845                         \n",
      "    6      0.405554   0.462406   0.851                         \n",
      "    7      0.324407   0.379418   0.8835                        \n",
      "    8      0.319293   0.378683   0.871                         \n",
      "    9      0.37745    0.3294     0.892                         \n",
      "    10     0.321874   0.388887   0.8675                        \n",
      "    11     0.333175   0.338951   0.8815                        \n",
      "    12     0.230425   0.320582   0.8955                        \n",
      "    13     0.268168   0.326834   0.894                         \n",
      "    14     0.267637   0.330666   0.889                         \n",
      "    15     0.223525   0.322936   0.895                         \n",
      "    16     0.237719   0.312673   0.8995                        \n",
      "    17     0.173687   0.318363   0.893                         \n",
      "    18     0.172206   0.313613   0.9015                        \n",
      "    19     0.185096   0.319949   0.9035                        \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([0.31995]), 0.903499997407198]"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 1.11 noENT on LM, ENT on clas\n",
    "wd=1e-7 #1e-6?\n",
    "bs=200  # try with truncated 200 words\n",
    "lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*1 \n",
    "#lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*2 ?\n",
    "learn.fit(lrs, n_cycle=1, wds=wd, cycle_len=20, use_clr=(10,20))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "dc795580a83d4ca3ad7f4a899811ee78",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=20), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                     \n",
      "    0      1.080745   1.154619   0.468     \n",
      "    1      0.912616   0.827696   0.6725                        \n",
      "    2      0.803956   0.725042   0.7115                        \n",
      "    3      0.660619   0.666951   0.7435                        \n",
      "    4      0.629606   0.552529   0.7945                        \n",
      "    5      0.595529   0.518047   0.812                         \n",
      "    6      0.480156   0.519156   0.8205                        \n",
      "    7      0.48899    0.514283   0.8215                        \n",
      "    8      0.365413   0.487099   0.83                          \n",
      "    9      0.4077     0.450899   0.8265                        \n",
      "    10     0.350877   0.443739   0.8295                        \n",
      "    11     0.395255   0.444448   0.8315                        \n",
      "    12     0.302502   0.432657   0.837                         \n",
      "    13     0.317949   0.464915   0.833                         \n",
      "    14     0.307394   0.449634   0.8355                        \n",
      "    15     0.300865   0.50981    0.828                         \n",
      "    16     0.272787   0.448975   0.836                         \n",
      "    17     0.269424   0.444318   0.8275                        \n",
      "    18     0.270619   0.454351   0.835                         \n",
      "    19     0.273345   0.453143   0.826                         \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([0.45314]), 0.8260000012814999]"
      ]
     },
     "execution_count": 112,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# was wrong, not ENT after all in clas data!\n",
    "# 1.11 noENT on LM, ENT on clas\n",
    ".\n",
    "wd=1e-7 #1e-6?\n",
    "bs=200  # try with truncated 200 words\n",
    "lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*1 \n",
    "#lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*2 ?\n",
    "learn.fit(lrs, n_cycle=1, wds=wd, cycle_len=20, use_clr=(10,20))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "270ac3eda48e48b48e0df782dc4a0f90",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=20), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                      \n",
      "    0      0.901914   0.737277   0.707     \n",
      "    1      0.843598   0.689649   0.742                         \n",
      "    2      0.785138   0.650254   0.7515                        \n",
      "    3      0.820657   0.626869   0.7535                        \n",
      "    4      0.766361   0.607883   0.773                         \n",
      "    5      0.715389   0.583788   0.784                         \n",
      "    6      0.781313   0.562963   0.7895                        \n",
      "    7      0.755657   0.559893   0.787                         \n",
      "    8      0.714867   0.550014   0.7935                        \n",
      "    9      0.693643   0.525307   0.804                         \n",
      "    10     0.705955   0.525282   0.804                         \n",
      "    11     0.671646   0.515538   0.8095                        \n",
      "    12     0.670022   0.50731    0.8145                        \n",
      "    13     0.678187   0.530717   0.8155                        \n",
      "    14     0.658997   0.502076   0.817                         \n",
      "    15     0.650093   0.515668   0.8115                        \n",
      "    16     0.667668   0.49834    0.8205                        \n",
      "    17     0.608823   0.504686   0.816                         \n",
      "    18     0.646065   0.48398    0.8215                        \n",
      "    19     0.651114   0.491967   0.8215                        \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([0.49197]), 0.821500001102686]"
      ]
     },
     "execution_count": 95,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 31.10 ENT on LM, ENT on clas\n",
    "wd=1e-7 #1e-6?\n",
    "bs=200  # try with truncated 200 words\n",
    "lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*1 \n",
    "#lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*2 ?\n",
    "learn.fit(lrs, n_cycle=1, wds=wd, cycle_len=20, use_clr=(10,20))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 246,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "0b42a28f1dd64b23a2f2859218899166",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=40), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                      \n",
      "    0      0.863732   0.673801   0.726333  \n",
      "    1      0.859651   0.676968   0.738333                     \n",
      "    2      0.778679   0.613633   0.757333                      \n",
      "    3      0.805469   0.589531   0.770333                      \n",
      "    4      0.771135   0.583629   0.777                         \n",
      "    5      0.690103   0.547105   0.788333                     \n",
      "    6      0.668815   0.543315   0.785333                      \n",
      "    7      0.64111    0.514515   0.805                         \n",
      "    8      0.678837   0.516527   0.797667                      \n",
      "    9      0.617058   0.486839   0.818                         \n",
      "    10     0.590986   0.491506   0.818667                      \n",
      "    11     0.573794   0.46893    0.821333                      \n",
      "    12     0.560596   0.472524   0.827333                      \n",
      "    13     0.538717   0.509057   0.819667                      \n",
      "    14     0.564312   0.462762   0.830667                      \n",
      "    15     0.509481   0.472528   0.824                         \n",
      "    16     0.482257   0.456902   0.84                          \n",
      "    17     0.497391   0.45432    0.834667                      \n",
      "    18     0.455137   0.455425   0.841                         \n",
      "    19     0.456449   0.453677   0.842667                      \n",
      "    20     0.421485   0.457388   0.842                         \n",
      "    21     0.425155   0.439947   0.846667                      \n",
      "    22     0.423556   0.445944   0.845667                      \n",
      "    23     0.438487   0.448922   0.845667                      \n",
      "    24     0.391038   0.439213   0.848                         \n",
      "    25     0.409661   0.449376   0.845333                     \n",
      "    26     0.406825   0.450119   0.849                         \n",
      "    27     0.401583   0.461765   0.845667                      \n",
      "    28     0.376793   0.454938   0.848333                      \n",
      "    29     0.388907   0.432229   0.853333                      \n",
      "    30     0.347631   0.445655   0.851333                      \n",
      "    31     0.326592   0.46281    0.852                         \n",
      "    32     0.33616    0.437167   0.85                          \n",
      "    33     0.352244   0.442012   0.857                         \n",
      "    34     0.338883   0.446027   0.853667                      \n",
      "    35     0.297821   0.447373   0.853                         \n",
      "    36     0.338237   0.453068   0.851667                      \n",
      "    37     0.34669    0.456658   0.854667                      \n",
      "    38     0.327242   0.444479   0.852667                      \n",
      "    39     0.283071   0.446997   0.854333                     \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([0.447]), 0.8543333311875662]"
      ]
     },
     "execution_count": 246,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 25.10, split entities, randomized, LM has not ENT-tag. CLAS data does have.\n",
    "# no trunc on LM\n",
    "\n",
    "# 16 epoch reaches 0.84, #33 is max 0.857\n",
    "wd=1e-7 #1e-6?\n",
    "bs=200\n",
    "lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*1 \n",
    "#lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*2 ?\n",
    "learn.fit(lrs, n_cycle=1, wds=wd, cycle_len=40, use_clr=(10,40))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.sched.plot_lr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 203,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7c049ca57ce7440b9efb7cbd1f3bd7e4",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=50), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                    \n",
      "    0      0.894836   0.727825   0.725     \n",
      "    1      0.867559   0.724577   0.707                       \n",
      "    2      0.874823   0.818662   0.7375                      \n",
      "    3      0.810144   2.448367   0.743                       \n",
      "    4      0.724066   0.737688   0.748                       \n",
      "    5      0.738162   0.709561   0.7605                      \n",
      "    6      0.649582   0.564669   0.7895                      \n",
      "    7      0.648562   0.559181   0.796                       \n",
      "    8      0.576893   0.545653   0.7945                      \n",
      "    9      0.586251   0.553139   0.807                       \n",
      "    10     0.566973   0.5306     0.8135                      \n",
      "    11     0.58191    0.54444    0.8195                      \n",
      "    12     0.509779   0.53532    0.819                       \n",
      "    13     0.481554   0.522785   0.8245                      \n",
      "    14     0.482669   0.586675   0.82                        \n",
      "    15     0.495074   0.551627   0.8165                      \n",
      "    16     0.454293   0.629963   0.829                       \n",
      "    17     0.414201   0.523589   0.825                       \n",
      "    18     0.411066   0.522083   0.835                       \n",
      "    19     0.389517   0.530466   0.8275                      \n",
      "    20     0.402478   0.504084   0.838                       \n",
      "    21     0.402106   0.512166   0.841                       \n",
      "    22     0.377341   0.506326   0.8285                      \n",
      "    23     0.365389   0.505241   0.8425                      \n",
      "    24     0.305999   0.53103    0.845                       \n",
      "    25     0.352485   0.506474   0.836                       \n",
      "    26     0.325986   0.532253   0.8295                      \n",
      "    27     0.281626   0.559055   0.8355                      \n",
      "    28     0.299721   0.666151   0.844                       \n",
      "    29     0.329575   0.52342    0.8405                      \n",
      "    30     0.250905   0.556687   0.8445                      \n",
      "    31     0.324334   0.5738     0.8465                      \n",
      "    32     0.278031   0.551263   0.8565                      \n",
      "    33     0.286023   0.565723   0.843                       \n",
      "    34     0.261455   0.556461   0.8425                      \n",
      "    35     0.236961   0.571934   0.842                       \n",
      "    36     0.214262   0.584887   0.8425                      \n",
      "    37     0.247358   0.601739   0.8465                      \n",
      "    38     0.231991   0.582968   0.8435                      \n",
      "    39     0.225189   0.572713   0.8385                      \n",
      "    40     0.236481   0.600274   0.845                       \n",
      "    41     0.260849   0.585878   0.8465                      \n",
      "    42     0.218946   0.605059   0.846                       \n",
      "    43     0.195728   0.597881   0.851                       \n",
      "    44     0.2069     0.605019   0.851                       \n",
      "    45     0.185254   0.599619   0.8525                      \n",
      "    46     0.165702   0.595408   0.8545                      \n",
      "    47     0.177576   0.578982   0.849                       \n",
      "    48     0.193969   0.592404   0.8485                      \n",
      "    49     0.205786   0.591041   0.85                        \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([0.59104]), 0.8499999923110009]"
      ]
     },
     "execution_count": 203,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 1cycle\n",
    "# - maximum lr from test\n",
    "# - minimum as 1/10 of max\n",
    "# - raising of lr to be chosen\n",
    "# - often big bs and higher lr preferred\n",
    "\n",
    "# - momentum, possible to use cyclical momentum\n",
    "# Short runs with momentum values of 0.99, 0.97, 0.95, and 0.9 will quickly show the best value for momentum\n",
    "# lower end: 0.8 or 0.85 (performance is almost independent of the minimum momentum value)\n",
    "\n",
    "wd=1e-7 #1e-6?\n",
    "bs=200\n",
    "lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*1.2 # higher!\n",
    "#lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*2 ?\n",
    "learn.fit(lrs, n_cycle=1, wds=wd, cycle_len=50, use_clr=(10,40))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.fit(lrs, n_cycle=4, wds=wd, cycle_len=10, use_clr=(32,10))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Best 0.85\n",
    "wd=1e-7\n",
    "bs=82\n",
    "learn.fit(lrs, n_cycle=2, wds=wd, cycle_len=10, use_clr=(32,10))\n",
    "learn.fit(lrs, n_cycle=2, wds=wd, cycle_len=10, use_clr=(32,10)) # .8425\n",
    "lrs=np.array([1e-4,1e-4,1e-4,1e-3,1e-2])*0.3\n",
    "learn.fit(lrs, n_cycle=1, wds=wd, cycle_len=10, use_clr=(32,10))\n",
    "learn.fit(lrs, n_cycle=1, wds=wd, cycle_len=10, use_clr=(32,10))\n",
    "bs = 100\n",
    "learn.fit(lrs, n_cycle=1, wds=wd, cycle_len=10, use_clr=(32,10)) #.85"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 178,
   "metadata": {},
   "outputs": [],
   "source": [
    "# bs 150 works, bs 200 works. 250 takes 6067 MB, 280 takes 6715 MB, 300=7,4GB, OBS later 300 gives 8095MB !! \n",
    "# variation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# clr_beta_ ratio 10 for high-low lr, 15% of epochs left for after, max and min momentum.\n",
    "# learn.fit(0.8, 1, cycle_len=95, use_clr_beta=(10, 13.68, 0.95, 0.85), wds=1e-4)\n",
    "wd = 0\n",
    "learn.fit(lrs, 1, wds=wd, cycle_len=15, use_clr_beta=(10, 15, 0.95, 0.85)) # add 0.95, 0.85 to use_clr_beta"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('clas_2_Classifier'+'_fullLM_'+'full')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('clas_2_Classifier'+'_fullLM_'+'full')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    ".learn.save('clas_2_Classifier'+'_ent_trainsize_'+str(trainsize)+'')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.sched.plot_loss()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "26200"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.sched.nb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Predict test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "targetfile = 'final_test.csv'\n",
    "targetfile = 'final_test15718entity.csv'\n",
    "bs = 14\n",
    "bs = 200"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n"
     ]
    }
   ],
   "source": [
    "chunksize=24000\n",
    "BOS = 'xbos'  # beginning-of-sentence tag\n",
    "FLD = 'xfld'  # data field tag\n",
    "df_test = pd.read_csv(CLAS_PATH/targetfile, header=None, chunksize=chunksize)\n",
    "\n",
    "# tokenize\n",
    "tokens_test, test_labels = get_all(df_test, 1)\n",
    "\n",
    "(CLAS_PATH/'tmp').mkdir(exist_ok=True)\n",
    "np.save(CLAS_PATH/'tmp'/'tok_test.npy', tokens_test)\n",
    "np.save(CLAS_PATH/'tmp'/'test_labels.npy', test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2000, 2000)"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# saving and loading changes from list to np.ndarray\n",
    "tokens_test = np.load(CLAS_PATH/'tmp'/'tok_test.npy')\n",
    "len(tokens_test), len(test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "# change strings to integers\n",
    "test_clas = np.array([[stoi[o] for o in p] for p in tokens_test])\n",
    "test_clas = np.expand_dims(test_clas, axis=1)\n",
    "np.save(CLAS_PATH/'tmp'/'test_ids.npy', test_clas)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_clas = np.load(CLAS_PATH/'tmp'/'test_ids.npy')\n",
    "test_clas = np.squeeze(test_clas) \n",
    "test_labels = np.squeeze(np.load(CLAS_PATH/'tmp'/'test_labels.npy'))\n",
    "n_class=int(test_labels.max())+1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "#build TextDataSet\n",
    "test_dataset = TextDataset(test_clas, test_labels)\n",
    "\n",
    "# build sortSampler - mixes labels and items, doesnt work\n",
    "#test_sampler = SortSampler(test_clas, key=lambda x: len(test_clas[x]))\n",
    "\n",
    "# build DataLoader\n",
    "# shuffle mixes labels wrongly\n",
    "#test_loader = DataLoader(test_dataset, bs, transpose=True, num_workers=1, pad_idx=1, sampler=test_sampler)\n",
    "test_loader = DataLoader(test_dataset, bs, transpose=True, num_workers=1, pad_idx=1, \n",
    "                         sampler=None, shuffle=False)\n",
    "\n",
    "# build TextData\n",
    "# TextData or fastai.nlp were not used when training mdoel\n",
    "from fastai.nlp import TextData\n",
    "md = TextData(PATH, None, test_loader)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "bs = 150 # out of memory\n",
    "bs = 100 # maybe ok from empty?, but oom after training\n",
    "bs = 35 # continue after training: 6,1GB usage\n",
    "\n",
    "bptt,em_sz,nhiddenu,nlayers = 70,400,1150,3\n",
    "vocsize = len(itos)\n",
    "optim_fn = partial(optim.Adam, betas=(0.8, 0.99))\n",
    "\n",
    "model = get_rnn_classifier(bptt, 20*70, n_class, vocsize, emb_sz=em_sz, n_hid=nhiddenu, n_layers=nlayers, pad_token=1,\n",
    "          layers=[em_sz*3, 50, n_class], drops=[dps[4], 0.1],\n",
    "          dropouti=dps[0], wdrop=dps[1], dropoute=dps[2], dropouth=dps[3])\n",
    "\n",
    "#model.eval  - automatically called?\n",
    "\n",
    "learner = RNN_Learner(\n",
    "    data=md,\n",
    "    models=TextModel(to_gpu(model)),    \n",
    "    opt_fn=optim_fn\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds_dist, true_labels = learner.predict_with_targs()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Plot Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "%reload_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZIAAAEKCAYAAAA4t9PUAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzt3Xd4VGXax/HvnQbSW1QEpAiIID3SiW0XAhZEUbH3giDFdRX33eK6u65tpQioYF8LIqig0nQtoUNA6SChKFmkaEIxERLgef+YE968MQmTZCYzk/w+1zVXJmfOmbmfTMjNOc85vzHnHCIiIiUVFeoCREQksqmRiIhIqaiRiIhIqaiRiIhIqaiRiIhIqaiRiIhIqaiRiIhIqaiRiIhIqaiRiIhIqcSEuoCyUK9ePdekSZNQlyEiEjFWrlz5o3Mu3p91K0QjadKkCSkpKaEuQ0QkYpjZd/6uq0NbIiJSKmokIiJSKmokIiJSKmokIiJSKmokIiJSKkFtJGaWZGabzSzVzEYX8HglM3vXe3yZmTXxltc1sy/M7Gczm5Bvm85mttbbZryZWTDHICIiRQtaIzGzaGAi0A9oDVxnZq3zrXYHkOGcaw6MAZ70lh8G/gQ8WMBTPw/cDbTwbkmBr15ERPwVzD2SLkCqc26bcy4bmAoMyLfOAOB17/504GIzM+dcpnNuIb6GcoKZ1QdqOOeWON9nBL8BXBHEMZTY2rQDJH+7L9RliIgEXTAbSQNgZ57v07xlBa7jnDsKHADqnuQ5007ynACY2d1mlmJmKfv2lf0f9FHTvuHmV5Yz/J2v2Z+VXeavLyJSVoLZSAqau3AlWKdE6zvnJjvnEpxzCfHxfl3lH1BHjx0HYPbaH+gzJpkvNu8t8xpERMpCMBtJGtAoz/cNgV2FrWNmMUBNIP0kz9nwJM8ZFg4dPsqN3c7kw6E9qVUlltteXcEj76/l5yNHQ12aiEhABbORrABamFlTM4sDBgOz8q0zC7jFuz8I+Nyb+yiQc+4H4JCZdfPO1roZmBn40kvn+HFHRlY2tavEcW6Dmswa1ot7EpsxdcX39BuXzPLtRfVKEZHIErRG4s15DAPmARuBac659Wb2mJld7q32MlDXzFKBB4ATpwib2Q7gWeBWM0vLc8bXEOAlIBXYCswJ1hhK6uDhHI47qF0lDoDKsdE80v8cpt3THcO4dvIS/vHJBg7nHAtxpSIipRfU9F/n3Gxgdr5lf85z/zBwdSHbNilkeQpwbuCqDLyMrBwA6lSN+3/Lz2tShzkjevP47I1MWbCdLzfv49lrOtC2Yc1QlCkiEhC6sj0I0jN9Z2nVztdIAKpWiuEfA9vy+u1dOHg4h4GTFjH2s2/J8SbnRUQijRpJEGTkNpIqsYWuc37LeOaPPJ9L29Vn7GdbuHLSYrbsOVRWJYqIBIwaSRBkZOU2kl/vkeRVs0osYwd35PkbOpGWkcUlzy3kpQXbOH68qDOgRUTCixpJEOQ2kvxzJIXp17Y+80edT2KLeP7+yUYGT1nKzvSsYJYoIhIwaiRBkJ6ZQ1x0FFXiov3eJr56Jabc3JmnB7Vj466DJI1N5p3l31PE2dAiImFBjSQIMjKzqV01luIGE5sZVyc0Yu6oRNo3qsUj76/lttdWsOfg4ZNvLCISImokQZB7MWJJNah1Cm/e0ZVHL2vN0m0/0WdMMrNWh+UF/CIiaiTBUNpGAhAVZdzasymzh/emab2qDH/na4a+verEGWEiIuFCjSQI0jOz/Z5oP5lm8dWYfm93ft/3bOav302fscl8vmlPQJ5bRCQQ1EiCYH9WDrWrFn4NSXHFREcx9MLmzBzai7pV47j9tRQenr6GQ4dzAvYaIiIlpUYSYLmBjXVKeWirIK3PqMHMYT0ZcsFZvLdyJ0ljF7Bk608Bfx0RkeJQIwmw3MDGWkFoJACVYqJ5OKkV793bndho47opS3nsIwVAikjoqJEEWG7OVqDmSArTuXEdZo/ozc3dG/PKou1cMn4Bq3fuD+priogURI0kwHKTfwsKbAy0KnExPDbgXN68oytZ2ce48vnFPDt/M9lHFQApImVHjSTA/AlsDLReLeoxd2QiAzqcwfjPUxk4aRGbdysAUkTKhhpJgKX7GdgYaDVPieXZazrw4k2d2X3gMJc9t5AXv9rKMQVAikiQqZEE2P5iBjYGWt82pzNvVCIXtornn3M2MXjyEr77KTMktYhIxaBGEmDpmTnExRQvsDHQ6lWrxAs3dubZa9qzafch+o1bwJtLv1MApIgEhRpJgGVkZlO7SvEDGwPNzLiyU0PmjUykc+Pa/PHDddzy6gp2H1AApIgElhpJgKUHIGcrkM6odQpv3N6Fvw1ow4rt6fQZ8xUffv1f7Z2ISMCokQTY/qzA5WwFiplxU/cmzB7RmxanVWfku99w31ur+OnnI6EuTUTKATWSAEvPDK89krya1qvKtHu683BSK/6zcS99xybz6QYFQIpI6aiRBFhGgAMbAy06yhhywVnMur8n8dUrc9cbKTz43moOKgBSREpIjSSAjh93vkNbYbpHkler02swc2hPhl3YnPdXpdFv7AIWp/4Y6rJEJAKpkQRQsAMbAy0uJooH+57NjCE9qBQTxfUvLePRWev5JVsBkCLiPzWSACqrwMZA63hmbT4Z3ptbezThtcU7uGT8AlZ9nxHqskQkQqiRBFBGbjxKhDUSgFPionn08ja8fWdXjhw9zqDnF/P0vE0KgBSRk1IjCaCMTN+EdSTMkRSmR/N6zBnZm6s6NWTiF1sZMHERG384GOqyRCSMqZEEUG5gY60yTP4NhhqVY3n66vZMuTmBfYeOcPmEhUz6MlUBkCJSIDWSAMqI0DmSwvy29WnMH5XIb1ufxlNzN3P1C4vZ/qMCIEXk/1MjCaCMrNAHNgZanapxTLy+E+MGdyB178/0H7eAN5bs4Lj2TkTEo0YSQOES2BhoZsaADg2YP+p8zmtahz/PXM/Nryxn1/5fQl2aiIQBNZIACrfAxkA7vWZlXr/tPP4x8FxWfZ9B37HJzFiZpgBIkQpOjSSAMjLDL7Ax0MyMG7o2Zs6I3rQ6vTq/e2819/x7JT8qAFKkwlIjCaCMcr5HklfjulWZend3/tC/FV9u3kffMcnMXbc71GWJSAgEtZGYWZKZbTazVDMbXcDjlczsXe/xZWbWJM9jj3jLN5tZ3zzLR5nZejNbZ2bvmFnlYI6hOMI9sDHQoqOMuxPP4uPhvahfqzL3vrmSB979hgO/KABSpCIJWiMxs2hgItAPaA1cZ2at8612B5DhnGsOjAGe9LZtDQwG2gBJwCQzizazBsBwIME5dy4Q7a0XcpEU2BhoLU+rzgf39WT4xS2YuXoXSWOTWbBlX6jLEpEyEsw9ki5AqnNum3MuG5gKDMi3zgDgde/+dOBi853yNACY6pw74pzbDqR6zwcQA5xiZjFAFWBXEMfgt9zAxkiMRwmE2OgoHvhtS94f0oMqcdHc9PJy/vThOrKyj4a6NBEJsmA2kgbAzjzfp3nLClzHOXcUOADULWxb59x/gWeA74EfgAPOufkFvbiZ3W1mKWaWsm9f8P93nBvYWFHmSArTvlEtPhnemzt6NeXNZd/Rf9wCVn6XHuqyRCSIgtlICrqYIv95ooWtU+ByM6uNb2+lKXAGUNXMbizoxZ1zk51zCc65hPj4+GKUXTKRHNgYaJVjo/nTpa15565uHD3uuPqFJTwxZxNHjiqeXqQ8CmYjSQMa5fm+Ib8+DHViHe9QVU0gvYhtfwNsd87tc87lAO8DPYJSfTGll4PAxkDr1qwuc0cmck1CI174aisDJixi/a4DoS5LRAIsmI1kBdDCzJqaWRy+SfFZ+daZBdzi3R8EfO58V7fNAgZ7Z3U1BVoAy/Ed0upmZlW8uZSLgY1BHIPfMspJYGOgVasUwxNXteOVWxP4KTObKyYuYsLnWzh6TPH0IuVF0BqJN+cxDJiH74/9NOfcejN7zMwu91Z7GahrZqnAA8Bob9v1wDRgAzAXGOqcO+acW4ZvUn4VsNarf3KwxlAc5S2wMdAuanUa80cm0rfN6Twz/1sGvbCErft+DnVZIhIAVhHiLRISElxKSkpQX+Ofczby6qIdbP5bUrnL2gq0j1bv4k8z13E45xgPJ7Xilu5NiIrSz0wknJjZSudcgj/r6sr2ANmfmUOdKnFqIn64rP0ZzB+ZSPdmdfnrRxu48eVlpGVkhbosESkhNZIASc/K1vxIMZxaozKv3HoeT1zZltU795M0dgHTUnYqAFIkAqmRBEhFCGwMNDNjcJczmTsykdZn1OCh6Wu4642V7D10ONSliUgxqJEESEZWtq4hKaFGdaow9a5u/PGSc0je4guAnL32h1CXJSJ+UiMJkIysHGrr0FaJRUUZd/ZuxuzhvWhUpwr3vbWKEVO/5kCWAiBFwp0aSQAcq8CBjYHW/NTqzBjSg1G/ackna36gz9iv+OpbBUCKhDM1kgA4+EvFDmwMtNjoKEb8pgUf3NeTGpVjueWV5fzhg7VkHlEApEg4UiMJgBM5W9ojCai2DWvy0f29uDuxGe8s/55+4xawYocCIEXCjRpJACiwMXgqx0bzh/7n8O7d3XE4rnlxCY/P3sjhHAVAioQLNZIAUGBj8HVpWoe5IxK5rsuZTE7exuUTFrLuvwqAFAkHaiQB8H97JDprK5iqVorh8YFtee228zjwSw5XTFzEuM+2kKMASJGQUiMJgAx9qFWZuuDsU5k3MpFL2tVnzGffctXzi0ndeyjUZYlUWGokAZCelU1cTBRV4qJDXUqFUatKHOMGd2TSDZ3YmZ7FJeMX8tKCbRw/rogVkbKmRhIAGZnZCmwMkf5t6zNvVCK9W9Tj759s5LopS9mZrgBIkbKkRhIAGVk5CmwMoVOrV2bKzQk8Nagd63cdJGlsMlOXf68ASJEyokYSAApsDD0z45qERswd2Zt2DWsx+v213PF6CnsPKgBSJNjUSAIgXYGNYaNh7Sq8dWdX/nJZaxal/kifscl8tHpXqMsSKdfUSAJgf1aOriEJI1FRxm09mzJ7RG8a163K/e98zbC3V504u05EAsvvRmJmVYNZSKTKDWxU8m/4OSu+GjPu7c6DfVoyd91u+oxN5otNe0Ndlki5c9JGYmY9zGwDsNH7vr2ZTQp6ZRFCgY3hLSY6imEXtWDmsJ7UqRLHba+tYPSMNfysAEiRgPFnj2QM0Bf4CcA5txpIDGZRkSTdu6pdk+3hrc0ZNZl1f0/uPf8spqXsJGlsMku3/RTqskTKBb8ObTnnduZbpMQ8z36vkdTSHEnYqxQTzeh+rZh2T3eio4zrpizlbx9vUACkSCn500h2mlkPwJlZnJk9iHeYSxTYGIkSmtRh9vDe3Ni1MS8v3M6lzy1kTdr+UJclErH8aST3AkOBBkAa0AG4L5hFRZITOVsKbIwoVSvF8LcrzuWN27vw8+GjDJy0mGc//VYBkCIl4E8jOds5d4Nz7jTn3KnOuRuBc4JdWKTQh1pFtsSW8cwbmciA9mcw/j9bGDhpEd/uUQCkSHH400ie83NZhaTAxshXs0osz17bgRdu7MSu/Ye59LmFTE7eyjEFQIr4JaawB8ysO9ADiDezB/I8VAPQX02PAhvLj6Rz69O5cR3+8MFaHp+9ic827OWZq9tzZt0qoS5NJKwVtUcSB1TD12yq57kdBAYFv7TIkJ6Zo2tIypH46pWYfFNn/nV1ezb+cJCkccm8tew7BUCKFKHQPRLn3FfAV2b2mnPuuzKsKaLoqvbyx8y4qnNDup1Vl4emr+Z/PljH/PV7ePKqdpxes3KoyxMJO/7MkWSZ2dNmNtvMPs+9Bb2yCKHAxvKrQa1T+PftXXlsQBuWbf+JvmOTmfnNf7V3IpKPP43kLWAT0BT4K7ADWBHEmiJK7hyJlE9RUcbN3Zswe3hvmsVXZcTUbxj69irSFQApcoI/jaSuc+5lIMc595Vz7nagW5DrigjHjjsO/JKjQ1sVQLP4aky/twcPJZ3Npxv20GdMMp9t2BPqskTCgj+NJMf7+oOZXWJmHYGGQawpYiiwsWKJjjLuu6A5s4b1ol61OO58I4Xfv7eaQ4dzTr6xSDnmTyP5u5nVBH4HPAi8BIwKalURQoGNFdM59Wswa1gvhl54FjNWpZE0dgGLt/4Y6rJEQqbIRmJm0UAL59wB59w659yFzrnOzrlZ/jy5mSWZ2WYzSzWz0QU8XsnM3vUeX2ZmTfI89oi3fLOZ9c2zvJaZTTezTWa20bveJSROxKNojqTCiYuJ4vd9WzF9SA/iYqK4fsoyHp21nl+yFQApFU+RjcQ5dwy4vCRP7DWhiUA/oDVwnZm1zrfaHUCGc645vrj6J71tWwODgTZAEjDJez6AccBc51wroD0hDJDMyPId0lAjqbg6nVmb2cN7c2uPJry2eAeXPLeAr7/PCHVZImXKn0Nbi81sgpn1NrNOuTc/tusCpDrntjnnsoGpwIB86wwAXvfuTwcuNt8l4gOAqc65I8657UAq0MXMauD7LJSXAZxz2c65kMW2KrBRAE6Ji+bRy9vw1p1dOZx9jKueX8wz8zaTfVQBkFIxFHpBYh49vK+P5VnmgItOsl0DIO/nmKQBXQtbxzl31MwOAHW95UvzbdsA+AXYB7xqZu2BlcAI51ymH+MIOM2RSF49m9dj7qhEHvtoAxO+SOXzTXt59tr2tDq9RqhLEwmqk+6RePMi+W8nayIABYVP5b+Sq7B1ClseA3QCnnfOdQQygV/NvQCY2d1mlmJmKfv27fOj3OLL8AIbT4lV9Jj41KgcyzNXt2fyTZ3Ze+gwlz+3iOe/VACklG9+fUJiCaUBjfJ83xDYVdg6ZhYD1ATSi9g2DUhzzi3zlk/H11h+xTk32TmX4JxLiI+PL+VQCqbARilMnzanM29kIhe1OpUn527imheXsOPHkOw4iwRdMBvJCqCFmTU1szh8k+f5z/aaBdzi3R8EfO58+ROzgMHeWV1NgRbAcufcbnyf2Hi2t83FwIYgjqFICmyUotStVonnb+zE2Gs7sGXPIfqNW8C/l+xQxIqUO/7MkZSIN+cxDJiHL3b+FefcejN7DEjxTiF+Gfi3maXi2xMZ7G273sym4WsSR4Gh3hlkAPcDb3nNaRtwW7DGcDIKbJSTMTOu6NiArs3q8ND0Nfxp5nrmb9jDU4PaUb/mKaEuTyQg7GT/OzKzKwtYfABY65zbG5SqAiwhIcGlpKQE/Hkv+teXnFO/BhOv9+ckNqnonHO8tex7/vHJRmKijb9e3oaBHRvo0KiEJTNb6ZxL8Gddfw5t3YHvavYbvNsU4AFgkZndVOIqywEFNkpxmBk3dmvM3JG9Ofu06jwwbTVD3lzFTz8fCXVpIqXiTyM5DpzjnLvKOXcVvosLj+A7lffhYBYXzo4dd+z/RXMkUnyN61bl3Xu680i/Vny+aS99xiQzb/3uUJclUmL+NJImzrm8Mad7gZbOuXT+L9Cxwjn4Sw7OoTkSKZHoKOOe88/io/t7cXrNytzz75U8MO0bDvxSYf9JSQTzp5EsMLOPzewWM7sFmAkkm1lVIGRXlYeaLkaUQDj79Op8cF9Phl/UnJnf7CJpbDILtygAUiKLP41kKPAa0AHoCLyB7yyqTOfchUGsLawpsFECJS4migf6nM2MIT04JS6aG19exp9nriMr+2ioSxPxy0lP//Wu65ju3cSjwEYJtA6NajF7eG+enLuJVxftIPnbffzrmg50blw71KWJFOmkeyRmdqWZbTGzA2Z20MwOmdnBsigunCmwUYKhcmw0f7msDW/f1ZWcY46rX1jMU3M3ceSo4uklfPlzaOsp4HLnXE3nXA3nXHXnXIVPodMciQRTj7PqMXdkbwZ1bsikL7cyYMIiNuyq8P9/kzDlTyPZ45wL2Wd+hKuMzGwqKbBRgqh65VieGtSel25O4MefsxkwcSETv0jl6DHF00t48SciJcXM3gU+xHf9CADOufeDVlUEyMjKprYCG6UM/Kb1acxvXJs/fbiOp+dt5rONe/jX1e1pFl8t1KWJAP7tkdQAsoA+wGXe7dJgFhUJFNgoZalO1TgmXN+RcYM7sG1fJv3HL+D1xTs4rnh6CQP+nLUVslDEcJaRlU0dTbRLGTIzBnRoQLdmdXlo+hr+Mms98zfs5qlB7WlQSwGQEjqF7pGY2UPe1+fMbHz+W9mVGJ4ysrKppVN/JQROq1GZ1247j8cHtuXr7/eTNCaZ6SvTFE8vIVPUHknuBHvgY3PLAQU2SiiZGdd3PZNezevx4HurefC91cxbv5vHB7YlvnqlUJcnFUyhjcQ595H39fWyKycyKLBRwsWZdavwzt3deGXhdp6ev5m+Y5P5xxXn0q9t/VCXJhWIPxcktjSzyWY238w+z72VRXHh6oACGyWMREcZdyU245P7e3FGrcoMeWsVo979hgNZCoCUsuHP6b/vAS/g+0wSXV6Lb34EdDGihJcWp/kCICd8nsqEL1JZsvUnnhrUjsSW8aEuTco5f07/Peqce945t9w5tzL3FvTKwpgCGyVcxUZHMeq3Lfngvh5UqxzDza8s548friXziAIgJXj8aSQfmdl9ZlbfzOrk3oJeWRhLz9QeiYS3dg1r8fH9vbizV1PeWvY9/ccvIGVHeqjLknLKn0ZyC/B7YDGw0rtV6DO59nvHnmtpjkTCWOXYaP54aWum3tWNY8cdV7+4hH/O2cjhHB2hlsAqspGYWRRwo3Ouab5bszKqLywpsFEiSddmdZk7MpHB553Ji19t4/IJC1n33wOhLkvKkSIbiXPuOPBMGdUSMRTYKJGmWqUY/nllW1699Tz2Z+VwxcRFPPefLQqAlIDw59DWfDO7ypROeEJ6pgIbJTJd2OpU5o9KpH/b+vzr02+56vnFpO79OdRlSYTzp5E8gO8U4CP6YCufjCxdjCiRq1aVOMZf15EJ13fku/QsLhm/gFcWblcApJTYSRuJ90FWUc65OH2wlY8CG6U8uLTdGcwfmUjP5vV47OMNXP/SUnamZ4W6LIlA/uyRYGa1zayLmSXm3oJdWDjL8A5tiUS6U2tU5uVbEnjqqnasTTtAv3ELeHfF9wqAlGLxJyLlTiAZmAf81fv6aHDLCm+5H2olUh6YGdec14i5IxM5t0ENHp6xljtfT2HvocOhLk0ihD97JCOA84DvnHMXAh2BfUGtKowpsFHKq0Z1qvD2nd3486WtWZj6I33GJPPxml2hLksigD+N5LBz7jCAmVVyzm0Czg5uWeErN7Cxji5GlHIoKsq4vVdTPhnem8Z1qjDs7a+5/52v2e9dOyVSEH8aSZqZ1cL3me2fmtlMoML+NyU3HkV7JFKeNT+1GjOG9OB3v23JnLU/0GdMMl9s3hvqsiRM+XPW1kDn3H7n3KPAn4CXgSuCXVi4yv2fmeZIpLyLiY7i/otb8OHQntSqEsttr67gkffX8LMCICUff8/a6mVmtznnvgKWAA2CW1b4UmCjVDTnNqjJrGG9uCexGVNX7KTfuGSWbfsp1GVJGPHnrK2/AA8Dj3iLYoE3g1lUOMv9LBIFNkpFUjk2mkf6n8O0e7pjGIOnLOXvH29QAKQA/u2RDAQuBzIBnHO7gOrBLCqcZXjJv9ojkYrovCZ1mDOiN9d3OZOXFm7n0ucWsiZtf6jLkhDzp5FkO9/VSQ7AzKoGt6TwpsBGqeiqVorhHwPb8vrtXTh0OIeBkxYz5tNvyVEAZIXlTyOZZmYvArXM7C7gM2CKP09uZklmttnMUs1sdAGPVzKzd73Hl5lZkzyPPeIt32xmffNtF21mX5vZx/7UEUjpmdnUqarARpHzW8Yzf+T5XNauPuP+s4UrJy1my55DoS5LQsCfs7aeAaYDM/BdP/Jn59xzJ9vOzKKBiUA/oDVwnZm1zrfaHUCGc645MAZ40tu2NTAYaAMkAZO858s1Ath4shqCISMrm1o6Y0sEgJpVYhk7uCPP39CJtIwsLnluIS8t2KYAyArGr7O2nHOfOud+75x70Dn3qZ/P3QVIdc5tc85lA1OBAfnWGQC87t2fDlzsxdUPAKY6544457YDqd7zYWYNgUuAl/ysI6AysnIU2CiST7+29Zk/6nwSW8Tz9082MniKAiArkkIbSW5cfAE3f2PkGwA783yfxq9PGz6xjnPuKHAAqHuSbccCDwEhOSCrwEaRgsVXr8SUmzvz9KB2bNx1kKSxybyzXAGQFUGhjSQ3Lr6Am78x8gVNIuT/jSpsnQKXm9mlwF7n3MqTvrjZ3WaWYmYp+/YFLhosXYGNIoUyM65OaMTcUYm0b1SLR95fy22vrWDPQQVAlmd+HdoqoTSgUZ7vG/LraJUT65hZDFATSC9i257A5Wa2A9+hsovMrMBrWpxzk51zCc65hPj4+NKPBl9g4wEFNoqcVINap/DmHV159LLWLN32E33GJDNrdYVNVir3gtlIVgAtzKypmcXhmzyflW+dWcAt3v1BwOfeqcazgMHeWV1NgRbAcufcI865hs65Jt7zfe6cuzGIY/h/FNgo4r+oKOPWnk2ZPbw3TetVZfg7XzP07VVkZCoAsrwJWiPx5jyG4fv8ko3ANOfcejN7zMwu91Z7GahrZqn4PtJ3tLftemAasAGYCwx1zoX8EloFNooUX7P4aky/tzu/73s289fvps/YZD7ftCfUZUkAWUWYCEtISHApKSmlfp6UHekMemEJb9zehcSWgTlcJlKRbNh1kAemfcOm3Ye4NqERf7z0HKpX1h5+ODKzlc65BH/WDeahrXJHgY0ipdP6jBrMHNaTIRecxXsrd5I0dgFLtioAMtKpkRRDbmCjDm2JlFylmGgeTmrFe/d2JzbauG7KUh77SAGQkUyNpBjSM32BjbU12S5Sap0b12H2iN7c3L0xryzaziXjF7B6pwIgI5EaSTHsz1Jgo0ggVYmL4bEB5/LmHV3Jyj7Glc8v5tn5m8k+qgDISKJGUgwKbBQJjl4t6jF3ZCIDOpzB+M9TGThpEZt3KwAyUqiRFIMCG0WCp+YpsTx7TQdevKkzuw8c5rLnFvLiV1s5pgDIsKdGUgwKbBQJvr5tTmfeqEQubBXPP+dsYvDkJXz3U2aoy5IiqJEUgwIbRcpGvWqVeOH8T6hgAAAQTklEQVTGzjx7TXs27T5Ev3ELeHPpdwqADFNqJMWQnpWta0hEyoiZcWWnhswbmUjnxrX544fruOXVFew+oADIcKNG4qfcwEbNkYiUrTNqncIbt3fhbwPasGJ7On3GfMWHX/9XeydhRI3ETwpsFAkdM+Om7k2YM6I3LU6rzsh3v+G+t1bx089HQl2aoEbiNwU2ioRek3pVmXZPdx5OasV/Nu6l79hkPt2gAMhQUyPx04l4FB3aEgmp6ChjyAVnMev+nsRXr8xdb6Tw4HurOXg4J9SlVVhqJH7KUGCjSFhpdXoNZg7tybALm/P+qjSSxiSzOPXHUJdVIamR+EmBjSLhJy4migf7ns2MIT2oHBvN9S8t49FZ6/klWwGQZUmNxE+5gY11dGhLJOx0PLM2nwzvza09mvDa4h1cMn4Bq77PCHVZFYYaiZ8ycgMb4xTYKBKOTomL5tHL2/D2nV05cvQ4g55fzNPzNikAsgyokfgpI1MXI4pEgh7N6zFnZG+u6tSQiV9sZcDERWz84WCoyyrX1Ej8lJGleBSRSFGjcixPX92eKTcnsO/QES6fsJBJX6YqADJI1Ej8lJ6ZTW0FNopElN+2Po35oxL5bevTeGruZq5+YTHbf1QAZKCpkfhpf1aO9khEIlCdqnFMvL4T4wZ3IHXvz/Qft4A3luzguPZOAkaNxE8KbBSJXGbGgA4NmD/qfM5rWoc/z1zPza8sZ9f+X0JdWrmgRuKHo8eOK7BRpBw4vWZlXr/tPP4x8FxWfZ9B37HJzFiZpgDIUlIj8YMCG0XKDzPjhq6NmTOiN61Or87v3lvNPf9eyY8KgCwxNRI/ZGT5LkbUVe0i5UfjulWZend3/tC/FV9u3kffMcnMXbc71GVFJDUSP+TGo2iORKR8iY4y7k48i4+H96J+rcrc++ZKHnj3Gw78ogDI4lAj8cOJCHnNkYiUSy1Pq84H9/Vk+MUtmLl6F0ljk1mwZV+oy4oYaiR+2K/ARpFyLzY6igd+25L3h/SgSlw0N728nD99uI6s7KOhLi3sqZH4QYGNIhVH+0a1+GR4b+7o1ZQ3l31H/3ELWPldeqjLCmtqJH5QYKNIxVI5Npo/Xdqad+7qxtHjjqtfWMITczZx5Kji6QuiRuKHdAU2ilRI3ZrVZe7IRK5JaMQLX21lwIRFrN91INRlhR01Ej/sV2CjSIVVrVIMT1zVjlduTeCnzGyumLiICZ9v4egxxdPnUiPxg/ZIROSiVqcxf2QifduczjPzv2XQC0vYuu/nUJcVFtRI/JCRlUMtXdUuUuHVrhrHhOs78dx1HdnxUyaXjF/Aq4u2V/gASDUSP2QosFFE8ris/RnMH5lI92Z1+etHG7jhpWWkZWSFuqyQCWojMbMkM9tsZqlmNrqAxyuZ2bve48vMrEmexx7xlm82s77eskZm9oWZbTSz9WY2Ipj1w/8FNmqORETyOrVGZV659TyeuLIta9L2kzR2AdNSdlbIAMigNRIziwYmAv2A1sB1ZtY632p3ABnOuebAGOBJb9vWwGCgDZAETPKe7yjwO+fcOUA3YGgBzxlQuYGNtXVoS0TyMTMGdzmTuSMTaXNGDR6avoa73khh76HDoS6tTAVzj6QLkOqc2+acywamAgPyrTMAeN27Px242MzMWz7VOXfEObcdSAW6OOd+cM6tAnDOHQI2Ag2COIYTOVu6ql1ECtOoThXeuasbf7zkHJK3/EjfMcnMXvtDqMsqM8FsJA2AnXm+T+PXf/RPrOOcOwocAOr6s613GKwjsCyANf9KbvKv5khEpChRUcadvZsxe3gvGtWpwn1vrWLE1K85kFX+AyCD2UisgGX5Dx4Wtk6R25pZNWAGMNI5d7DAFze728xSzCxl376Sh68psFFEiqP5qdWZMaQHo37Tkk/W/ECfsV/x5ea9oS4rqILZSNKARnm+bwjsKmwdM4sBagLpRW1rZrH4mshbzrn3C3tx59xk51yCcy4hPj6+xIPIyNShLREpntjoKEb8pgUf3NeTGpVjufXVFfzhg7VkHimfAZDBbCQrgBZm1tTM4vBNns/Kt84s4Bbv/iDgc+c75WEWMNg7q6sp0AJY7s2fvAxsdM49G8TaTzhxaEt7JCJSTG0b1uSj+3txd2Iz3ln+Pf3GLWD59vIXABm0RuLNeQwD5uGbFJ/mnFtvZo+Z2eXeai8Ddc0sFXgAGO1tux6YBmwA5gJDnXPHgJ7ATcBFZvaNd+sfrDGAb7K9cqwCG0WkZCrHRvOH/ufw7t3dcTiunbyEx2dv5HBO+QmAtIpwznNCQoJLSUkp0bYPvreaRak/suSRiwNclYhUNJlHjvKP2Rt5e9n3tDytGs9e04FzG9QMdVkFMrOVzrkEf9bVle0nocBGEQmUqpVieHxgW1677TwO/JLDFRMXMe6zLeREeACkGslJKLBRRALtgrNPZd7IRC5pV58xn33LVc8vJnXvoVCXVWJqJCehwEYRCYZaVeIYN7gjk27oxM70LPqPX8hLC7ZFZACkGslJaI9ERIKpf9v6zBuVSGKLevz9k41cN2UpO9MjKwBSjaQIR48d5+BhBTaKSHCdWr0yU25O4KlB7Vi/6yBJY5OZuvz7iAmAVCMpggIbRaSsmBnXJDRi7sjetGtYi9Hvr+WO11PYezD8AyDVSIqgwEYRKWsNa1fhrTu78pfLWrMo9Uf6jE3mo9X5Q0HCixpJERTYKCKhEBVl3NazKbNH9KZx3arc/87XDHt71YnIpnCjRlIEBTaKSCidFV+NGfd258E+LZm7bjd9xibzxabwC4BUIymCAhtFJNRioqMYdlELZg7rSZ0qcdz22gpGz1jDz2EUAKlGUoR0b45EgY0iEmptzqjJrPt7cu/5ZzEtZSdJY5NZuu2nUJcFqJEUaX9WjgIbRSRsVIqJZnS/Vrx3b3eio4zrpizlbx9vCHkApBpJEdIzlbMlIuGnc+M6zBnRmxu7Nublhdu59LmFrEnbH7J61EiKkKFGIiJhqkpcDH+74lzeuL0LPx8+ysBJi3n2029DEgCpRlKEjCzFo4hIeEtsGc+8UYkMaH8G4/+zhYGTFvHtnrINgFQjKUJGVo7O2BKRsFfzlFievbYDL9zYiR/2H+bS5xYyOXkrx8ooAFKNpAi+ORLFo4hIZEg61xcAeUHLeB6fvYnrJi8lKzv4pwnHBP0VIpRzjgvPjqdDo1qhLkVExG/1qlXixZs68/6q/5LyXTpV4oL/Z14ftSsiIr+ij9oVEZEyo0YiIiKlokYiIiKlokYiIiKlokYiIiKlokYiIiKlokYiIiKlokYiIiKlUiEuSDSzfcB3Jdy8HvBjAMsJVxpn+aJxli+hGGdj51y8PytWiEZSGmaW4u/VnZFM4yxfNM7yJdzHqUNbIiJSKmokIiJSKmokJzc51AWUEY2zfNE4y5ewHqfmSEREpFS0RyIiIqWiRlIIM0sys81mlmpmo0NdT0mY2Q4zW2tm35hZiresjpl9amZbvK+1veVmZuO98a4xs055nucWb/0tZnZLqMaTp55XzGyvma3Lsyxg4zKzzt7PLdXb1sp2hCfqKGicj5rZf7339Bsz65/nsUe8mjebWd88ywv8XTazpma2zBv/u2YWks+VNrNGZvaFmW00s/VmNsJbXq7e0yLGGfnvqXNOt3w3IBrYCjQD4oDVQOtQ11WCcewA6uVb9hQw2rs/GnjSu98fmAMY0A1Y5i2vA2zzvtb27tcO8bgSgU7AumCMC1gOdPe2mQP0C6NxPgo8WMC6rb3f00pAU+/3N7qo32VgGjDYu/8CMCRE46wPdPLuVwe+9cZTrt7TIsYZ8e+p9kgK1gVIdc5tc85lA1OBASGuKVAGAK97918Hrsiz/A3nsxSoZWb1gb7Ap865dOdcBvApkFTWReflnEsG0vMtDsi4vMdqOOeWON+/xjfyPFeZKmSchRkATHXOHXHObQdS8f0eF/i77P2P/CJgurd93p9ZmXLO/eCcW+XdPwRsBBpQzt7TIsZZmIh5T9VICtYA2Jnn+zSKfsPDlQPmm9lKM7vbW3aac+4H8P1iA6d6ywsbc6T8LAI1rgbe/fzLw8kw75DOK7mHeyj+OOsC+51zR/MtDykzawJ0BJZRjt/TfOOECH9P1UgKVtDx00g8va2nc64T0A8YamaJRaxb2Jgj/WdR3HGF+3ifB84COgA/AP/ylkf8OM2sGjADGOmcO1jUqgUsi5ixFjDOiH9P1UgKlgY0yvN9Q2BXiGopMefcLu/rXuADfLvEe7xdfbyve73VCxtzpPwsAjWuNO9+/uVhwTm3xzl3zDl3HJiC7z2F4o/zR3yHhGLyLQ8JM4vF98f1Lefc+97icveeFjTO8vCeqpEUbAXQwjsDIg4YDMwKcU3FYmZVzax67n2gD7AO3zhyz2a5BZjp3Z8F3OydEdMNOOAdTpgH9DGz2t4udx9vWbgJyLi8xw6ZWTfvmPPNeZ4r5HL/sHoG4ntPwTfOwWZWycyaAi3wTTAX+LvszRV8AQzyts/7MytT3s/5ZWCjc+7ZPA+Vq/e0sHGWi/e0LGb0I/GG78yQb/GdHfE/oa6nBPU3w3c2x2pgfe4Y8B1H/Q+wxftax1tuwERvvGuBhDzPdTu+ib5U4LYwGNs7+A4B5OD739kdgRwXkIDvH/NWYALehbthMs5/e+NYg+8PTf086/+PV/Nm8pyVVNjvsvc7stwb/3tApRCNsxe+QzBrgG+8W//y9p4WMc6If091ZbuIiJSKDm2JiEipqJGIiEipqJGIiEipqJGIiEipqJGIiEipqJGInISZLfa+NjGz6wP83H8o6LVEIolO/xXxk5ldgC+l9dJibBPtnDtWxOM/O+eqBaI+kVDRHonISZjZz97dJ4De3mdGjDKzaDN72sxWeIF793jrX+B97sTb+C40w8w+9MIz1+cGaJrZE8Ap3vO9lfe1vKu2nzazdeb7HI1r8zz3l2Y23cw2mdlb3hXTmNkTZrbBq+WZsvwZScUWc/JVRMQzmjx7JF5DOOCcO8/MKgGLzGy+t24X4Fzni/8GuN05l25mpwArzGyGc260mQ1zznUo4LWuxBfi1x6o522T7D3WEWiDL0dpEdDTzDbgi9do5ZxzZlYr4KMXKYT2SERKrg++zKdv8MWB18WXhwSwPE8TARhuZquBpfgC91pQtF7AO84X5rcH+Ao4L89zpzlfyN83QBPgIHAYeMnMrgSySj06ET+pkYiUnAH3O+c6eLemzrncPZLMEyv55lZ+A3R3zrUHvgYq+/HchTmS5/4xIMb5PoOiC75k2SuAucUaiUgpqJGI+O8Qvo9IzTUPGOJFg2NmLb2k5fxqAhnOuSwza4Xv42Fz5eRun08ycK03DxOP72N3lxdWmPcZFzWdc7OBkfgOi4mUCc2RiPhvDXDUO0T1GjAO32GlVd6E9z4K/mjTucC9ZrYGX4rr0jyPTQbWmNkq59wNeZZ/gO8zxlfjS4x9yDm322tEBakOzDSzyvj2ZkaVbIgixafTf0VEpFR0aEtEREpFjUREREpFjUREREpFjUREREpFjUREREpFjUREREpFjUREREpFjURERErlfwFmH+ZOyYdNCQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.sched.plot_lr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "# points from individual losess along the way. \n",
    "epochs = 20\n",
    "points = list(range(epochs, len(learn.sched.losses), int(len(learn.sched.losses)/epochs)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "p_val_losses = learn.sched.val_losses\n",
    "#p_losses = [learn.sched.losses[i] for i in range(epochs,150)]\n",
    "p_losses = [learn.sched.losses[i] for i in points ]\n",
    "p_rec_metrics = learn.sched.rec_metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "20"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(learn.sched.val_losses)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<fastai.sgdr.CircularLR at 0x7f41137d76d8>"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.sched"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0,0.5,'Accuracy')"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x864 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "epochs = 20\n",
    "\n",
    "# Losses\n",
    "fig,ax = plt.subplots(2,1,figsize=(8,12))\n",
    "ax[0].plot(list(range(epochs)),learn.sched.val_losses, label='Validation loss')\n",
    "ax[0].plot(list(range(epochs)),[learn.sched.losses[i] for i in range(epochs)], label='Training loss')\n",
    "ax[0].set_xlabel('Epoch')\n",
    "ax[0].set_ylabel('Loss')\n",
    "ax[0].legend(loc='upper right')\n",
    "\n",
    "# Accuracy\n",
    "ax[1].plot(list(range(epochs)),learn.sched.rec_metrics)\n",
    "ax[1].set_xlabel('Epoch')\n",
    "ax[1].set_ylabel('Accuracy')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 166,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0,0.5,'Accuracy')"
      ]
     },
     "execution_count": 166,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x864 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "epochs = 15\n",
    "\n",
    "# Losses\n",
    "fig,ax = plt.subplots(2,1,figsize=(8,12))\n",
    "ax[0].plot(list(range(epochs)),learn.sched.val_losses, label='Validation loss')\n",
    "ax[0].plot(list(range(epochs)),[learn.sched.losses[i] for i in range(epochs)], label='Training loss')\n",
    "ax[0].set_xlabel('Epoch')\n",
    "ax[0].set_ylabel('Loss')\n",
    "ax[0].legend(loc='upper right')\n",
    "\n",
    "# Accuracy\n",
    "ax[1].plot(list(range(epochs)),learn.sched.rec_metrics)\n",
    "ax[1].set_xlabel('Epoch')\n",
    "ax[1].set_ylabel('Accuracy')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [],
   "source": [
    "epochs = 15\n",
    "xvalidate_loss = learn.sched.val_losses\n",
    "xtrain_loss = [learn.sched.losses[i] for i in range(epochs)]\n",
    "xaccuracy = learn.sched.rec_metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "8000"
      ]
     },
     "execution_count": 96,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trainsize"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [],
   "source": [
    "results[trainsize]= pd.DataFrame({'val': xvalidate_loss, 'train':xtrain_loss, 'acc':xaccuracy})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 161,
   "metadata": {},
   "outputs": [],
   "source": [
    "#results1000 = pd.DataFrame({'val': xvalidate_loss, 'train':xtrain_loss, 'acc':xaccuracy})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Results with different sized training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>val</th>\n",
       "      <th>train</th>\n",
       "      <th>acc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[4.82410333840549]</td>\n",
       "      <td>0.499924</td>\n",
       "      <td>0.6485</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[0.9181957594454289]</td>\n",
       "      <td>0.592864</td>\n",
       "      <td>0.7395</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[1.155655806310475]</td>\n",
       "      <td>0.605727</td>\n",
       "      <td>0.7775</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[0.8125892618820071]</td>\n",
       "      <td>0.634543</td>\n",
       "      <td>0.8100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[5.843671057600528]</td>\n",
       "      <td>0.614488</td>\n",
       "      <td>0.7745</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>[6.287118505634368]</td>\n",
       "      <td>0.622923</td>\n",
       "      <td>0.7820</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>[12.47612617594935]</td>\n",
       "      <td>0.648743</td>\n",
       "      <td>0.7910</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>[111.34638942733035]</td>\n",
       "      <td>0.660557</td>\n",
       "      <td>0.7775</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>[1.3332453550454229]</td>\n",
       "      <td>0.660923</td>\n",
       "      <td>0.8205</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>[7.197082749731838]</td>\n",
       "      <td>0.652427</td>\n",
       "      <td>0.8255</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>[183.9017787543712]</td>\n",
       "      <td>0.664865</td>\n",
       "      <td>0.8140</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>[3.341987658651546]</td>\n",
       "      <td>0.738382</td>\n",
       "      <td>0.8195</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>[29.879887756768614]</td>\n",
       "      <td>0.726256</td>\n",
       "      <td>0.8280</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>[2.2274074643151835]</td>\n",
       "      <td>0.740831</td>\n",
       "      <td>0.8325</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>[7.722132334040012]</td>\n",
       "      <td>0.724308</td>\n",
       "      <td>0.8095</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     val     train     acc\n",
       "0     [4.82410333840549]  0.499924  0.6485\n",
       "1   [0.9181957594454289]  0.592864  0.7395\n",
       "2    [1.155655806310475]  0.605727  0.7775\n",
       "3   [0.8125892618820071]  0.634543  0.8100\n",
       "4    [5.843671057600528]  0.614488  0.7745\n",
       "5    [6.287118505634368]  0.622923  0.7820\n",
       "6    [12.47612617594935]  0.648743  0.7910\n",
       "7   [111.34638942733035]  0.660557  0.7775\n",
       "8   [1.3332453550454229]  0.660923  0.8205\n",
       "9    [7.197082749731838]  0.652427  0.8255\n",
       "10   [183.9017787543712]  0.664865  0.8140\n",
       "11   [3.341987658651546]  0.738382  0.8195\n",
       "12  [29.879887756768614]  0.726256  0.8280\n",
       "13  [2.2274074643151835]  0.740831  0.8325\n",
       "14   [7.722132334040012]  0.724308  0.8095"
      ]
     },
     "execution_count": 98,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results[8000]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load\n",
    "results = np.load('resultsx.npy').item()"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "# while 200 labeled items give 0.5850, 300 give 0.6 acuracy. 1,5% increase with 100 more data.\n",
    "# 100 gave 0.54\n",
    "# \n",
    "# 500 0.625\n",
    "\n",
    "#dict\n",
    "results = {}\n",
    "\n",
    "results['100']= results100\n",
    "#results['200']= results200\n",
    "results['300']= results300\n",
    "results['500']= results500\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [],
   "source": [
    "#dictionary = {'hello':'world'}\n",
    "np.save('resultsx1.npy', results) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "results100.to_csv('results100.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 155,
   "metadata": {},
   "outputs": [],
   "source": [
    "results100 = pd.read_csv('results100.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 149,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>val</th>\n",
       "      <th>train</th>\n",
       "      <th>acc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[0.9872420721650124]</td>\n",
       "      <td>0.863131</td>\n",
       "      <td>0.6270</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[1.0959983005225657]</td>\n",
       "      <td>0.711867</td>\n",
       "      <td>0.5940</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[1.1319324245899915]</td>\n",
       "      <td>0.657229</td>\n",
       "      <td>0.6450</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[1.0007186770141125]</td>\n",
       "      <td>0.718754</td>\n",
       "      <td>0.6755</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[0.9728661698848009]</td>\n",
       "      <td>0.791027</td>\n",
       "      <td>0.6680</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>[1.0017748506218196]</td>\n",
       "      <td>0.814120</td>\n",
       "      <td>0.6905</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>[0.9709802632331848]</td>\n",
       "      <td>0.852151</td>\n",
       "      <td>0.6965</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>[0.9872805901020766]</td>\n",
       "      <td>0.830495</td>\n",
       "      <td>0.6780</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>[1.007214888587594]</td>\n",
       "      <td>0.840957</td>\n",
       "      <td>0.6870</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>[1.0563348397463561]</td>\n",
       "      <td>0.891591</td>\n",
       "      <td>0.6785</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>[1.0370394280478359]</td>\n",
       "      <td>0.874175</td>\n",
       "      <td>0.7070</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>[1.054195993117988]</td>\n",
       "      <td>0.927292</td>\n",
       "      <td>0.7170</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>[0.9998700297549367]</td>\n",
       "      <td>0.951997</td>\n",
       "      <td>0.7215</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>[1.0190723473578691]</td>\n",
       "      <td>0.952755</td>\n",
       "      <td>0.7135</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>[1.0098433435857297]</td>\n",
       "      <td>0.968196</td>\n",
       "      <td>0.7225</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     val     train     acc\n",
       "0   [0.9872420721650124]  0.863131  0.6270\n",
       "1   [1.0959983005225657]  0.711867  0.5940\n",
       "2   [1.1319324245899915]  0.657229  0.6450\n",
       "3   [1.0007186770141125]  0.718754  0.6755\n",
       "4   [0.9728661698848009]  0.791027  0.6680\n",
       "5   [1.0017748506218196]  0.814120  0.6905\n",
       "6   [0.9709802632331848]  0.852151  0.6965\n",
       "7   [0.9872805901020766]  0.830495  0.6780\n",
       "8    [1.007214888587594]  0.840957  0.6870\n",
       "9   [1.0563348397463561]  0.891591  0.6785\n",
       "10  [1.0370394280478359]  0.874175  0.7070\n",
       "11   [1.054195993117988]  0.927292  0.7170\n",
       "12  [0.9998700297549367]  0.951997  0.7215\n",
       "13  [1.0190723473578691]  0.952755  0.7135\n",
       "14  [1.0098433435857297]  0.968196  0.7225"
      ]
     },
     "execution_count": 149,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results1000 = pd.DataFrame({'val': xvalidate_loss, 'train':xtrain_loss, 'acc':xaccuracy})\n",
    "results1000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>val</th>\n",
       "      <th>train</th>\n",
       "      <th>acc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[1.0498955637216567]</td>\n",
       "      <td>1.034174</td>\n",
       "      <td>0.5725</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[1.1106152303516865]</td>\n",
       "      <td>0.917990</td>\n",
       "      <td>0.5845</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[1.108517288863659]</td>\n",
       "      <td>0.972473</td>\n",
       "      <td>0.6020</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[1.197882125020027]</td>\n",
       "      <td>0.933784</td>\n",
       "      <td>0.5960</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[2.1856168991476297]</td>\n",
       "      <td>0.848695</td>\n",
       "      <td>0.5845</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>[3.929511443644762]</td>\n",
       "      <td>0.880784</td>\n",
       "      <td>0.5810</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>[1.5566950822472572]</td>\n",
       "      <td>0.871992</td>\n",
       "      <td>0.5965</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>[2.5267729723006487]</td>\n",
       "      <td>0.914394</td>\n",
       "      <td>0.6080</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>[4.981768222734332]</td>\n",
       "      <td>1.000357</td>\n",
       "      <td>0.6215</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>[5.416245001584292]</td>\n",
       "      <td>0.948081</td>\n",
       "      <td>0.6080</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>[4.4219001570642]</td>\n",
       "      <td>0.963638</td>\n",
       "      <td>0.6015</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>[4.377095339268446]</td>\n",
       "      <td>0.993501</td>\n",
       "      <td>0.6130</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>[5.09209756077826]</td>\n",
       "      <td>0.970043</td>\n",
       "      <td>0.6250</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>[3.804072525113821]</td>\n",
       "      <td>0.962595</td>\n",
       "      <td>0.6210</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>[5.591898477390409]</td>\n",
       "      <td>0.937253</td>\n",
       "      <td>0.6145</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     val     train     acc\n",
       "0   [1.0498955637216567]  1.034174  0.5725\n",
       "1   [1.1106152303516865]  0.917990  0.5845\n",
       "2    [1.108517288863659]  0.972473  0.6020\n",
       "3    [1.197882125020027]  0.933784  0.5960\n",
       "4   [2.1856168991476297]  0.848695  0.5845\n",
       "5    [3.929511443644762]  0.880784  0.5810\n",
       "6   [1.5566950822472572]  0.871992  0.5965\n",
       "7   [2.5267729723006487]  0.914394  0.6080\n",
       "8    [4.981768222734332]  1.000357  0.6215\n",
       "9    [5.416245001584292]  0.948081  0.6080\n",
       "10     [4.4219001570642]  0.963638  0.6015\n",
       "11   [4.377095339268446]  0.993501  0.6130\n",
       "12    [5.09209756077826]  0.970043  0.6250\n",
       "13   [3.804072525113821]  0.962595  0.6210\n",
       "14   [5.591898477390409]  0.937253  0.6145"
      ]
     },
     "execution_count": 102,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results500 = pd.DataFrame({'val': xvalidate_loss, 'train':xtrain_loss, 'acc':xaccuracy})\n",
    "results500"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>val</th>\n",
       "      <th>train</th>\n",
       "      <th>acc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[1.148422669172287]</td>\n",
       "      <td>1.120210</td>\n",
       "      <td>0.5775</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[1.2495238498449326]</td>\n",
       "      <td>0.912188</td>\n",
       "      <td>0.5135</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[1.3291830311119557]</td>\n",
       "      <td>0.922584</td>\n",
       "      <td>0.5190</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[1.2065900976359845]</td>\n",
       "      <td>0.835061</td>\n",
       "      <td>0.5440</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[1.298182210713625]</td>\n",
       "      <td>0.897600</td>\n",
       "      <td>0.5600</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>[1.3334138369262218]</td>\n",
       "      <td>0.852542</td>\n",
       "      <td>0.5685</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>[1.4095449968278408]</td>\n",
       "      <td>0.891998</td>\n",
       "      <td>0.5610</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>[1.5145190207064152]</td>\n",
       "      <td>0.857893</td>\n",
       "      <td>0.5980</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>[1.417078743264079]</td>\n",
       "      <td>0.832469</td>\n",
       "      <td>0.6000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>[1.4897609430849552]</td>\n",
       "      <td>0.833281</td>\n",
       "      <td>0.5930</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>[1.5607994010448456]</td>\n",
       "      <td>0.838134</td>\n",
       "      <td>0.6005</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>[1.7070065824985503]</td>\n",
       "      <td>0.868666</td>\n",
       "      <td>0.5945</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>[1.640056502968073]</td>\n",
       "      <td>0.871638</td>\n",
       "      <td>0.5885</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>[1.5934658929109573]</td>\n",
       "      <td>0.901595</td>\n",
       "      <td>0.5940</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>[1.6655278148800134]</td>\n",
       "      <td>0.927699</td>\n",
       "      <td>0.5920</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     val     train     acc\n",
       "0    [1.148422669172287]  1.120210  0.5775\n",
       "1   [1.2495238498449326]  0.912188  0.5135\n",
       "2   [1.3291830311119557]  0.922584  0.5190\n",
       "3   [1.2065900976359845]  0.835061  0.5440\n",
       "4    [1.298182210713625]  0.897600  0.5600\n",
       "5   [1.3334138369262218]  0.852542  0.5685\n",
       "6   [1.4095449968278408]  0.891998  0.5610\n",
       "7   [1.5145190207064152]  0.857893  0.5980\n",
       "8    [1.417078743264079]  0.832469  0.6000\n",
       "9   [1.4897609430849552]  0.833281  0.5930\n",
       "10  [1.5607994010448456]  0.838134  0.6005\n",
       "11  [1.7070065824985503]  0.868666  0.5945\n",
       "12   [1.640056502968073]  0.871638  0.5885\n",
       "13  [1.5934658929109573]  0.901595  0.5940\n",
       "14  [1.6655278148800134]  0.927699  0.5920"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results300 = pd.DataFrame({'val': xvalidate_loss, 'train':xtrain_loss, 'acc':xaccuracy})\n",
    "results300"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>val</th>\n",
       "      <th>train</th>\n",
       "      <th>acc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[1.195070844143629]</td>\n",
       "      <td>1.041893</td>\n",
       "      <td>0.5360</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[1.3940935591161252]</td>\n",
       "      <td>1.094646</td>\n",
       "      <td>0.5690</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[1.4975435910969972]</td>\n",
       "      <td>0.992466</td>\n",
       "      <td>0.5460</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[1.3699843916893006]</td>\n",
       "      <td>0.969821</td>\n",
       "      <td>0.5745</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[1.4847442761063576]</td>\n",
       "      <td>0.998617</td>\n",
       "      <td>0.5510</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>[1.6594213886260987]</td>\n",
       "      <td>0.936941</td>\n",
       "      <td>0.5570</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>[1.7852485501468183]</td>\n",
       "      <td>0.986875</td>\n",
       "      <td>0.5560</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>[1.7116678630709647]</td>\n",
       "      <td>0.982495</td>\n",
       "      <td>0.5495</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>[1.5778338895440103]</td>\n",
       "      <td>0.975589</td>\n",
       "      <td>0.5430</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>[1.5465353992581368]</td>\n",
       "      <td>0.944386</td>\n",
       "      <td>0.5615</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>[1.61626892131567]</td>\n",
       "      <td>0.948060</td>\n",
       "      <td>0.5850</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>[1.6108183098435402]</td>\n",
       "      <td>0.906177</td>\n",
       "      <td>0.5755</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>[1.669055641502142]</td>\n",
       "      <td>0.890414</td>\n",
       "      <td>0.5745</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>[1.658261726796627]</td>\n",
       "      <td>0.906241</td>\n",
       "      <td>0.5845</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>[1.7189348363280297]</td>\n",
       "      <td>0.886755</td>\n",
       "      <td>0.5830</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     val     train     acc\n",
       "0    [1.195070844143629]  1.041893  0.5360\n",
       "1   [1.3940935591161252]  1.094646  0.5690\n",
       "2   [1.4975435910969972]  0.992466  0.5460\n",
       "3   [1.3699843916893006]  0.969821  0.5745\n",
       "4   [1.4847442761063576]  0.998617  0.5510\n",
       "5   [1.6594213886260987]  0.936941  0.5570\n",
       "6   [1.7852485501468183]  0.986875  0.5560\n",
       "7   [1.7116678630709647]  0.982495  0.5495\n",
       "8   [1.5778338895440103]  0.975589  0.5430\n",
       "9   [1.5465353992581368]  0.944386  0.5615\n",
       "10    [1.61626892131567]  0.948060  0.5850\n",
       "11  [1.6108183098435402]  0.906177  0.5755\n",
       "12   [1.669055641502142]  0.890414  0.5745\n",
       "13   [1.658261726796627]  0.906241  0.5845\n",
       "14  [1.7189348363280297]  0.886755  0.5830"
      ]
     },
     "execution_count": 89,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results200 = pd.DataFrame({'val': xvalidate_loss, 'train':xtrain_loss, 'acc':xaccuracy})\n",
    "results200"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 172,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "results100 = pd.DataFrame({'val': validate_loss, 'train':train_loss, 'acc':accuracy})\n",
    "results100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 210,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "results100.to_csv('results100.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# http://forums.fast.ai/t/fast-ai-predict-array-for-nlp-classification/17180\n",
    "\n",
    "# df_pred = pd.read_csv(LM_PATH/'predict.csv', header=None, chunksize=chunksize)\n",
    "# tok_pred, pred_labels = get_all(df_pred, 1)\n",
    "# pred_ids = np.array([[stoi[o] for o in p] for p in tok_pred])\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}