-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsolver_FC.py
163 lines (143 loc) · 7.21 KB
/
solver_FC.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
import os
os.environ['KMP_DUPLICATE_LIB_OK']='True'
import time
import numpy as np
import torch
import torch.nn as nn
from matplotlib import pyplot as plt
DELTA_CLIP = 50.0
class BSDESolver(object):
"""The fully connected neural network model."""
def __init__(self, config, bsde): #传入字典config和equation类bdse
self.eqn_config = config["eqn_config"]
self.net_config = config["net_config"]
self.bsde = bsde
self.model = NonsharedModel(config, bsde) #model
self.y_init = self.model.y_init #
# lr_schedule = tf.keras.optimizers.schedules.PiecewiseConstantDecay(
# self.net_config.lr_boundaries, self.net_config.lr_values)
self.optimizer = torch.optim.Adam(self.model.parameters(), lr=0.01)
#optimizer_1 = torch.optim.Adam(net_1.parameters(), lr=initial_lr)
#scheduler_1 = toLambdaLR(optimizer_1, lr_lambda=lambda epoch: 1 / (epoch + 1))
def train(self):
start_time = time.time()
training_history = []
training_time=[]
# begin sgd iteration
for step in range(self.net_config["num_iterations"]+1):
valid_data = self.bsde.sample(self.net_config["valid_size"])
loss = self.loss_fn(valid_data,True)
if step % self.net_config["logging_frequency"] == 0:
y_init = self.model.y_init
elapsed_time = time.time() - start_time
training_history.append(y_init.item())
training_time.append(step)
if self.net_config["verbose"]:
print("step: %d, loss: %.4f, Y0: %.4f, elapsed time: %3f" % (
step, loss.data.item(), float(y_init), elapsed_time))
#将history数据加载到列表中，没有打印
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
plt.plot(np.array(training_time),np.array(training_history))
plt.show()
# self.train_step(self.bsde.sample(self.net_config.batch_size))
#训练函数，返回training_history列表
return np.array(training_history)
def loss_fn(self, inputs, training):
dw, x = inputs
y_terminal = self.model(inputs, training)
delta = y_terminal - self.bsde.g_tf(self.bsde.total_time, x[:, :, -1])
# use linear approximation outside the clipped range
loss = torch.sum(torch.where(torch.abs(delta) < DELTA_CLIP, torch.square(delta),
2 * DELTA_CLIP * torch.abs(delta) - DELTA_CLIP ** 2))
#一种奇怪的loss函数
return loss
# def grad(self, inputs, training):
# with tf.GradientTape(persistent=True) as tape:
# loss = self.loss_fn(inputs, training)
# grad = tape.gradient(loss, self.model.trainable_variables)
# del tape
# return grad
# @tf.function
# def train_step(self, train_data):
# grad = self.grad(train_data, training=True)
# self.optimizer.apply_gradients(zip(grad, self.model.trainable_variables))
#这几个函数类似于torch中optimizer.step
class NonsharedModel(nn.Module): #相当于nn.Module
def __init__(self, config, bsde):
super(NonsharedModel, self).__init__()
self.eqn_config = config["eqn_config"]
self.net_config = config["net_config"]
self.bsde = bsde
self.y_init = nn.Parameter(torch.from_numpy(np.random.uniform(low=self.net_config["y_init_range"][0],
high=self.net_config["y_init_range"][1],
size=[1])).float()
) #u0 类似于nn.parameter
self.z_init = nn.Parameter(torch.from_numpy(np.random.uniform(low=-.1, high=.1,
size=[1, self.eqn_config["dim"]])).float()
) #delta_u0:1*N
self.subnet = nn.ModuleList([FeedForwardSubNet(config) for _ in range(self.bsde.num_time_interval-1)]
) #建立N-1个自网络
def forward(self, inputs, training): #forward
dw, x = inputs #dw:bat*dim*N x:bat*dim*N+1
time_stamp = np.arange(0, self.eqn_config["num_time_interval"]) * self.bsde.delta_t
#[delt,2delt....N-1 delt]
all_one_vec = torch.ones(self.net_config["batch_size"],1)
#这里函数调用太奇怪了
#all_one_vec:bat*1
y = all_one_vec * self.y_init
#u0:batch*1
z = torch.matmul(all_one_vec, self.z_init)
#delta_u0:batch*N
for t in range(0, self.bsde.num_time_interval-1):
y = y - self.bsde.delta_t * (
self.bsde.f_tf(time_stamp[t], x[:, :, t], y, z)
) + torch.sum(z * dw[:, :, t], 1, keepdim=True)
z = self.subnet[t](x[:, :, t + 1], training) / self.bsde.dim
# terminal time
y = y - self.bsde.delta_t * self.bsde.f_tf(time_stamp[-1], x[:, :, -2], y, z) + \
torch.sum(z * dw[:, :, -1], 1, keepdim=True)
return y
#batch*1
class FeedForwardSubNet(nn.Module):
def __init__(self, config):
super(FeedForwardSubNet, self).__init__()
dim = config["eqn_config"]["dim"]
num_hiddens = config["net_config"]["num_hiddens"]
#len(num_hiddens) + 2层batchnormalize
self.bn_layers=nn.ModuleList([])
self.bn_layers.append(nn.BatchNorm1d(
num_features=dim,
momentum=0.99,
eps=1e-6,
)
)
for k in range(0,len(num_hiddens)):
self.bn_layers.append(nn.BatchNorm1d(num_features=num_hiddens[k],momentum=0.99,eps=1e-6,))
self.bn_layers.append(nn.BatchNorm1d(
num_features=dim,
momentum=0.99,
eps=1e-6,
))
# self.dense_layers = nn.ModuleList([tf.keras.layers.Dense(num_hiddens[i],
# use_bias=False,
# activation=None)
# for i in range(len(num_hiddens))]) #全连接层？'''
self.dense_layers=nn.ModuleList([])
self.dense_layers.append(nn.Linear(dim,num_hiddens[0]))
for i in range(len(num_hiddens)-1):
self.dense_layers.append(nn.Linear(num_hiddens[i],num_hiddens[i+1]))
self.dense_layers.append(nn.Linear(num_hiddens[len(num_hiddens)-1],dim))
# final output should be gradient of size dim
# self.dense_layers.append(tf.keras.layers.Dense(dim, activation=None))
def forward(self, x, training):
"""structure: bn -> (dense -> bn -> relu) * len(num_hiddens) -> dense -> bn"""
x = self.bn_layers[0](x)
for i in range(len(self.dense_layers) - 1):
x = self.dense_layers[i](x)
x = self.bn_layers[i+1](x)
x = nn.functional.relu(x)
x = self.dense_layers[-1](x)
x = self.bn_layers[-1](x)
return x #每个自网络经多层全连接与batchnorm层