-
Notifications
You must be signed in to change notification settings - Fork 0
/
P3.3.py
194 lines (175 loc) · 7.08 KB
/
P3.3.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
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
import torch
from torch import nn, optim
from torch.utils.data import TensorDataset, DataLoader
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
import numpy as np
LR_1 = 0.0001
LR_2 = 0.001
MAX_EPOCH = 10
BATCH_SIZE = 64
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
def simpleFunction(x):
val = (2 * np.sin(5 * np.pi * x)) / (5 * np.pi * x)
return val
def accuracy(y_pred, y_true):
if isinstance(y_true, torch.Tensor):
y_true = y_true.detach().cpu().numpy()
if isinstance(y_pred, torch.Tensor):
y_pred = y_pred.detach().cpu().numpy()
correct = sum(1 for true, pred in zip(y_true, y_pred) if ((pred >= true - 1) and (pred <= true + 1)))
total = len(y_true)
acc = correct / total
return acc
class FunctionApproximator1(nn.Module):
def __init__(self):
super(FunctionApproximator1, self).__init__()
self.regressor = nn.Sequential(nn.Linear(1, 20),
nn.ReLU(inplace=True),
nn.Linear(20, 40),
nn.ReLU(inplace=True),
nn.Linear(40, 40),
nn.ReLU(inplace=True),
nn.Linear(40, 20),
nn.ReLU(inplace=True),
nn.Linear(20, 1))
def forward(self, x):
output = self.regressor(x)
return output
def interpolate_models(model_1, model_2, alpha):
new_model = FunctionApproximator1()
for (name1, param1), (name2, param2) in zip(model_1.named_parameters(), model_2.named_parameters()):
new_param = alpha * param1.data + (1 - alpha) * param2.data
setattr(new_model, 'param', nn.Parameter(new_param))
return new_model
X = np.random.rand(10 ** 5)
y = simpleFunction(X)
X_train, X_val, y_train, y_val = map(torch.tensor, train_test_split(X, y, test_size=0.2))
train_dataloader = DataLoader(TensorDataset(X_train.unsqueeze(1), y_train.unsqueeze(1)), batch_size=BATCH_SIZE,
pin_memory=True, shuffle=True)
val_dataloader = DataLoader(TensorDataset(X_val.unsqueeze(1), y_val.unsqueeze(1)), batch_size=BATCH_SIZE,
pin_memory=True, shuffle=True)
model1 = FunctionApproximator1().to(device)
model2 = FunctionApproximator1().to(device)
optimizer1 = optim.Adam(model1.parameters(), lr=LR_1)
optimizer2 = optim.Adam(model2.parameters(), lr=LR_2)
criterion = nn.MSELoss(reduction="mean")
# training loop for model 1
train_loss_list = []
val_loss_list = []
for epoch in range(MAX_EPOCH):
print("epoch %d / %d" % (epoch + 1, MAX_EPOCH))
model1.train()
# training loop
temp_loss_list = []
for X_train, y_train in train_dataloader:
X_train = X_train.type(torch.float32).to(device)
y_train = y_train.type(torch.float32).to(device)
optimizer1.zero_grad()
score = model1(X_train)
loss = criterion(input=score, target=y_train)
loss.backward()
optimizer1.step()
temp_loss_list.append(loss.detach().cpu().numpy())
train_loss_list.append(np.average(temp_loss_list))
# validation
model1.eval()
temp_loss_list = []
for X_val, y_val in val_dataloader:
X_val = X_val.type(torch.float32).to(device)
y_val = y_val.type(torch.float32).to(device)
score = model1(X_val)
loss = criterion(input=score, target=y_val)
temp_loss_list.append(loss.detach().cpu().numpy())
val_loss_list.append(np.average(temp_loss_list))
print(" train loss: %.5f" % train_loss_list[-1])
print(" val loss: %.5f" % val_loss_list[-1])
# training loop for model 2 (trained same way just using different models to hopefully get different result)
train_loss_list = []
val_loss_list = []
for epoch in range(MAX_EPOCH):
print("epoch %d / %d" % (epoch + 1, MAX_EPOCH))
model2.train()
# training loop
temp_loss_list = []
for X_train, y_train in train_dataloader:
X_train = X_train.type(torch.float32).to(device)
y_train = y_train.type(torch.float32).to(device)
optimizer2.zero_grad()
score = model2(X_train)
loss = criterion(input=score, target=y_train)
loss.backward()
optimizer2.step()
temp_loss_list.append(loss.detach().cpu().numpy())
train_loss_list.append(np.average(temp_loss_list))
# validation
model2.eval()
temp_loss_list = []
for X_val, y_val in val_dataloader:
X_val = X_val.type(torch.float32).to(device)
y_val = y_val.type(torch.float32).to(device)
score = model2(X_val)
loss = criterion(input=score, target=y_val)
temp_loss_list.append(loss.detach().cpu().numpy())
val_loss_list.append(np.average(temp_loss_list))
alpha_vals = np.linspace(0,1,10)
fin_tr_loss = []
fin_te_loss = []
fin_tr_acc = []
fin_te_acc = []
for alpha in alpha_vals:
lin_model = interpolate_models(model1, model2, alpha)
train_loss_list = []
val_loss_list = []
train_acc_list = []
test_acc_list = []
temp_loss_list = []
temp_tr_acc_list = []
temp_te_acc_list = []
for epoch in range(MAX_EPOCH):
print("epoch %d / %d" % (epoch + 1, MAX_EPOCH))
lin_model.train()
# training loop
correct = 0
temp_acc = []
for X_train, y_train in train_dataloader:
X_train = X_train.type(torch.float32).to(device)
y_train = y_train.type(torch.float32).to(device)
optimizer2.zero_grad()
score = lin_model(X_train)
loss = criterion(input=score, target=y_train)
loss.backward()
temp_acc.append(accuracy(score, y_train))
optimizer2.step()
temp_loss_list.append(loss.detach().cpu().numpy())
train_loss_list.append(np.average(temp_loss_list))
temp_tr_acc_list.append(np.average(temp_acc))
# validation
lin_model.eval()
temp_loss_list = []
temp_acc = []
for X_val, y_val in val_dataloader:
X_val = X_val.type(torch.float32).to(device)
y_val = y_val.type(torch.float32).to(device)
score = lin_model(X_val)
loss = criterion(input=score, target=y_val)
temp_acc.append(accuracy(score, y_val))
temp_loss_list.append(loss.detach().cpu().numpy())
val_loss_list.append(np.average(temp_loss_list))
temp_te_acc_list.append(np.average(temp_acc))
fin_tr_loss.append(train_loss_list[-1])
fin_te_loss.append(val_loss_list[-1])
fin_tr_acc.append(np.average(temp_tr_acc_list))
fin_te_acc.append(np.average(temp_te_acc_list))
fig1, (ax1, ax2) = plt.subplots(1,2)
ax1.plot(alpha_vals, fin_tr_loss, color='r', label='train_loss')
ax1.plot(alpha_vals, fin_te_loss, color='g', label='test_loss')
ax1.set_xlabel('alpha')
ax1.set_ylabel('loss')
ax1.legend()
ax2.plot(alpha_vals, fin_tr_acc, color='r', label='train_acc')
ax2.plot(alpha_vals, fin_te_acc, color='g', label='test_acc')
ax2.set_xlabel('alpha')
ax2.set_ylabel('accuracy')
ax2.legend()
fig1.savefig("P3.3.1.png")