File size: 5,156 Bytes
ec69f9c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
# %%
# import libraries
import numpy as np
import torch 
import torch.nn as nn
import matplotlib.pyplot as plt
from IPython import display
display.set_matplotlib_formats('svg')
# %%
# create data

N = 30
x = torch.randn(N, 1)
y = x + torch.randn(N, 1)/2

# and plot 
plt.plot(x,y,'s')
plt.show()
# %%
# build model 
ANNreg = nn.Sequential(
    nn.Linear(1, 1), # input layer
    nn.ReLU(), # activation function
    nn.Linear(1, 1) # output layer
)
ANNreg
# %%
# learning rate 
learning_rate = .05

# loss function
loss_fn = nn.MSELoss()

# optimizer (the flavor of gradient descent to implement)
optimizer = torch.optim.SGD(ANNreg.parameters(), lr=learning_rate)
# %%
# train the model
num_epochs = 500
losses = torch.zeros(num_epochs)

# Train the model 
for epochi in range(num_epochs):

    # forward pass
    yhat = ANNreg(x)

    # compute loss
    loss = loss_fn(yhat, y)
    losses[epochi] = loss

    # backpropagation
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()
# %%
# show the losses

# manually compute losses
# final forward pass
predictions = ANNreg(x)

# final loss (MSE)
test_loss = loss_fn(predictions, y).pow(2).mean()

plt.plot(losses.detach(), 'o', markerfacecolor='w', linewidth=.1)
plt.plot(num_epochs, test_loss.detach(), 'ro')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.title('Final loss: %.3f' % test_loss.item())
plt.show()
# %%
test_loss.item()
# %%
# plot the data 
plt.plot(x,y,'bo', label='Real data')
plt.plot(x, predictions.detach(), 'rs', label='Predictions')
plt.title(f'preiction-data r = {np.corrcoef(y.T, predictions.detach().T)[0,1]:.2f}')
plt.legend()
plt.show()
# %%
def Model(x,y, num_epochs, learning_rate):

    # build model
    ANNreg =nn.Sequential(
        nn.Linear(1,1),
        nn.ReLU(),
        nn.Linear(1,1)
    )

    # loss function
    loss_fn = nn.MSELoss()

    # optimizer
    optimizer = torch.optim.SGD(ANNreg.parameters(), lr=learning_rate)

    # train the model
    losses = torch.zeros(num_epochs)

    for epoch in range(num_epochs):
        # forward pass
        yHat = ANNreg(x)

        # compute loss
        loss = loss_fn(yHat,y)
        losses[epoch] = loss

        # backpropagation
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

    final_predictions = ANNreg(x)


    return final_predictions, losses

N = 30
x = torch.randn(N, 1)
y = x + torch.randn(N, 1)/2
final_predictions, losses = Model(x,y, num_epochs=500, learning_rate=.05)
final_predictions 

# %%
# A function that creates  and trains the model

def buildAndTrainTheModel(x,y):

    # build the model
    ANNreg = nn.Sequential(
        nn.Linear(1,1), # input layer
        nn.ReLU(),      # activation function
        nn.Linear(1,1)  # output layer
    )

    # loss and optimizer functions
    loss_fn = nn.MSELoss()
    optimizer = torch.optim.SGD(ANNreg.parameters(), lr=.05)

    # train the model
    num_epochs = 500
    losses = torch.zeros(num_epochs)

    for epochi in range(num_epochs):

        # forward pass
        yHat = ANNreg(x)

        # compute loss
        loss = loss_fn(yHat, y)
        losses[epochi] = loss

        # backpropagation
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        # end traing loop

    # compute model predictions
    predictions = ANNreg(x)

    return predictions, losses

# %%
# A function that creates the data

def createTheData(m):
    N = 30
    x = torch.randn(N,1)
    y = m*x + torch.randn(N,1)/2
    return x, y
# %%
# Test it once 

# create a dataset
x, y = createTheData(.8)

# run the model 
yhat, losses = buildAndTrainTheModel(x,y)

fig, ax = plt.subplots(1,2, figsize=(12,4))

ax[0].plot(losses.detach(),'o',markerfacecolor='w',linewidth=.1)
ax[0].set_xlabel('Epoch')
ax[0].set_title('loss')

ax[1].plot(x,y,'bo',label='Real data')
ax[1].plot(x,yhat.detach(),'rs',label='Predictions')
ax[1].set_xlabel('x')
ax[1].set_ylabel('y')
ax[1].set_title(f'prediction-data corr = {np.corrcoef(y.T, yhat.detach().T)[0,1]:.2f}')
ax[1].legend()
plt.show()

# %%
# Now for the expriment!

# (takes  3 mns with 21 slopes and 50 exps)

# the slopes to simulate
slopes = np.linspace(-2,2,21)

numExps = 50

# intialize output matrix
results = np.zeros((len(slopes), numExps,2))

for slopi in range(len(slopes)):
    for expi in range(numExps):

        # create data
        x, y = createTheData(slopes[slopi])

        # run the model
        yhat, losses = buildAndTrainTheModel(x,y)

        # store the results
        results[slopi, expi, 0] = losses[-1]
        results[slopi, expi, 1] = np.corrcoef(y.T, yhat.detach().T)[0,1]

# correlation can be 0 if the model didn't do well. Set nan's -> 0
results[np.isnan(results)] = 0
# %%
# plot the results!

fig, ax = plt.subplots(1,2, figsize=(12,4))

ax[0].plot(slopes, np.mean(results[:,:,0], axis=1), 'ko-', markerfacecolor='w', markersize=10)
ax[0].set_xlabel('Slope')
ax[0].set_title('Loss')

ax[1].plot(slopes, np.mean(results[:,:,1], axis=1), 'ms-', markerfacecolor='w', markersize=10)
ax[1].set_xlabel('Slope')
ax[1].set_ylabel('Real-predicted Correlation')
ax[1].set_title('Model performance')

plt.show()
# %%