-
Notifications
You must be signed in to change notification settings - Fork 0
/
log_regression_test.py
37 lines (29 loc) · 1.19 KB
/
log_regression_test.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
from models.logistic_regression import LogisticRegression
from utils.loss import Loss
import matplotlib.pyplot as plt
from utils.utils import get_data, get_binary_data, normalize_data
"""
A demonstration of the LogisticRegression Model from the models directory.
Uses the MNIST dataset with the 0 and 1 class label data to perform binary
predictions.
Even with this simple model, the costs converge near 0 for train and
test sets and the classification rate is nearly 100% for both.
"""
def main():
# Load in data for classes 0 & 1 and normalize
Xtrain, Xtest, Ytrain, Ytest = get_data()
Xtrain, Ytrain = get_binary_data(Xtrain, Ytrain)
Xtest, Ytest = get_binary_data(Xtest, Ytest)
Xtrain, Xtest = normalize_data(Xtrain, Xtest)
model = LogisticRegression()
model.compile(loss=Loss.BCE)
model.fit(Xtrain, Ytrain, validation_data=(Xtest, Ytest), epochs=1000, learning_rate=0.00001)
# Plot the costs over training epochs
plt.plot(model.history["train_costs"], label="train_costs")
plt.plot(model.history["test_costs"], label="test_costs")
plt.legend()
plt.show()
# Log the classification rate
model.score(Ytrain, Ytest)
if __name__ == '__main__':
main()