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
|
#!/bin/python3
import sys
import pandas as pd
import numpy as np
from dataset import Dataset
def sigmoid(x):
return 1.0 / (1.0 * np.exp(-x))
def hypothesis(x, theta):
return sigmoid(x.dot(theta))
def gradient(ys, xs, theta):
g = np.zeros(len(xs[0]))
for j in range(len(theta)):
g[j] = sum([(hypothesis(x, theta) - y) * x[j] for y, x in zip(ys, xs)]) / len(xs)
return g
def gradient_descent(ys, xs, alpha, epoch):
theta = np.random.randn(len(xs[0]))
for i in range(epoch):
print("Gradient descent: {:02}%\r".format(int((i / epoch) * 100.0)), end="")
theta = theta - alpha * gradient(ys, xs, theta)
return theta
def train(ys, xs):
thetas = []
# print(np.unique(ys))
for trained in np.unique(ys):
print(f"Trainning against {trained}")
ys_ally = ys.copy()
ys_ally[ys == trained] = 0 # opposite?
ys_ally[ys != trained] = 1
thetas.append((trained, gradient_descent(ys_ally, xs, 1, 100)))
return thetas
if __name__ == '__main__':
if len(sys.argv) != 2:
raise 'Usage: {} dataset_path'.format(sys.argv[0])
d = Dataset(sys.argv[1])
X = d.df_scores.values
X = np.hstack([X, np.ones((X.shape[0], 1))])
X = (X - X.min()) / (X.max() - X.min())
Y = d.df["house"].values
thetas = train(Y, X)
with open("weights", "w") as f:
for label, t in thetas:
f.write("{}: {}\n".format(label, ','.join([str(x) for x in t])))
|
