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
|
class Model:
def __init__(self, weights_filename='weights'):
self.weights_filename = weights_filename
def train(self, xs, ys, alpha=1, epoch=1000):
for _ in range(epoch):
theta = theta - alpha * self.gradient(xs, ys)
def train_against(self, xs, ys, theta, one, alpha, epoch):
ys_ally = ys.copy()
ys_ally[ys == one] = 0
ys_ally[ys != one] = 1
return gradient_descent(xs, ys_ally, theta, alpha, epoch)
def train_thetas(xs, ys, theta, alpha=1, epoch=1000):
thetas = []
for i in np.unique(ys):
thetas.append(train_against(xs, ys, theta, i, alpha, epoch))
return thetas
def gradient(self, xs, ys):
return np.array([self.partial(xs, ys, i) for i in range(len(self.theta))])
def partial(self, xs, ys, theta_j):
total = 0
for x_i, y_i in zip(xs, ys):
temp = self.hypothesis(x_i) - y_i
if theta_j != 0:
temp *= x_i[theta_j - 1]
total += temp
return total / len(xs)
def predict(self, x):
return 1 if self.hypothesis(x) >= 0.5 else 0
def hypothesis(self, x):
return self._sigmoid(x.dot(self.theta))
def logloss(self, x, y):
if y == 1:
return -np.ln(self.hypothesis(x))
elif y == 0:
return -np.ln(1 - self.hypothesis(x))
else:
raise "y != 1 and y != 0"
def cost(self, xs, ys):
return sum([self.logloss(x, y) for x, y in zip(xs, ys)]) / len(xs)
def _sigmoid(self, x):
return 1 / (1 + np.exp(-x))
def _normalize(self, x):
return (x - x.min()) / (x.max() - x.min())
def _read_weights(self):
try:
with open(self.weights_filename, 'r') as file:
self.weights = np.array(
[float(s) for s in file.read().strip().split(',')])
except IOError:
raise 'Couldn\'t read weights file at: {}'.format(self.weights_filename)
def _write_weights(self):
try:
with open(self.weights_filename, 'w') as file:
file.write(','.join([str(w) for w in self.weights])
except IOError:
raise 'Couldn\'t write weights file at: {}'.format(self.weights_filename)
|
