-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathprinters.py
148 lines (131 loc) · 6.19 KB
/
printers.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
import matplotlib.pyplot as plt
import numpy as np
class ShowTraining():
def __init__(self, epochs_num = None, weights_list = None):
plt.ion()
self.fig1 = plt.figure(1)
self.ax = self.fig1.add_subplot(111)
self.ax.set_title('Errors History (J)')
self.train, = self.ax.plot(xrange(len([])), [], color='green', marker='^', label='Training')
self.test, = self.ax.plot(xrange(len([])), [], color='blue', marker='s', label='Test')
self.ax.set_xlabel('Epochs')
self.ax.set_ylabel(r'$||J||_2/N$')
self.ax.legend()
if epochs_num is not None:
self.ax.set_xlim([0,epochs_num])
self.fig2 = plt.figure(2)
self.ax2 = self.fig2.add_subplot(111)
self.ax2.set_title('Loss Gradient History (dJ/dy)')
self.dJdy, = self.ax2.plot(xrange(len([])), [], color='red', marker='o')
self.ax2.set_xlabel('Epochs')
self.ax2.set_ylabel(r'$||\delta J/\delta y||_2/N$')
if epochs_num is not None:
self.ax2.set_xlim([0,epochs_num])
self.weightsline = []
self.weights = []
if weights_list is not None:
self.fig3 = plt.figure(3)
self.ax3 = self.fig3.add_subplot(111)
self.ax3.set_title('Weights Norm2')
for weight in weights_list:
self.weights.append(weights_list[weight])
line, = self.ax3.plot(xrange(len([])), [], marker='o', label=weight)
self.weightsline.append(line)
self.ax3.set_xlabel('Epochs')
self.ax3.set_ylabel(r'$||W_i||_2$')
if epochs_num is not None:
self.ax3.set_xlim([0,epochs_num])
self.ax3.legend()
plt.ioff()
def show(self, epoch, J_train_list, dJdy_list = None, J_test_list = None):
plt.ion()
self.train.set_xdata(xrange(len(J_train_list[:epoch+1])))
self.train.set_ydata(J_train_list[:epoch+1])
if J_test_list is not None:
self.test.set_xdata(xrange(len(J_test_list[:epoch+1])))
self.test.set_ydata(J_test_list[:epoch+1])
self.ax.set_ylim([0,max(max(J_train_list)+max(J_train_list)*0.1,max(J_test_list)+max(J_test_list)*0.1)])
else:
self.ax.set_ylim([0,max(J_train_list)+max(J_train_list)*0.1])
self.fig1.canvas.draw()
self.dJdy.set_xdata(xrange(len(dJdy_list[:epoch+1])))
self.dJdy.set_ydata(dJdy_list[:epoch+1])
self.ax2.set_ylim([min(dJdy_list)-min(dJdy_list)*0.1,max(dJdy_list)+max(dJdy_list)*0.1])
self.fig2.canvas.draw()
if len(self.weights) > 0:
minval = 0
maxval = 0
for weight,weightline in zip(self.weights,self.weightsline):
if epoch == 0:
weightline.set_ydata([])
weightline.set_xdata(xrange(len(dJdy_list[:epoch+1])))
weightline.set_ydata(np.append(weightline.get_ydata(),np.linalg.norm(weight.get())))
minval = minval if minval < min(weightline.get_ydata()) else min(weightline.get_ydata())
maxval = maxval if maxval > max(weightline.get_ydata()) else max(weightline.get_ydata())
self.ax3.set_ylim([minval-minval*0.1,maxval+maxval*0.1])
self.fig3.canvas.draw()
plt.ioff()
class Printer2D():
def forward_all(self, model, xs):
return map(lambda x: model.forward(x), xs)
def draw_decision_surface(self, figure_ind, model, data):
# max = np.max([i[0] for i in data],0)
# min = np.min([i[0] for i in data],0)
max = np.max(data,0)
min = np.min(data,0)
x_range = np.linspace(min[0]-0.5,max[0]+0.5,100)
y_range = np.linspace(min[1]-0.5,max[1]+0.5,100)
X, Y = np.meshgrid(x_range, y_range)
Z = np.argmax(self.forward_all(model, np.c_[X.ravel(), Y.ravel()]), axis=1)
Z = Z.reshape(X.shape)
# cs = plt.contourf(xx, yy, Z, cmap='Paired')
plt.figure(figure_ind)
plt.set_cmap(plt.cm.Paired)
plt.pcolormesh(X, Y, Z)
plt.colorbar()
def print_model(self, figure_ind, model, x_list, print_layers = None):
max = np.max(x_list,0)
min = np.min(x_list,0)
x_range = np.linspace(min[0],max[0],100)
y_range = np.linspace(min[1],max[1],100)
X, Y = np.meshgrid(x_range, y_range)
if print_layers == None:
print_layers = range(len(model.elements))
for ind in print_layers:
plt.figure(figure_ind+ind)
z_vett = []
for x_ind in xrange(X.shape[0]):
for y_ind in xrange(Y.shape[1]):
aux_x = np.array([X[x_ind][y_ind],Y[x_ind][y_ind]])
for layer in model.elements[:ind]:
aux_x = layer.forward(aux_x)
z_vett.append(aux_x)
z_array = np.array(z_vett)
for exit in range(z_array.shape[1]):
z_array_out = z_array[:,exit]
Z = z_array_out.reshape(X.shape)
plt.subplot(z_array.shape[1], 1, exit+1)
try:
CS = plt.contour(X, Y, Z)
except ValueError:
pass
def compare_data(self, figure_ind, train, output, num_classes, colors, classes):
xy = range(num_classes)
for type in range(num_classes):
xy[type] = []
for type in range(num_classes):
for i, (x,t) in enumerate(train):
xy[np.argmax(t)].append((x[0],x[1],np.argmax(output[i])))
plt.figure(figure_ind)
for type in range(num_classes):
for x,y,c in xy[type]:
plt.scatter(x, y, s=100, color=colors[c], marker=classes[type])
def print_data(self, figure_ind, data, targets, num_classes, colors, classes):
x = range(num_classes)
y = range(num_classes)
for type in range(num_classes):
x[type] = [point[0] for i, point in enumerate(data) if np.argmax(targets[i]) == type]
y[type] = [point[1] for i, point in enumerate(data) if np.argmax(targets[i]) == type]
plt.figure(figure_ind)
for type in range(num_classes):
plt.scatter(x[type], y[type], s=100, color=colors[type], marker=classes[type])