-
Notifications
You must be signed in to change notification settings - Fork 5
/
testing_autoencoder_and_neural_network.py
199 lines (150 loc) · 7.62 KB
/
testing_autoencoder_and_neural_network.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
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
import numpy as np
import pandas as pd
from tensorflow.keras.models import Sequential
from tensorflow.keras import Model
from tensorflow.keras.layers import Dense, Dropout, BatchNormalization
from pandas import read_csv
from pandas import DataFrame
from tensorflow.keras import backend as K
import matplotlib.pyplot as plt
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import train_test_split
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.callbacks import EarlyStopping
def rmse(y_true, y_pred):
return K.sqrt(K.mean(K.square(y_pred - y_true), axis=-1))
def get_euclidean_distances_and_mean(points1, points2):
x1, y1 = points1
x2, y2 = points2
x1, y1 = np.array(x1), np.array(y1)
x2, y2 = np.array(x2), np.array(y2)
distance_x = x1 - x2
distance_y = y1 - y2
squared_distance_x = distance_x ** 2
squared_distance_y = distance_y ** 2
distances = squared_distance_x + squared_distance_y
euclidean_distances = np.sqrt(distances)
return euclidean_distances, np.mean(euclidean_distances)
def create_neural_network(input_dimension):
model = Sequential()
model.add(Dense(50, input_dim=input_dimension, activation='sigmoid'))
model.add(BatchNormalization())
model.add(Dense(50, activation='relu'))
model.add(Dense(50, activation='relu'))
model.add(Dense(2, activation='relu'))
model.compile(loss='mse', optimizer=Adam(.001), metrics=['mse'])
return model
def prepare_data_for_training(path_to_file):
data = read_csv(path_to_file, index_col=None)
y = data.iloc[:, -2:]
x = data.iloc[:, :-2]
return train_test_split(x, y, test_size = .2, shuffle = False)
def train_neural_network(neural_network, train_data, validation_data, number_of_epochs, batch_size):
es = EarlyStopping(monitor = 'val_loss', patience = 100, verbose = 0, mode = 'auto', restore_best_weights = True)
return neural_network.fit(x = train_data[0], y = train_data[1], validation_data = validation_data, epochs=number_of_epochs, batch_size=batch_size, verbose=0, callbacks = [es])
def plot_history(history):
if 'mse' in history.history.keys():
plt.plot(history.history['mse'])
plt.plot(history.history['val_mse'])
plt.title('Mean squared error')
plt.ylabel('Error')
plt.xlabel('Epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
if 'loss' in history.history.keys():
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('Model loss')
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
if 'accuracy' in history.history.keys():
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('Model accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
def plot_cumulative_density_function(distances, save_file_name):
sorted_distances = np.sort(distances)
prob_deep = 1. * np.arange(len(sorted_distances))/(len(sorted_distances) - 1)
_, axes = plt.subplots()
axes.plot(sorted_distances, prob_deep, color='black')
plt.title('CDF of Euclidean distance error')
plt.xlabel('Distance (m)')
plt.ylabel('Probability')
plt.grid(True)
gridlines = axes.get_xgridlines() + axes.get_ygridlines()
for line in gridlines:
line.set_linestyle('-.')
# 'Figure_CDF_error.png'
plt.savefig(save_file_name, dpi=300)
plt.show()
plt.close()
def run_script_on_data(filename):
train_x, validation_x, train_y, validation_y = prepare_data_for_training(filename)
neural_network = create_neural_network(train_x.shape[1])
history = train_neural_network(neural_network, (train_x, train_y), (validation_x, validation_y), 1000, 1000)
predictions = neural_network.predict(validation_x)
_, mean_distance = get_euclidean_distances_and_mean((predictions[:, 0], predictions[:, 1]),
(validation_y["x"], validation_y["y"]))
print('Mean distance: ' + str(mean_distance))
plot_history(history)
def create_autoencoder(input_dimension):
model = Sequential()
model.add(Dense(50, input_dim=input_dimension, activation='sigmoid'))
model.add(BatchNormalization())
model.add(Dense(25, activation='relu'))
model.add(Dense(8, activation='relu'))
model.add(Dense(25, activation='relu'))
model.add(Dense(50, activation='relu'))
model.add(Dense(input_dimension, activation='sigmoid')) #13
model.compile(optimizer=Adam(.0001, clipnorm = .5, clipvalue = .5),
loss='mse', metrics=['accuracy'])
return model
def build_autoencoder_neural_network(autoencoder):
predictions = Dense(2, activation = 'relu')(autoencoder.layers[3].output)
neural_network = Model(inputs=autoencoder.input, outputs=predictions)
return neural_network
def set_trainable_layers(model):
for layer in model.layers[:-2]:
layer.trainable = False
def set_new_trainable_layers(model):
for layer in model.layers[:-2]:
layer.trainable = True
def run_autoencoder_scenario_on_data(path_to_unlabeled_data, path_to_labeled_data):
unlabeled_data = read_csv(path_to_unlabeled_data, index_col=None)
train_unlabeled, validation_unlabeled = train_test_split(unlabeled_data, test_size = .2, shuffle = False)
autoencoder = create_autoencoder(train_unlabeled.shape[1])
history = train_neural_network(autoencoder, (train_unlabeled, train_unlabeled), (validation_unlabeled, validation_unlabeled), 1000, 1000)
plot_history(history)
neural_network = build_autoencoder_neural_network(autoencoder)
set_trainable_layers(neural_network)
neural_network.compile(optimizer=Adam(.001, clipnorm = .5, clipvalue = .5),
loss=rmse, metrics=['accuracy'])
train_x, validation_x, train_y, validation_y = prepare_data_for_training(path_to_labeled_data)
history = train_neural_network(neural_network, (train_x, train_y), (validation_x, validation_y), 1000, 50)
plot_history(history)
predictions = neural_network.predict(validation_x)
distances, mean_distance = get_euclidean_distances_and_mean((predictions[:, 0], predictions[:, 1]),
(validation_y["x"], validation_y["y"]))
plot_cumulative_density_function(distances, 'CDF_train_new_decoder.png')
print('Mean distance: ' + str(mean_distance))
set_new_trainable_layers(neural_network)
neural_network.compile(optimizer=Adam(.001, clipnorm = .5, clipvalue = .5),
loss=rmse, metrics=['accuracy'])
train_x, validation_x, train_y, validation_y = prepare_data_for_training(path_to_labeled_data)
history = train_neural_network(neural_network, (train_x, train_y), (validation_x, validation_y), 1000, 50)
plot_history(history)
predictions = neural_network.predict(validation_x)
distances, mean_distance = get_euclidean_distances_and_mean((predictions[:, 0], predictions[:, 1]),
(validation_y["x"], validation_y["y"]))
plot_cumulative_density_function(distances, 'CDF_train_full_neural_network.png')
print('Mean distance: ' + str(mean_distance))
if __name__ == "__main__":
run_script_on_data('data_labeled.csv')
run_script_on_data('labeled_data_scaled.csv')
run_autoencoder_scenario_on_data('extracted_features_scaled_unlabeled.csv', 'extracted_features_scaled.csv')
run_autoencoder_scenario_on_data('unlabeled_data_scaled.csv', 'labeled_data_scaled.csv')