-
Notifications
You must be signed in to change notification settings - Fork 1
/
VAE_Model_1_1.py
230 lines (190 loc) · 7.99 KB
/
VAE_Model_1_1.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
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
import numpy as np
import pandas as pd
import tensorflow as tf
from tensorflow import keras
from keras import layers
from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split
# Read label encoded dataframe from pickle file
label_encoded_df = pd.read_pickle('data\label_encoded_df.pkl')
print(label_encoded_df.head(4))
# Normalize data between 0 and 1
scaler = MinMaxScaler()
scaled_data = scaler.fit_transform(label_encoded_df)
scaled_label_encoded_df = pd.DataFrame(scaled_data, columns = label_encoded_df.columns)
# Split data into train and test sets
train_df, test_df = train_test_split(scaled_label_encoded_df, test_size=0.2, random_state=42)
# Drop column names from train and test dataframes
train_array = train_df.values
test_array = test_df.values
# Drop ID column from the array
train_array = train_array[:, 1:]
test_array = test_array[:, 1:]
# create a new array with shape (2, 5, 6)
reshaped_train_data = np.zeros((train_array.shape[0], 5, 6))
for i in range(reshaped_train_data.shape[0]):
for j in range(reshaped_train_data.shape[1]):
for k in range(reshaped_train_data.shape[2]):
if j == 0:
reshaped_train_data[i,j,k] = int(train_array[i, k])
else:
reshaped_train_data[i,j,k] = int(train_array[i, j+5])
# create a new array with shape (2, 5, 6)
reshaped_test_data = np.zeros((test_array.shape[0], 5, 6))
for i in range(reshaped_test_data.shape[0]):
for j in range(reshaped_test_data.shape[1]):
for k in range(reshaped_test_data.shape[2]):
if j == 0:
reshaped_test_data[i,j,k] = int(test_array[i, k])
else:
reshaped_test_data[i,j,k] = int(test_array[i, j+5])
class VAE(tf.keras.Model):
def __init__(self, latent_dim):
super(VAE, self).__init__()
self.latent_dim = latent_dim
# Encoder
self.encoder = tf.keras.Sequential([
layers.LSTM(16, input_shape=(5,6), return_sequences=False),
layers.Dense(latent_dim + latent_dim),
])
# Decoder
self.decoder = tf.keras.Sequential([
layers.Dense(5*6, activation='relu', input_shape=(latent_dim,)),
layers.Reshape((5, 6)),
layers.Conv1DTranspose(16, 3, activation='relu', padding='same'),
layers.Conv1DTranspose(6, 3, activation='sigmoid', padding='same')
])
def encode(self, x):
mean, logvar = tf.split(self.encoder(x), num_or_size_splits=2, axis=1)
return mean, logvar
def reparameterize(self, mean, logvar):
eps = tf.random.normal(shape=mean.shape)
return eps * tf.exp(logvar * 0.5) + mean
def decode(self, z):
return self.decoder(z)
def call(self, x):
mean, logvar = self.encode(x)
z = self.reparameterize(mean, logvar)
x_recon = self.decode(z)
return x_recon, mean, logvar
# Load your data here
train_data = reshaped_train_data
test_data = reshaped_test_data
# Define the loss function
def vae_loss(x, x_recon, mean, logvar):
# Use mean squared error as the loss function with KL divergence
reconstruction_loss = tf.reduce_mean(tf.square(x - x_recon))
kl_divergence = -0.5 * tf.reduce_mean(1 + logvar - tf.square(mean) - tf.exp(logvar))
return reconstruction_loss + kl_divergence
# Create an instance of the VAE model
latent_dim = 6
vae = VAE(latent_dim)
# Define the optimizer and the batch size
optimizer = tf.keras.optimizers.Adam(learning_rate=0.00001)
batch_size = 32
# Create empty lists to store loss and accuracy
losses = []
accuracies = []
# Train the model
epochs = 30
for epoch in range(epochs):
print('Epoch:', epoch+1)
seed = epoch
for step in range(train_data.shape[0] // batch_size):
x = train_data[step*batch_size : (step+1)*batch_size]
with tf.GradientTape() as tape:
x_recon, mean, logvar = vae(x)
loss = vae_loss(x, x_recon, mean, logvar)
gradients = tape.gradient(loss, vae.trainable_variables)
optimizer.apply_gradients(zip(gradients, vae.trainable_variables))
# Append the loss value to the list
losses.append(loss.numpy())
if step % 100 == 0:
print('Step:', step, 'Loss:', loss.numpy())
# Evaluate the model on the train set after each epoch
x_train = train_data
x_train_recon, _, _ = vae(x_train)
train_loss = vae_loss(x_train, x_train_recon, *vae.encode(x_train))
losses.append(train_loss.numpy())
# Calculate the accuracy of the generated samples
rng = np.random.RandomState(seed)
generated_data = vae.decode(rng.normal(0, 1, size=(train_data.shape[0], latent_dim)))
generated_data = np.where(generated_data > 0.5, 1, 0)
accuracy = np.mean(np.all(generated_data == train_data, axis=(1, 2)))
accuracies.append(accuracy)
print('Train Loss:', train_loss.numpy())
print('Accuracy:', accuracy)
# Plot the loss values over time
import matplotlib.pyplot as plt
step_to_epoch_ratio = int(len(losses) / len(accuracies))
fig, ax = plt.subplots()
ax.plot(losses[::step_to_epoch_ratio], color='b')
#add x-axis label
ax.set_xlabel('Epoch', fontsize=14)
# Set x-axis to use integers
plt.xticks(range(1,epochs+1), rotation=45)
#add y-axis label
ax.set_ylabel('Loss', color='b', fontsize=16)
#define second y-axis that shares x-axis with current plot
ax2 = ax.twinx()
#add second line to plot
ax2.plot(accuracies, color='r')
#add second y-axis label
ax2.set_ylabel('Accuracy', color='r', fontsize=16)
fig.suptitle('Training Loss vs Accuracy')
plt.show()
x_test = test_data
x_test_recon, _, _ = vae(x_test)
test_loss = vae_loss(x_test, x_test_recon, *vae.encode(x_test))
# Calculate the accuracy of the generated samples
generated_data = vae.decode(np.random.normal(0, 1, size=(test_data.shape[0], latent_dim)))
generated_data = np.where(generated_data > 0.5, 1, 0)
print(generated_data.shape)
print(test_array.shape)
val_accuracy = np.mean(np.all(generated_data == test_data, axis=(1, 2)))
print("Validation Loss: " + str(test_loss))
print("Validation accuracy: " + str(val_accuracy))
# Load the test data
test_data = reshaped_test_data
n = 1 # number of samples
mu, sigma = 0, 1 # mean and standard deviation of the noise
shape = (n, 5, 6) # shape of the noise array
noise = np.random.normal(mu, sigma, shape)
# Encode the test data and get the mean and logvar
mean, logvar = vae.encode(noise)
# Use the mean to decode the test data
decoded_data = vae.decode(mean)
print(decoded_data)
scaled_output_array = np.zeros((test_array.shape[0], 10))
attack_average_var = 0
for i in range(decoded_data.shape[0]):
for j in range(decoded_data.shape[1]):
for k in range(decoded_data.shape[2]):
if j == 0:
scaled_output_array[i, k] = decoded_data[i,j,k]
else:
attack_average_var = attack_average_var + decoded_data[i,j,k]
if k == 5:
attack_average_var = attack_average_var/6
scaled_output_array[i, k+j] = attack_average_var
attack_average_var = 0
def decode_df(decoding, encoded):
for column, label_encoder in decoding.items():
if column != 'name':
print(column)
encoded[column] = encoded[column].map(dict(zip(label_encoder.transform(label_encoder.classes_), label_encoder.classes_)))
return encoded
# Import decoding dictionary
decoding_dict = pd.read_pickle('data\label_encoder_objects.pkl')
decoded_df = pd.DataFrame(scaled_output_array, columns=label_encoded_df.columns[1:])
print(decoded_df.head(10))
# Add a nameless column for inverse transformation
decoded_df.insert(0, '', 0)
unscaled_output_array = scaler.inverse_transform(decoded_df)
print(unscaled_output_array)
unscaled_df = pd.DataFrame(unscaled_output_array, columns=label_encoded_df.columns)
unscaled_df = unscaled_df.apply(lambda x: round(x)).astype(int)
decoded_df = decode_df(decoding_dict, unscaled_df)
print(decoded_df.head(1))
# Save the decoded predictions
decoded_df.to_csv('data.csv')