-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathmain.py
More file actions
155 lines (128 loc) · 8.33 KB
/
main.py
File metadata and controls
155 lines (128 loc) · 8.33 KB
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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Thu Nov 15 12:19:17 2018
@author: hananhindy
"""
import itertools
from dataset_processor import DatasetHandler
from siamese_net import SiameseNet
from args_handler import Arguments
import matplotlib.pyplot as plt
import numpy as np
import time
import os
if __name__ == "__main__":
args = Arguments()
args.parse()
dataset_handler = DatasetHandler(args.path, args.dataset_name, args.verbose)
all_classes = list(dataset_handler.get_classes())
if args.verbose:
print('\nClasses are:\n{}'.format(all_classes))
if args.train_with_all:
training_categories = all_classes
testing_categories = all_classes
else:
all_conbinations = list(itertools.combinations(all_classes, args.number_of_training_categories))
training_categories = all_conbinations[args.comb_index]
testing_categories = list(set(all_classes) - set(training_categories))
dataset_handler.encode_split(training_categories, testing_categories, args.verbose)
if args.verbose:
print('!Starting!')
with open(args.output_file_name, "a") as file_writer:
file_writer.write("Dataset, {}\n".format(args.dataset_name))
file_writer.write("Network ID, {}\n".format(args.network_id))
file_writer.write("Training Batch:Testing Batch, {}:{}\n".format(args.batch_size, args.testing_batch_size))
file_writer.write("No of iterations, {}\n".format(args.niterations))
file_writer.write("acc_not_in_training, acc_added_labels\n")
file_writer.write(", ".join(training_categories) + "\n")
for run in range(args.nruns):
print(args.network_id)
if args.verbose:
print("Run #{}".format(run))
wrapper = SiameseNet((dataset_handler.number_of_features,), args.network_id, args.dataset_name, args.verbose)
if args.network_path != '':
wrapper.load_saved_model(args.network_path)
if args.train_with_all or args.test_vs_all:
dataset_handler.evaluate_classisfication(
'pairs/{}_{}_{}_Classification_Pairs.csv'.format(args.dataset_name, args.batch_size,args.comb_index),
wrapper.siamese_net,
args.batch_size,
len(all_classes),
all_classes, args.output_file_name)
else:
index_of_zero_day_category = all_classes.index([item for item in all_classes if item not in training_categories][0])
dataset_handler.evaluate_zero_day_new(
'pairs/{}_{}_{}_Classification_Pairs.csv'.format(args.dataset_name, args.batch_size, args.comb_index),
wrapper.siamese_net,
args.batch_size,
len(all_classes),
index_of_zero_day_category,
training_categories,
args.output_file_name)
else:
(inputs1, targets1) = dataset_handler.load_batch(args.batch_size, 'pairs/{}_{}_{}_Training_Pairs.csv'.format(args.dataset_name, args.batch_size, args.comb_index))
if args.train_with_all:
(inputs_val, targets_val) = dataset_handler.load_batch(args.batch_size, 'pairs/{}_{}_{}_Validation_Pairs.csv'.format(args.dataset_name, args.batch_size,args.comb_index))
loss = np.zeros(args.niterations)
validation_loss = np.zeros(args.niterations)
tr_acc = np.zeros(args.niterations)
val_acc = np.zeros(args.niterations)
for i in range(1, args.niterations + 1):
if args.use_sub_batches:
if args.train_with_all:
hist = wrapper.siamese_net.fit(inputs1,targets1, batch_size = args.batch_size//1000, validation_data= (inputs_val,targets_val)).history
val_acc[i-1] = round(hist['val_accuracy'][0]*100, 2)
else:
hist = wrapper.siamese_net.fit(inputs1,targets1, batch_size = args.batch_size//1000).history
loss[i-1] = hist['loss'][0]
tr_acc[i-1] = round(hist['accuracy'][0]*100, 2)
else:
loss[i-1] = wrapper.siamese_net.train_on_batch(inputs1,targets1)
if args.train_with_all:
if args.use_sub_batches:
validation_loss[i-1] = hist['val_loss'][0]
else:
validation_loss[i-1] = wrapper.siamese_net.test_on_batch(inputs_val, targets_val)
print('{} -> Loss = {} validation loss = {}'.format(i, loss[i-1], validation_loss[i-1]))
if i >= args.evaluate_every and i%args.evaluate_every == 0:
if args.train_with_all or args.test_vs_all:
plt.clf()
training_plot, = plt.plot(loss[0:i])
val_plot, = plt.plot(validation_loss[0:i])
plt.xlabel('Iteration no')
plt.ylabel('Loss')
plt.legend([training_plot, val_plot], ['Training Loss', 'Validation Loss'])
plt.savefig('loss_Run_{}.png'.format(args.network_id))
dataset_handler.evaluate_classisfication(
'pairs/{}_{}_{}_Classification_Pairs.csv'.format(args.dataset_name, args.batch_size,args.comb_index),
wrapper.siamese_net,
args.batch_size,
len(all_classes),
all_classes, args.output_file_name)
else:
index_of_zero_day_category = all_classes.index([item for item in all_classes if item not in training_categories][0])
accuracy_zero_day, conf_mat = dataset_handler.evaluate_zero_day_new('pairs/{}_{}_{}_Classification_Pairs.csv'.format(args.dataset_name, args.batch_size, args.comb_index),
wrapper.siamese_net,
args.batch_size,
len(all_classes),
index_of_zero_day_category,
training_categories,
args.output_file_name)
wrapper.siamese_net.save(os.path.join('/home/hananhindy/dump_networks/', '{}_{}_{}_{}'.format(args.dataset_name, args.comb_index, args.network_id, time.strftime("%Y%m%d-%H%M%S"))))
if args.train_with_all or args.test_vs_all:
plt.clf()
training_plot, = plt.plot(loss[loss>0])
val_plot, = plt.plot(validation_loss[validation_loss>0])
plt.xlabel('Iteration no')
plt.ylabel('Loss')
plt.legend([training_plot, val_plot], ['Training Loss', 'Validation Loss'])
plt.savefig('loss_Run_{}_{}_{}.png'.format(args.dataset_name, args.network_id, time.strftime("%Y%m%d-%H%M%S")))
plt.ylim(bottom=0, top=3) # adjust the top leaving bottom unchanged
plt.savefig('limit_loss_Run_{}_{}.png'.format(args.network_id, time.strftime("%Y%m%d-%H%M%S")))
plt.clf()
training_plot, = plt.plot(tr_acc[tr_acc>0])
val_plot, = plt.plot(val_acc[val_acc>0])
plt.xlabel('Iteration no')
plt.ylabel('Accuracy')
plt.legend([training_plot, val_plot], ['Training Acc', 'Validation Acc'])