-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathClassification_pruning.py
153 lines (118 loc) · 6.92 KB
/
Classification_pruning.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
import tensorflow as tf
import numpy as np
import argparse
import csv
import random
import os
current_path = os.getcwd()
# ==================================================================================
# To use tensor cores and speed up training (uncomment only if available)
# ==================================================================================
# from tensorflow.keras import mixed_precision
# policy = mixed_precision.Policy('mixed_float16')
# mixed_precision.set_global_policy(policy)
# print('Compute dtype: %s' % policy.compute_dtype)
# print('Variable dtype: %s' % policy.variable_dtype)
# ==================================================================================
# This is used to launch all the runs with the required parameters.
# ==================================================================================
argParser = argparse.ArgumentParser()
argParser.add_argument('--N_samples', type = int, required = True)
argParser.add_argument('--K', type = int, required = True)
argParser.add_argument('--type_prune', type = str, required = True)
argParser.add_argument('--ratio', type = float, required = True)
args = argParser.parse_args()
N_samples = args.N_samples
K = args.K
type_prune = args.type_prune
ratio = args.ratio
# for sanity checking
print('Number of samples used : {}'.format(N_samples))
print('K = {}'.format(K))
print('type_prune : {}'.format(type_prune))
print('Prune_ratio = {}'.format(ratio))
# ==================================================================================
# We load directly the MFCCs into the memory, again, to speed up the process.
# ==================================================================================
X_train_features = np.load(os.path.join(current_path, 'MFCCs' , 'train_MFCCs.npy'))
train_normal_y = np.load(os.path.join(current_path , 'Dataset_as_numpy', 'train_y.npy'))
X_valid_features = np.load(os.path.join(current_path, 'MFCCs' , 'valid_MFCCs.npy'))
valid_normal_y = np.load(os.path.join(current_path , 'Dataset_as_numpy', 'valid_y.npy'))
X_test_features = np.load(os.path.join(current_path, 'MFCCs' , 'test_MFCCs.npy'))
test_normal_y = np.load(os.path.join(current_path , 'Dataset_as_numpy', 'test_y.npy'))
# ==================================================================================
# select parts of the trainig set only, pruned using Wav2vec + kmeans + number final
# ==================================================================================
# If the pruning is not "normal" ie strictly random pruning we select according to our pruning results
if type_prune != 'normal':
y_pruned = np.load(os.path.join(current_path, 'y_pruned', 'y_{}_pruning_FULL_k{}_{}.npy'.format(type_prune , K , ratio)))
print('There are ', len(y_pruned), 'samples in the pruned dataset')
# prune the dataset according to the labels previously selected
X_train_features = X_train_features[y_pruned,:]
train_normal_y = train_normal_y[y_pruned]
# randomly select a certain number of the dataset, random pruning always occurs, even if we pruned the dataset before
labels_random_prune = random.sample(range(1, len(train_normal_y)), N_samples)
X_train_features = X_train_features[labels_random_prune,:]
train_normal_y = train_normal_y[labels_random_prune]
print('There are ', len(train_normal_y), 'samples in the final train set')
# ==================================================================================
# Tensorflow classifiers, uncomment the desired model
# ==================================================================================
Batch_size = 512
epoch = 170
initial_learn_rate = 5e-3
Callback_epoch_number = 15
dropout_rate = 0.1
N_classes = (np.max(train_normal_y)+1)
print('there are {} classes in the dataset'.format(N_classes))
# # LeNet original model
# model = tf.keras.Sequential()
# model.add(tf.keras.layers.Conv2D(filters=6, kernel_size=(2, 2), activation='relu', input_shape=(99,32,1)))
# model.add(tf.keras.layers.AveragePooling2D())
# model.add(tf.keras.layers.Conv2D(filters=16, kernel_size=(5, 3), activation='relu'))
# model.add(tf.keras.layers.AveragePooling2D())
# model.add(tf.keras.layers.Flatten())
# model.add(tf.keras.layers.Dense(units=120, activation='relu'))
# model.add(tf.keras.layers.Dropout(dropout_rate))
# model.add(tf.keras.layers.Dense(units=84, activation='relu'))
# model.add(tf.keras.layers.Dropout(dropout_rate))
# model.add(tf.keras.layers.Dense(units=N_classes, activation = 'softmax'))
# model.summary()
# # Smallest model
# model = tf.keras.Sequential()
# model.add(tf.keras.layers.Conv2D(filters=3, kernel_size=(2, 2), activation='relu', input_shape=(99,32,1)))
# model.add(tf.keras.layers.AveragePooling2D())
# model.add(tf.keras.layers.Conv2D(filters=8, kernel_size=(5, 3), activation='relu'))
# model.add(tf.keras.layers.AveragePooling2D())
# model.add(tf.keras.layers.Conv2D(filters=16, kernel_size=(5, 3), activation='relu'))
# model.add(tf.keras.layers.AveragePooling2D())
# model.add(tf.keras.layers.Flatten())
# model.add(tf.keras.layers.Dense(units=84, activation='relu'))
# model.add(tf.keras.layers.Dropout(dropout_rate))
# model.add(tf.keras.layers.Dense(units=N_classes, activation = 'softmax'))
# model.summary()
# Tiny model
model = tf.keras.Sequential()
model.add(tf.keras.layers.Conv2D(filters=4, kernel_size=(2, 2), activation='relu', input_shape=(99,32,1)))
model.add(tf.keras.layers.AveragePooling2D())
model.add(tf.keras.layers.Conv2D(filters=8, kernel_size=(2, 2), activation='relu'))
model.add(tf.keras.layers.AveragePooling2D())
model.add(tf.keras.layers.Conv2D(filters=8, kernel_size=(4, 2), activation='relu'))
model.add(tf.keras.layers.AveragePooling2D())
model.add(tf.keras.layers.Conv2D(filters=16, kernel_size=(4, 2), activation='relu'))
model.add(tf.keras.layers.AveragePooling2D())
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dropout(dropout_rate))
model.add(tf.keras.layers.Dense(units=N_classes, activation = 'softmax'))
model.summary()
model.compile(optimizer=tf.optimizers.Adam(lr=initial_learn_rate),loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), metrics=['accuracy'])
lr_reduce = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=3, min_delta=0.0001)
Stopping_condition = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=Callback_epoch_number, restore_best_weights=True, min_delta=0.0001)
history = model.fit(X_train_features, train_normal_y , batch_size=Batch_size, shuffle=True, callbacks = [lr_reduce,Stopping_condition], epochs=epoch, validation_data=(X_valid_features, valid_normal_y), verbose = 1)
test_loss, test_acc = model.evaluate(X_test_features, test_normal_y, verbose=2)
print("The final accuracy is : ")
print(test_acc)
with open('results/history_tiny_prune_{}_K{}_{}_{}.csv'.format(N_samples, K, type_prune, ratio), 'a') as f_object:
writer_object = csv.writer(f_object)
# writer_object.writerow(['test_loss', 'test_acc', 'N_epochs'])
writer_object.writerow([test_loss, test_acc, len(history.history['loss'])])