-
Notifications
You must be signed in to change notification settings - Fork 0
/
v4wn.py
150 lines (102 loc) · 4.49 KB
/
v4wn.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
#%% setup
import tensorflow as tf
import numpy as np
import datetime
import os
import pickle
from tensorflow import keras
from tensorflow.keras.layers import Dense, Flatten, BatchNormalization, Dropout, TimeDistributed, LSTM, GRU, Activation
from tensorflow.keras.models import Model, load_model
from tensorflow.keras.utils import plot_model
from tensorflow.keras import backend as K
from tensorflow.keras.callbacks import ModelCheckpoint, TensorBoard
from tensorflow.keras.optimizers import Adam, RMSprop, Adadelta, SGD
tf.keras.backend.clear_session()
tf.compat.v1.disable_eager_execution()
tf.keras.mixed_precision.experimental.set_policy('mixed_bfloat16')
import tensorflow_addons as tfa
from load_gcs_data import IMG_WIDTH, IMG_HEIGHT, get_datasets, num_classes
from helpers import f1_m, precision_m, recall_m, get_weighted_loss, weighted_categorical_crossentropy
##CONFIG
USE_TPU = True
INPUT_SHAPE = (IMG_WIDTH, IMG_HEIGHT, 3)
NUM_OUTPUTS = num_classes
SEQUENCE_LENGTH = 16
PER_CORE_BATCH_SIZE = 16
# WEIGHT_FILE = '/root/WoW/starting_weights/weights.tf'
WEIGHT_FILE = 'gs://uswow/models/7-lstm_resnetv2_AdamW_heavy_reg/weights.tf'
SAVE_FILE = 'gs://uswow/models/7-lstm_resnetv2_AdamW_heavy_reg_10/weights.tf'
setparam={
'stateful':False,
'train_cnn': True,
'dropout':0.4,
'initial_lr': .001,
'epochs': 5 * 2
}
#print(f'Number of outputs: {num_classes}')
class_weights = pickle.load(open('weights.p', 'rb'))
class_weights = np.array(list(class_weights.values()))
#print(f'Class weights: {class_weights}')
if USE_TPU:
resolver = tf.distribute.cluster_resolver.TPUClusterResolver(tpu='node-1', project='openset-ml', zone='us-central1-a')
tf.config.experimental_connect_to_host(resolver.master())
tf.tpu.experimental.initialize_tpu_system(resolver)
strategy = tf.distribute.experimental.TPUStrategy(resolver)
print("Number of accelerators: ", strategy.num_replicas_in_sync)
BATCH_SIZE = PER_CORE_BATCH_SIZE * strategy.num_replicas_in_sync
else:
BATCH_SIZE = PER_CORE_BATCH_SIZE
train_dataset, val_dataset, train_steps, val_steps = get_datasets(batch_size=BATCH_SIZE,sequence_length=SEQUENCE_LENGTH, rand_buffer=8196//4)
def batch(params):
if USE_TPU:
with strategy.scope():
model = make_model(params)
else:
model = make_model(params)
# log_dir="gs://nnwow/logs/fit/" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
# tensorboard_callback = TensorBoard(log_dir=log_dir)
callbacks_list = []#tensorboard_callback]
out = model.fit(train_dataset,
steps_per_epoch=train_steps,
validation_data=val_dataset,
validation_steps=val_steps,
epochs=params['epochs'],
verbose=1,
#validation_freq=[params['epochs']],
callbacks=callbacks_list
)
return out, model
def make_model(params):
#CNN DEFININITION
cnn_model = keras.applications.resnet_v2.ResNet50V2(
include_top=False,
weights=None,#doesnt work
input_shape=INPUT_SHAPE,
pooling='avg')
if not params['train_cnn']:
for layer in cnn_model.layers:
layer.trainable = False
#COMBINED MODEL DEFINITION
input_layer = keras.Input(shape=((SEQUENCE_LENGTH,)+INPUT_SHAPE), batch_size=BATCH_SIZE)
cnn_layer = TimeDistributed(cnn_model, input_shape=(SEQUENCE_LENGTH,)+INPUT_SHAPE, name='cnn_timedist')(input_layer)
#ADDED DROPOUT!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
lstm_layer = LSTM(1024, return_sequences=True, stateful=params['stateful'], dropout=params['dropout'])(cnn_layer)
predictions = Dense(NUM_OUTPUTS, activation='softmax', dtype=tf.float32)(lstm_layer)
model = Model(inputs = input_layer, outputs = predictions)
print('Compiling model...')
loss = weighted_categorical_crossentropy(class_weights)
#opt = tfa.optimizers.extend_with_decoupled_weight_decay(SGD)(weight_decay=0.0001, learning_rate=['initial_lr'], nesterov=True, momentum=.9)
opt = tfa.optimizers.AdamW(0.0001, learning_rate=params['initial_lr'])
model.compile(
opt,
loss=loss,
#loss='categorical_crossentropy',
#loss=get_weighted_loss(class_weights),
metrics=['accuracy',f1_m])
print(model.summary())
print('loading weights')
model.load_weights(WEIGHT_FILE)
return model
_, model = batch(setparam)
print('saving weights')
model.save_weights(SAVE_FILE)