/
DeepMicrobes.py
executable file
·366 lines (317 loc) · 15.6 KB
/
DeepMicrobes.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
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
#!/usr/bin/env python
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from absl import flags
from absl import app as absl_app
import numpy as np
import tensorflow as tf
from models import embed_pool, embed_cnn, cnn_lstm, resnet_cnn, \
embed_lstm, embed_lstm_attention, seq2species
from models.input_pipeline import input_function_train_kmer, input_function_train_one_hot, \
input_function_predict_kmer, input_function_predict_one_hot, \
input_function_train_kmer_pad_to_fixed_len, input_function_predict_kmer_pad_to_fixed_len
from models.define_flags import universal_flags, model_specific_flags_embed_cnn, \
model_specific_flags_embed_lstm, flags_of_mode
from models.format_prediction import prob2npy, top_n_class, paired_report, \
prob2npy_paired, single_report
from utils.logs import hooks_helper
from utils.logs import logger
# import sys
# sys.path.append('models')
def config(model_name, params):
if model_name == 'embed_pool': # Embed + Pool
model = embed_pool.EmbedPool(num_classes=params['num_classes'],
vocab_size=params['vocab_size'],
embedding_dim=params['embedding_dim'],
mlp_dim=params['mlp_dim'],
kmer=params['kmer'],
max_len=params['max_len'])
elif model_name == 'embed_cnn': # Embed + CNN
model = embed_cnn.EmbedCNN(num_classes=params['num_classes'],
vocab_size=params['vocab_size'],
embedding_dim=params['embedding_dim'],
mlp_dim=params['mlp_dim'],
cnn_filter_sizes=list(map(int, params['cnn_filter_sizes'].split(","))),
cnn_num_filters=params['cnn_num_filters'],
kmer=params['kmer'],
max_len=params['max_len'])
elif model_name == 'embed_cnn_no_pool': # deprecated due to lower performance than embed_cnn
model = embed_cnn.EmbedCNNnoPool(num_classes=params['num_classes'],
vocab_size=params['vocab_size'],
embedding_dim=params['embedding_dim'],
mlp_dim=params['mlp_dim'],
cnn_filter_sizes=list(map(int, params['cnn_filter_sizes'].split(","))),
cnn_num_filters=params['cnn_num_filters'],
kmer=params['kmer'],
max_len=params['max_len'])
elif model_name == 'embed_lstm': # Embed + LSTM
model = embed_lstm.EmbedLSTM(num_classes=params['num_classes'],
vocab_size=params['vocab_size'],
embedding_dim=params['embedding_dim'],
mlp_dim=params['mlp_dim'],
lstm_dim=params['lstm_dim'],
pooling_type=params['pooling_type'],
kmer=params['kmer'],
max_len=params['max_len'])
elif model_name == 'cnn_lstm': # CNN + LSTM
model = cnn_lstm.ConvLSTM(num_classes=params['num_classes'],
max_len=params['max_len'])
elif model_name == 'cnn_2lstm': # deprecated due to lower performance than cnn_lstm
model = cnn_lstm.Conv2LSTM(num_classes=params['num_classes'],
max_len=params['max_len'])
elif model_name == 'deep_cnn': # ResNet-like CNN
model = resnet_cnn.DeepCNN(num_classes=params['num_classes'],
max_len=params['max_len'])
elif model_name == 'deep_cnn_13layer': # deprecated due to lower performance than deep_cnn
model = resnet_cnn.DeepCNN13(num_classes=params['num_classes'],
max_len=params['max_len'])
elif model_name == 'deep_cnn_9layer': # deprecated due to lower performance than deep_cnn
model = resnet_cnn.DeepCNN9(num_classes=params['num_classes'],
max_len=params['max_len'])
elif model_name == 'seq2species': # Seq2species
model = seq2species.Seq2species(num_classes=params['num_classes'],
max_len=params['max_len'])
else: # the best performing model, DeepMicrobes, Embed + LSTM + Attention
model = embed_lstm_attention.EmbedAttention(num_classes=params['num_classes'],
vocab_size=params['vocab_size'],
embedding_dim=params['embedding_dim'],
mlp_dim=params['mlp_dim'],
lstm_dim=params['lstm_dim'],
row=params['row'],
da=params['da'],
keep_prob=params['keep_prob'])
return model
def model_fn(features, labels, mode, params):
model = config(flags.FLAGS.model_name, params)
logits = model(features)
predictions = {
'classes': tf.argmax(logits, axis=1),
'probabilities': tf.nn.softmax(logits)
}
if mode == tf.estimator.ModeKeys.PREDICT:
# Return the predictions and the specification for serving a SavedModel
return tf.estimator.EstimatorSpec(
mode=mode,
predictions=predictions,
export_outputs={
'predict': tf.estimator.export.PredictOutput(predictions)
})
loss = tf.losses.sparse_softmax_cross_entropy(
logits=logits, labels=labels)
# Create a tensor named cross_entropy for logging purposes.
tf.identity(loss, name='cross_entropy')
tf.summary.scalar('cross_entropy', loss)
if mode == tf.estimator.ModeKeys.TRAIN:
global_step = tf.train.get_or_create_global_step()
learning_rate = tf.train.exponential_decay(params['lr'], global_step, 400000,
params['lr_decay'], staircase=False)
# Create a tensor named learning_rate for logging purposes
tf.identity(learning_rate, name='learning_rate')
tf.summary.scalar('learning_rate', learning_rate)
optimizer = tf.train.AdamOptimizer(learning_rate)
minimize_op = optimizer.minimize(loss, global_step)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
train_op = tf.group(minimize_op, update_ops)
else:
train_op = None
accuracy = tf.metrics.accuracy(labels, predictions['classes'])
metrics = {'accuracy': accuracy}
# Create a tensor named train_accuracy for logging purposes
tf.identity(accuracy[1], name='train_accuracy')
tf.summary.scalar('train_accuracy', accuracy[1])
return tf.estimator.EstimatorSpec(
mode=mode,
predictions=predictions,
loss=loss,
train_op=train_op,
eval_metric_ops=metrics)
def train(flags_obj, model_function, dataset_name):
run_config = tf.estimator.RunConfig(save_checkpoints_steps=100000, keep_checkpoint_max=1000)
classifier = tf.estimator.Estimator(
model_fn=model_function, model_dir=flags_obj.model_dir, config=run_config,
params={
'num_classes': flags_obj.num_classes,
'vocab_size': flags_obj.vocab_size,
'embedding_dim': flags_obj.embedding_dim,
'mlp_dim': flags_obj.mlp_dim,
'kmer': flags_obj.kmer,
'max_len': flags_obj.max_len,
'lr': flags_obj.lr,
'lr_decay': flags_obj.lr_decay,
'cnn_num_filters': flags_obj.cnn_num_filters,
'cnn_filter_sizes': flags_obj.cnn_filter_sizes,
'lstm_dim': flags_obj.lstm_dim,
'pooling_type': flags_obj.pooling_type,
'row': flags_obj.row,
'da': flags_obj.da,
'keep_prob': flags_obj.keep_prob
})
run_params = {
'batch_size': flags_obj.batch_size,
'train_epochs': flags_obj.train_epochs,
}
benchmark_logger = logger.config_benchmark_logger(flags_obj)
benchmark_logger.log_run_info('model', dataset_name, run_params)
train_hooks = hooks_helper.get_train_hooks(
flags_obj.hooks,
batch_size=flags_obj.batch_size)
def input_fn_train():
if flags_obj.encode_method == 'kmer':
input_fn = input_function_train_kmer(
flags_obj.input_tfrec,
flags_obj.train_epochs, flags_obj.batch_size,
flags_obj.cpus
)
if flags_obj.model_name in ['embed_pool', 'embed_cnn', 'embed_lstm',
'embed_cnn_no_pool']:
input_fn = input_function_train_kmer_pad_to_fixed_len(
flags_obj.input_tfrec,
flags_obj.train_epochs, flags_obj.batch_size,
flags_obj.cpus, flags_obj.max_len, flags_obj.kmer
)
else:
input_fn = input_function_train_one_hot(
flags_obj.input_tfrec,
flags_obj.train_epochs, flags_obj.batch_size,
flags_obj.cpus, flags_obj.max_len
)
return input_fn
classifier.train(input_fn=input_fn_train, hooks=train_hooks)
def evaluate(flags_obj, model_function):
classifier = tf.estimator.Estimator(
model_fn=model_function, model_dir=flags_obj.model_dir,
params={
'num_classes': flags_obj.num_classes,
'vocab_size': flags_obj.vocab_size,
'embedding_dim': flags_obj.embedding_dim,
'mlp_dim': flags_obj.mlp_dim,
'kmer': flags_obj.kmer,
'max_len': flags_obj.max_len,
'lr': flags_obj.lr,
'lr_decay': flags_obj.lr_decay,
'cnn_num_filters': flags_obj.cnn_num_filters,
'cnn_filter_sizes': flags_obj.cnn_filter_sizes,
'lstm_dim': flags_obj.lstm_dim,
'pooling_type': flags_obj.pooling_type,
'row': flags_obj.row,
'da': flags_obj.da,
'keep_prob': flags_obj.keep_prob
})
def input_fn_eval():
if flags_obj.encode_method == 'kmer':
input_fn = input_function_train_kmer(
flags_obj.input_tfrec,
1, flags_obj.batch_size,
flags_obj.cpus
)
if flags_obj.model_name in ['embed_pool', 'embed_cnn', 'embed_lstm',
'embed_cnn_no_pool']:
input_fn = input_function_train_kmer_pad_to_fixed_len(
flags_obj.input_tfrec,
1, flags_obj.batch_size,
flags_obj.cpus, flags_obj.max_len, flags_obj.kmer
)
else:
input_fn = input_function_train_one_hot(
flags_obj.input_tfrec,
1, flags_obj.batch_size,
flags_obj.cpus, flags_obj.max_len
)
return input_fn
classifier.evaluate(input_fn=input_fn_eval)
def predict(flags_obj, model_function):
classifier = tf.estimator.Estimator(
model_fn=model_function, model_dir=flags_obj.model_dir,
params={
'num_classes': flags_obj.num_classes,
'vocab_size': flags_obj.vocab_size,
'embedding_dim': flags_obj.embedding_dim,
'mlp_dim': flags_obj.mlp_dim,
'kmer': flags_obj.kmer,
'max_len': flags_obj.max_len,
'lr': flags_obj.lr,
'lr_decay': flags_obj.lr_decay,
'cnn_num_filters': flags_obj.cnn_num_filters,
'cnn_filter_sizes': flags_obj.cnn_filter_sizes,
'lstm_dim': flags_obj.lstm_dim,
'pooling_type': flags_obj.pooling_type,
'row': flags_obj.row,
'da': flags_obj.da,
'keep_prob': flags_obj.keep_prob
})
def input_fn_predict():
if flags_obj.encode_method == 'kmer':
input_fn = input_function_predict_kmer(
flags_obj.input_tfrec,
flags_obj.batch_size,
flags_obj.cpus
)
if flags_obj.model_name in ['embed_pool', 'embed_cnn', 'embed_lstm',
'embed_cnn_no_pool']:
input_fn = input_function_predict_kmer_pad_to_fixed_len(
flags_obj.input_tfrec,
flags_obj.batch_size,
flags_obj.cpus,
flags_obj.max_len,
flags_obj.kmer
)
else:
input_fn = input_function_predict_one_hot(
flags_obj.input_tfrec,
flags_obj.batch_size,
flags_obj.cpus,
flags_obj.max_len
)
return input_fn
return classifier.predict(input_fn=input_fn_predict, yield_single_examples=False)
def main(_):
if flags.FLAGS.running_mode == 'eval':
evaluate(flags.FLAGS, model_fn)
elif flags.FLAGS.running_mode == 'predict_prob':
predict_out = predict(flags.FLAGS, model_fn)
prob_matrix = prob2npy(
predict_out,
flags.FLAGS.num_classes,
flags.FLAGS.strands_average)
np.save(flags.FLAGS.pred_out, prob_matrix)
elif flags.FLAGS.running_mode == 'predict_top_n':
predict_out = predict(flags.FLAGS, model_fn)
top_n_indexes, top_n_probs = top_n_class(
predict_out,
flags.FLAGS.num_classes,
flags.FLAGS.top_n_class,
flags.FLAGS.strands_average)
np.savetxt(flags.FLAGS.pred_out+'.category.txt', top_n_indexes, fmt='%d', delimiter='\t')
np.savetxt(flags.FLAGS.pred_out+'.prob.txt', top_n_probs, fmt='%.2f', delimiter='\t')
elif flags.FLAGS.running_mode == 'predict_single_class':
predict_out = predict(flags.FLAGS, model_fn)
classes, probs = single_report(predict_out,
flags.FLAGS.num_classes,
flags.FLAGS.label_file,
flags.FLAGS.translate,
flags.FLAGS.strands_average)
np.savetxt(flags.FLAGS.pred_out + '.category_single.txt', classes, fmt='%d', delimiter='\t')
np.savetxt(flags.FLAGS.pred_out + '.prob_single.txt', probs, fmt='%.2f', delimiter='\t')
elif flags.FLAGS.running_mode == 'predict_paired_class':
predict_out = predict(flags.FLAGS, model_fn)
classes, probs = paired_report(predict_out,
flags.FLAGS.num_classes,
flags.FLAGS.label_file,
flags.FLAGS.translate)
np.savetxt(flags.FLAGS.pred_out + '.category_paired.txt', classes, fmt='%d', delimiter='\t')
np.savetxt(flags.FLAGS.pred_out + '.prob_paired.txt', probs, fmt='%.2f', delimiter='\t')
elif flags.FLAGS.running_mode == 'predict_paired_prob':
predict_out = predict(flags.FLAGS, model_fn)
prob_matrix = prob2npy_paired(predict_out,
flags.FLAGS.num_classes)
np.save(flags.FLAGS.pred_out, prob_matrix)
else:
train(flags.FLAGS, model_fn, 'dataset_name')
if __name__ == "__main__":
tf.logging.set_verbosity(tf.logging.INFO)
universal_flags()
model_specific_flags_embed_cnn()
model_specific_flags_embed_lstm()
flags_of_mode()
absl_app.run(main)