-
Notifications
You must be signed in to change notification settings - Fork 0
/
TensorflowModelCreator.py
394 lines (352 loc) · 22.6 KB
/
TensorflowModelCreator.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
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
# Plots
import matplotlib.pyplot as plt
import seaborn as sns
# TensorFlow
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout
from tensorflow.keras.optimizers import SGD, RMSprop, Adagrad, Adadelta, Adam, Adamax, Nadam
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint, TensorBoard, ReduceLROnPlateau
from tensorflow.keras import regularizers
# Usuals
import pandas as pd
sns.set()
plt.rcParams['figure.figsize'] = (8, 6)
plt.rcParams['axes.grid'] = False
tf.compat.v1.enable_eager_execution()
print(tf.__version__)
print(pd.__version__)
# Create a Stratey
class Tensorflow_Model_Generator_Class():
def __init__(self, ):
# ---------------------------------------------------------------------------- #
# Tensorflow lists of hyperparam types #
# ---------------------------------------------------------------------------- #
self.initializers = [None,
tf.keras.initializers.Constant(),
tf.keras.initializers.GlorotNormal(),
tf.keras.initializers.GlorotUniform(),
tf.keras.initializers.HeNormal(),
tf.keras.initializers.HeUniform(),
tf.keras.initializers.Identity(),
tf.keras.initializers.LecunNormal(),
tf.keras.initializers.LecunUniform(),
tf.keras.initializers.Ones(),
tf.keras.initializers.Orthogonal(),
tf.keras.initializers.RandomNormal(),
tf.keras.initializers.RandomUniform(),
tf.keras.initializers.TruncatedNormal(),
tf.keras.initializers.VarianceScaling(),
tf.keras.initializers.Zeros(),
]
self.activators = [None,
tf.keras.activations.tanh,
tf.keras.activations.softmax,
tf.keras.activations.elu,
tf.keras.activations.softplus,
tf.keras.activations.softsign,
tf.keras.activations.relu,
tf.keras.activations.sigmoid,
tf.keras.activations.hard_sigmoid,
tf.keras.activations.linear,
tf.keras.activations.exponential,
tf.keras.activations.selu,
tf.keras.activations.swish,
]
self.regularizers = [None,
tf.keras.regularizers.L1(),
tf.keras.regularizers.L1L2(),
tf.keras.regularizers.L2()]
self.constraints = [None,
tf.keras.constraints.MaxNorm(),
tf.keras.constraints.MinMaxNorm(),
tf.keras.constraints.NonNeg(),
tf.keras.constraints.RadialConstraint(),
tf.keras.constraints.UnitNorm(),
]
self.optimizers = [tf.optimizers.Nadam(),
tf.optimizers.Ftrl(),
tf.optimizers.Adam(),
tf.optimizers.SGD(),
tf.optimizers.RMSprop(),
tf.optimizers.Adagrad(),
tf.optimizers.Adadelta(),
tf.optimizers.Adamax(),
]
self.dropouts = [None, 0.1, 0.25, 0.5]
# ---------------------------------------------------------------------------- #
# Initial Tensorflow model gene space #
# ---------------------------------------------------------------------------- #
self.prelim_gene_space = {
# Universal hyperparams
# "epochs": [1000],
# "early_stopping_patience": [0.1],
"batch_size": [16, 32, 64, 128, 256, 512],
"n_layers": [1, 2, 4, 8],
"n_neurons": [5, 3, 2, 1.5, 1, 0.75, 0.5, 0.25], # multipliers
"dropout": self.dropouts,
# "losses": ["mae"],
# "metrics": ["mae"],
"learning_rate": [0.1, 0.01, 0.001, 0.0001, 0.00001, 0.000001],
"optimizers": self.optimizers,
# Dense: First Layer
"first_layer_activation": self.activators,
"first_layer_use_bias": [True, False],
"first_layer_kernel_initializer": self.initializers,
"first_layer_bias_initializer": self.initializers,
"first_layer_kernel_regularizer": self.regularizers,
"first_layer_bias_regularizer": self.regularizers,
"first_layer_activity_regularizer": self.regularizers,
"first_layer_kernel_constraint": self.constraints,
"first_layer_bias_constraint": self.constraints,
# Dense: Subsequent Layer
"sub_layer_activation": self.activators,
"sub_layer_use_bias": [True, False],
"sub_layer_kernel_initializer": self.initializers,
"sub_layer_bias_initializer": self.initializers,
"sub_layer_kernel_regularizer": self.regularizers,
"sub_layer_bias_regularizer": self.regularizers,
"sub_layer_activity_regularizer": self.regularizers,
"sub_layer_kernel_constraint": self.constraints,
"sub_layer_bias_constraint": self.constraints,
# Dense: Final Layer
"final_layer_activation": self.activators,
"final_layer_use_bias": [True, False],
"final_layer_kernel_initializer": self.initializers,
"final_layer_bias_initializer": self.initializers,
"final_layer_kernel_regularizer": self.regularizers,
"final_layer_bias_regularizer": self.regularizers,
"final_layer_activity_regularizer": self.regularizers,
"final_layer_kernel_constraint": self.constraints,
"final_layer_bias_constraint": self.constraints,
# # LSTM: First Layer (additionals)
# "first_layer_recurrent_activation":self.activators,
# "first_layer_recurrent_initializer":self.initializers,
# "first_layer_unit_forget_bias":[True, False],
# "first_layer_recurrent_regularizer":self.regularizers,
# "first_layer_recurrent_constraint":self.constraints,
# "first_layer_recurrent_dropout":dropouts,
# "first_layer_return_sequences":[True, False],
# "first_layer_return_state":[True, False],
# "first_layer_go_backwards":[True, False],
# "first_layer_stateful":[True, False],
# "first_layer_time_major":[True, False],
# "first_layer_unroll":[True, False],
# # GRU: First Layer
# "first_layer_reset_after":[True, False],
# # LSTM: Subsequent Layer (additionals)
# "sub_layer_recurrent_activation":self.activators,
# "sub_layer_recurrent_initializer":self.initializers,
# "sub_layer_unit_forget_bias":[True, False],
# "sub_layer_recurrent_regularizer":self.regularizers,
# "sub_layer_recurrent_constraint":self.constraints,
# "sub_layer_recurrent_dropout":dropouts,
# "sub_layer_return_sequences":[True, False],
# "sub_layer_return_state":[True, False],
# "sub_layer_go_backwards":[True, False],
# "sub_layer_stateful":[True, False],
# "sub_layer_time_major":[True, False],
# "sub_layer_unroll":[True, False],
# # GRU: Subsequent Layer
# "sub_layer_reset_after":[True, False],
}
# Convert lists to final gene space
self.gene_space = []
for key in self.prelim_gene_space.keys():
self.gene_space.append({"low": int(0), "high": int(len(self.prelim_gene_space[key])), "step": int(1)})
self.list_of_keys = [key for key in self.prelim_gene_space.keys()]
def generate_dense_model(self, current_hyperparams, num_of_training_features, model_output_size):
"""
:param current_hyperparams:
:return:
"""
# Make sure current_hyperparams is all int's re: pygad changes them to floats
current_hyperparams = [int(a) for a in current_hyperparams]
# Reconstruct the variables we need, makes it easier to ref variables below
batch_size = self.prelim_gene_space['batch_size'][current_hyperparams[self.list_of_keys.index('batch_size')]]
n_layers = self.prelim_gene_space['n_layers'][current_hyperparams[self.list_of_keys.index('n_layers')]]
n_neurons = self.prelim_gene_space['n_neurons'][current_hyperparams[self.list_of_keys.index('n_neurons')]]
dropout = self.prelim_gene_space['dropout'][current_hyperparams[self.list_of_keys.index('dropout')]]
optimizer = self.prelim_gene_space['optimizers'][current_hyperparams[self.list_of_keys.index('optimizers')]]
learning_rate = self.prelim_gene_space['learning_rate'][current_hyperparams[self.list_of_keys.index('learning_rate')]]
# Dense
first_layer_activation = self.prelim_gene_space['first_layer_activation'][
current_hyperparams[self.list_of_keys.index('first_layer_activation')]]
first_layer_use_bias = self.prelim_gene_space['first_layer_use_bias'][
current_hyperparams[self.list_of_keys.index('first_layer_use_bias')]]
first_layer_kernel_initializer = self.prelim_gene_space['first_layer_kernel_initializer'][
current_hyperparams[self.list_of_keys.index('first_layer_kernel_initializer')]]
first_layer_bias_initializer = self.prelim_gene_space['first_layer_bias_initializer'][
current_hyperparams[self.list_of_keys.index('first_layer_bias_initializer')]]
first_layer_kernel_regularizer = self.prelim_gene_space['first_layer_kernel_regularizer'][
current_hyperparams[self.list_of_keys.index('first_layer_kernel_regularizer')]]
first_layer_bias_regularizer = self.prelim_gene_space['first_layer_bias_regularizer'][
current_hyperparams[self.list_of_keys.index('first_layer_bias_regularizer')]]
first_layer_activity_regularizer = self.prelim_gene_space['first_layer_activity_regularizer'][
current_hyperparams[self.list_of_keys.index('first_layer_activity_regularizer')]]
first_layer_kernel_constraint = self.prelim_gene_space['first_layer_kernel_constraint'][
current_hyperparams[self.list_of_keys.index('first_layer_kernel_constraint')]]
first_layer_bias_constraint = self.prelim_gene_space['first_layer_bias_constraint'][
current_hyperparams[self.list_of_keys.index('first_layer_bias_constraint')]]
sub_layer_activation = self.prelim_gene_space['sub_layer_activation'][
current_hyperparams[self.list_of_keys.index('sub_layer_activation')]]
sub_layer_use_bias = self.prelim_gene_space['sub_layer_use_bias'][
current_hyperparams[self.list_of_keys.index('sub_layer_use_bias')]]
sub_layer_kernel_initializer = self.prelim_gene_space['sub_layer_kernel_initializer'][
current_hyperparams[self.list_of_keys.index('sub_layer_kernel_initializer')]]
sub_layer_bias_initializer = self.prelim_gene_space['sub_layer_bias_initializer'][
current_hyperparams[self.list_of_keys.index('sub_layer_bias_initializer')]]
sub_layer_kernel_regularizer = self.prelim_gene_space['sub_layer_kernel_regularizer'][
current_hyperparams[self.list_of_keys.index('sub_layer_kernel_regularizer')]]
sub_layer_bias_regularizer = self.prelim_gene_space['sub_layer_bias_regularizer'][
current_hyperparams[self.list_of_keys.index('sub_layer_bias_regularizer')]]
sub_layer_activity_regularizer = self.prelim_gene_space['sub_layer_activity_regularizer'][
current_hyperparams[self.list_of_keys.index('sub_layer_activity_regularizer')]]
sub_layer_kernel_constraint = self.prelim_gene_space['sub_layer_kernel_constraint'][
current_hyperparams[self.list_of_keys.index('sub_layer_kernel_constraint')]]
sub_layer_bias_constraint = self.prelim_gene_space['sub_layer_bias_constraint'][
current_hyperparams[self.list_of_keys.index('sub_layer_bias_constraint')]]
final_layer_activation = self.prelim_gene_space['final_layer_activation'][
current_hyperparams[self.list_of_keys.index('final_layer_activation')]]
final_layer_use_bias = self.prelim_gene_space['final_layer_use_bias'][
current_hyperparams[self.list_of_keys.index('final_layer_use_bias')]]
final_layer_kernel_initializer = self.prelim_gene_space['final_layer_kernel_initializer'][
current_hyperparams[self.list_of_keys.index('final_layer_kernel_initializer')]]
final_layer_bias_initializer = self.prelim_gene_space['final_layer_bias_initializer'][
current_hyperparams[self.list_of_keys.index('final_layer_bias_initializer')]]
final_layer_kernel_regularizer = self.prelim_gene_space['final_layer_kernel_regularizer'][
current_hyperparams[self.list_of_keys.index('final_layer_kernel_regularizer')]]
final_layer_bias_regularizer = self.prelim_gene_space['final_layer_bias_regularizer'][
current_hyperparams[self.list_of_keys.index('final_layer_bias_regularizer')]]
final_layer_activity_regularizer = self.prelim_gene_space['final_layer_activity_regularizer'][
current_hyperparams[self.list_of_keys.index('final_layer_activity_regularizer')]]
final_layer_kernel_constraint = self.prelim_gene_space['final_layer_kernel_constraint'][
current_hyperparams[self.list_of_keys.index('final_layer_kernel_constraint')]]
final_layer_bias_constraint = self.prelim_gene_space['final_layer_bias_constraint'][
current_hyperparams[self.list_of_keys.index('final_layer_bias_constraint')]]
# # Inspect the current_hyperparams' params
# print("batch_size:", batch_size,
# "n_layers:", n_layers,
# "n_neurons:", n_neurons,
# "dropout:", dropout,
# "optimizer:", optimizer,
# "learning_rate:", learning_rate, "\n",
# "first_layer_activation:", first_layer_activation,
# "first_layer_use_bias:", first_layer_use_bias,
# "first_layer_kernel_initializer:", first_layer_kernel_initializer,
# "first_layer_bias_regularizer:", first_layer_bias_regularizer,
# "first_layer_activity_regularizer:", first_layer_activity_regularizer,
# "first_layer_kernel_constraint:", first_layer_kernel_constraint,
# "first_layer_bias_constraint:", first_layer_bias_constraint,
# "sub_layer_activation:", sub_layer_activation,
# "sub_layer_use_bias:", sub_layer_use_bias,
# "sub_layer_kernel_initializer:", sub_layer_kernel_initializer,
# "sub_layer_bias_initializer:", sub_layer_bias_initializer,
# "sub_layer_kernel_regularizer:", sub_layer_kernel_regularizer,
# "sub_layer_bias_regularizer:", sub_layer_bias_regularizer,
# "sub_layer_activity_regularizer:", sub_layer_activity_regularizer,
# "sub_layer_kernel_constraint:", sub_layer_kernel_constraint,
# "sub_layer_bias_constraint:", sub_layer_bias_constraint,
# "final_layer_activation:", final_layer_activation,
# "final_layer_use_bias:", final_layer_use_bias,
# "final_layer_kernel_initializer:", final_layer_kernel_initializer,
# "final_layer_bias_initializer:", final_layer_bias_initializer,
# "final_layer_kernel_regularizer:", final_layer_kernel_regularizer,
# "final_layer_bias_regularizer:", final_layer_bias_regularizer,
# "final_layer_activity_regularizer:", final_layer_activity_regularizer,
# "final_layer_kernel_constraint:", final_layer_kernel_constraint,
# "final_layer_bias_constraint:", final_layer_bias_constraint)
# # Make new dataset re: batch size
# window = WindowGenerator(input_width=past_lookback_window_timesteps,
# label_width=past_lookback_window_timesteps,
# shift=future_prediction_timesteps,
# label_columns=['Close'],
# batch_size=batch_size)
# print('Input shape:', window.example[0].shape)
# Early stopping callback
early_stopping = tf.keras.callbacks.EarlyStopping(monitor='val_loss',
# This should be val_loss in future
patience=5,
mode='min')
# The n_layers passed to us is simply describing a shape. it does not
# know the num_cols we are dealing with, it is simply values of 0.5, 1, and 2,
# which need to be multiplied by the num_cols
# UPDATE. CHANGE FROM NUM_COLS TO TOTAL DATAPOINTS!!!!! NOT SURE WHAT I PUT THIS?
neurons_per_layer = []
for _ in range(int(n_layers)):
neurons_per_layer.append(max(int(num_of_training_features * n_neurons), 2))
# Begin model building
model = tf.keras.Sequential()
# Add first layer
model.add(tf.keras.layers.Dense(neurons_per_layer[0],
activation=first_layer_activation,
# input_dim=train_df.shape[1], # Maybe not needed?
use_bias=first_layer_use_bias,
kernel_initializer=first_layer_kernel_initializer,
bias_initializer=first_layer_bias_initializer,
kernel_regularizer=first_layer_kernel_regularizer,
bias_regularizer=first_layer_bias_regularizer,
activity_regularizer=first_layer_activity_regularizer,
kernel_constraint=first_layer_kernel_constraint,
bias_constraint=first_layer_bias_constraint,
# **kwargs
)) # , kernel_regularizer=regularizers.l2(0.01)
# First Dropout
if dropout != None:
model.add(tf.keras.layers.Dropout(dropout))
# Second layer
if len(neurons_per_layer) <= 2:
model.add(tf.keras.layers.Dense(neurons_per_layer[-1],
activation=sub_layer_activation,
# input_dim=train_df.shape[1], # Maybe not needed?
use_bias=sub_layer_use_bias,
kernel_initializer=sub_layer_kernel_initializer,
bias_initializer=sub_layer_bias_initializer,
kernel_regularizer=sub_layer_kernel_regularizer,
bias_regularizer=sub_layer_bias_regularizer,
activity_regularizer=sub_layer_activity_regularizer,
kernel_constraint=sub_layer_kernel_constraint,
bias_constraint=sub_layer_bias_constraint,
# **kwargs
)) # , kernel_regularizer=regularizers.l2(0.01)
# Second Dropout(s)
if dropout != None:
model.add(tf.keras.layers.Dropout(dropout))
# Subsequent layers
else:
for layer_size in neurons_per_layer[1:-1]:
model.add(tf.keras.layers.Dense(layer_size,
activation=sub_layer_activation,
# input_dim=train_df.shape[1], # Maybe not needed?
use_bias=sub_layer_use_bias,
kernel_initializer=sub_layer_kernel_initializer,
bias_initializer=sub_layer_bias_initializer,
kernel_regularizer=sub_layer_kernel_regularizer,
bias_regularizer=sub_layer_bias_regularizer,
activity_regularizer=sub_layer_activity_regularizer,
kernel_constraint=sub_layer_kernel_constraint,
bias_constraint=sub_layer_bias_constraint,
# **kwargs
)) # , kernel_regularizer=regularizers.l2(0.01)
# Final Dropout
if dropout != None:
model.add(tf.keras.layers.Dropout(dropout))
# Final layer
model.add(tf.keras.layers.Dense(model_output_size, activation=final_layer_activation,
# input_dim=len(train_df.columns.to_list()), # Maybe not needed?
use_bias=final_layer_use_bias,
kernel_initializer=final_layer_kernel_initializer,
bias_initializer=final_layer_bias_initializer,
kernel_regularizer=final_layer_kernel_regularizer,
bias_regularizer=final_layer_bias_regularizer,
activity_regularizer=final_layer_activity_regularizer,
kernel_constraint=final_layer_kernel_constraint,
bias_constraint=final_layer_bias_constraint))
# Compile
optimizer.learning_rate = learning_rate
if learning_rate != None:
model.compile(loss=tf.losses.MeanAbsoluteError(), optimizer=optimizer,
metrics=tf.metrics.MeanAbsoluteError())
else: # binary_crossentropy, categorical_crossentropy
model.compile(loss=tf.losses.MeanAbsoluteError(), optimizer=optimizer, metrics=tf.metrics.MeanAbsoluteError())
return model