-
Notifications
You must be signed in to change notification settings - Fork 3
/
PredictEncoding_randomly_generated.py
514 lines (408 loc) · 22.8 KB
/
PredictEncoding_randomly_generated.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
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
import numpy as np
import pickle
from SmallParser import SmallDatasetParser, get_doubling_xy, get_x, read_data
from Processing import correct_slopes, decorrect_slopes, normalize_features, normalize_with_norms, \
denormalize_with_norms, normalize_total, split_train_test
import AutoEncoders
import GeneLists
import CompoundLists
import matplotlib.pyplot as plt
from keras.layers import Input, Dense, concatenate, Conv2D, SpatialDropout2D, MaxPooling2D, Flatten, Dropout, Reshape
from keras.models import Model, load_model
from keras.optimizers import SGD, Adam
from keras import regularizers
from keras.callbacks import EarlyStopping
import scipy.io as sio
from keras.metrics import mse, mae, mape
from keras import backend as K
import tensorflow as tf
from sklearn.ensemble import RandomForestRegressor
from sklearn.linear_model import MultiTaskElasticNet
from sklearn.neighbors import KNeighborsRegressor
def cnn_model(x_train, x_valid, y_train, y_valid, norms_X, norms_Y):
"""A Convolutional Neural Network consisting of three coupled convolutional and maxpooling layers and one Dense layer, with data preprocessing like in the Autoencoders."""
#determine the shape of the source and target domain data.
trg_timepoints = 3
if x_train.shape[1]!=y_train.shape[1]:
trg_timepoints = 4
padded = False
genes = x_train.shape[1] // 3
MIN_GENES = 50
#if the number of genes in the gene set are less than 50 we intoduce zero padding to insure input data has a dimension of at least 50x3
if genes < MIN_GENES:
padded = True
# Assume y_train has the same amount of genes...
# shape of train: (num_samples, genes x timepoints)
x_pad = np.zeros((x_train.shape[0], (MIN_GENES - genes) * 3))
x_v_pad = np.zeros((x_valid.shape[0], (MIN_GENES - genes) * 3))
y_pad = np.zeros((y_train.shape[0], (MIN_GENES - genes) * trg_timepoints))
y_v_pad = np.zeros((y_valid.shape[0], (MIN_GENES - genes) * trg_timepoints))
x_train = np.append(x_train, x_pad, axis=1)
x_valid = np.append(x_valid, x_v_pad, axis=1)
y_train = np.append(y_train, y_pad, axis=1)
y_valid = np.append(y_valid, y_v_pad, axis=1)
res_x_train = x_train.reshape((x_train.shape[0], x_train.shape[1] // 3, 3, 1))
res_x_valid = x_valid.reshape((x_valid.shape[0], x_valid.shape[1] // 3, 3, 1))
print(y_train.shape)
print(y_train.shape)
res_y_train = y_train.reshape((y_train.shape[0], y_train.shape[1] // trg_timepoints, trg_timepoints, 1))
res_y_valid = y_valid.reshape((y_valid.shape[0], y_valid.shape[1] // trg_timepoints, trg_timepoints, 1))
print(res_y_valid.shape)
print(res_y_train.shape)
input_layer = Input(shape=(x_train.shape[1] // 3, 3, 1))
# define convolutional+maxpooling layer 1
x = Conv2D(16, kernel_size=(10, 1), activation='relu', data_format='channels_last')(input_layer)
x = SpatialDropout2D(0.1)(x)
x = MaxPooling2D(pool_size=(2, 1), data_format='channels_last')(x)
# define convolutional+maxpooling layer 2
x = Conv2D(8, kernel_size=(10, 1), activation='relu', data_format='channels_last')(x)
x = SpatialDropout2D(0.1)(x)
x = MaxPooling2D(pool_size=(2, 1), data_format='channels_last')(x)
#define convolutional+maxpooling layer 3
x = Conv2D(4, kernel_size=(2, 1), activation='relu', data_format='channels_last')(x)
x = SpatialDropout2D(0.1)(x)
x = MaxPooling2D(pool_size=(2, 1), data_format='channels_last')(x)
#flatten output into 1 dimensional vector.
x = Flatten()(x)
x = Dense(20, activation='relu', activity_regularizer=regularizers.l1(10e-6))(x)
#up-sampling
x = Dense(30, activation='relu')(x)
x = Dropout(0.1)(x)
x = Dense(y_train.shape[1], activation='sigmoid')(x)
x = Reshape((y_train.shape[1] // trg_timepoints, trg_timepoints, 1))(x)
model = Model(input_layer, x)
#define coset function mean absolute error (mae) and optimiser Adam.
adam = Adam()
model.compile(loss='mae', optimizer=adam, metrics=['mae', 'mse'])
#model fit
model.fit(res_x_train, res_y_train, epochs=10000, batch_size=128, shuffle=True,
validation_data=(res_x_valid, res_y_valid),
verbose=0,
callbacks=[EarlyStopping(monitor='val_loss', min_delta=0.00001, patience=40, verbose=0, mode='auto')])
recon_train = model.predict(res_x_train)
recon_valid = model.predict(res_x_valid)
recon_train_res = recon_train.reshape((res_y_train.shape[0], -1))
recon_valid_res = recon_valid.reshape((res_y_valid.shape[0], -1))
# Drop the padding for calculating true error
if padded:
drop = MIN_GENES - genes
x_train = np.delete(x_train, np.s_[-drop * 3 - 1:-1], 1)
y_train = np.delete(y_train, np.s_[-drop * trg_timepoints - 1:-1], 1)
x_valid = np.delete(x_valid, np.s_[-drop * 3 - 1:-1], 1)
y_valid = np.delete(y_valid, np.s_[-drop * trg_timepoints - 1:-1], 1)
recon_train_res = np.delete(recon_train_res, np.s_[-drop * trg_timepoints - 1:-1], 1)
recon_valid_res = np.delete(recon_valid_res, np.s_[-drop * trg_timepoints - 1:-1], 1)
X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid = _correct_data(x_train, x_valid, y_train, y_valid,
recon_train_res, recon_valid_res,
norms_X,
norms_Y)
train_mae = np.average(np.absolute(Y_train - recon_train))
train_mse = np.average(np.square(Y_train - recon_train))
val_mae = np.average(np.absolute(Y_valid - recon_valid))
val_mse = np.average(np.square(Y_valid - recon_valid))
return X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid, np.array(
[train_mae, train_mse, val_mae, val_mse])
def merge_autoencoders(inputs, encoder, decoder):
model_input = Input(shape=(inputs,))
encoder_output = encoder(model_input)
decoder_output = decoder(encoder_output)
network = Model(model_input, decoder_output)
adam = Adam(lr=0.0005, decay=0.00001)
network.compile(loss='mae', optimizer=adam, metrics=['mae', 'mse'])
return network
def mod_autoencoder_model(inputs): # this defines the modified autoencoder model
# define three encoder layers
input_layer = Input(shape=(inputs,))
encoded = Dense(70, activation='relu')(input_layer)
encoded = Dense(70, activation='relu', activity_regularizer=regularizers.l1(10e-7))(encoded)
encoded = Dense(60, activation='relu', activity_regularizer=regularizers.l1(10e-9))(encoded)
encoder = Model(input_layer, encoded)
# define three decoder layers
encoded_input = Input(shape=(60,))
decoded = Dense(70, activation='relu', activity_regularizer=regularizers.l1(10e-7))(encoded_input)
decoded = Dense(70, activation='relu')(decoded)
decoded = Dense(inputs, activation="sigmoid")(decoded)
decoder = Model(encoded_input, decoded)
# define full model network; input, endcoder layers, decoder layers, and output.
model_input = Input(shape=(inputs,))
encoder_output = encoder(model_input)
decoder_output = decoder(encoder_output)
network = Model(model_input, decoder_output)
#define optimiser - Adam and loss function mean absolute error (mae)
adam = Adam(lr=0.0005, decay=0.00001)
network.compile(loss='mae', optimizer=adam, metrics=['mae', 'mse'])
return network, encoder, decoder
def _correct_data(X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid, norms_X, norms_Y):
# Denormalize all data
X_train = denormalize_with_norms(X_train, norms_X)
X_valid = denormalize_with_norms(X_valid, norms_X)
Y_train = denormalize_with_norms(Y_train, norms_Y)
Y_valid = denormalize_with_norms(Y_valid, norms_Y)
recon_train = denormalize_with_norms(recon_train, norms_Y)
recon_valid = denormalize_with_norms(recon_valid, norms_Y)
# Decorrect for slopes, otherwise normalization won't be correct
X_train = decorrect_slopes(X_train, x_vivo)
X_valid = decorrect_slopes(X_valid, x_vivo)
Y_train = decorrect_slopes(Y_train, y_vivo)
Y_valid = decorrect_slopes(Y_valid, y_vivo)
recon_train = decorrect_slopes(recon_train, y_vivo)
recon_valid = decorrect_slopes(recon_valid, y_vivo)
# Normalize again, but using min/max over entire dataset instead of per feature.
# This is to have a consistent error measurement across different experiments.
combined = np.append(X_train, X_valid, axis=0)
combined, norms = normalize_total(combined)
X_train = combined[:X_train.shape[0]]
X_valid = combined[X_train.shape[0]:]
combined = np.append(Y_train, Y_valid, axis=0)
combined, norms = normalize_total(combined)
Y_train = combined[:Y_train.shape[0]]
Y_valid = combined[Y_train.shape[0]:]
recon_train = normalize_with_norms(recon_train, norms)
recon_valid = normalize_with_norms(recon_valid, norms)
# Correct for slopes again
X_train = correct_slopes(X_train, x_vivo)
X_valid = correct_slopes(X_valid, x_vivo)
Y_train = correct_slopes(Y_train, y_vivo)
Y_valid = correct_slopes(Y_valid, y_vivo)
recon_train = correct_slopes(recon_train, y_vivo)
recon_valid = correct_slopes(recon_valid, y_vivo)
return X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid
def train_network(network, X_train, X_valid, Y_train, Y_valid):
batch_size = 32
# Fitting the data
network.fit(X_train, Y_train, epochs=5000, batch_size=batch_size, shuffle=False, validation_data=(X_valid, Y_valid),
verbose=0,
callbacks=[EarlyStopping(monitor='val_loss', min_delta=0.00001, patience=100, verbose=0, mode='auto')])
# Given X, create Y such that the instances now have doubling (r1->r1, r1->r2, r2->r1, r2->r2)
def X_to_doubled(X):
d1_r1 = np.repeat(X[0::4], 2, axis=0) # split data into first and second replicate
d1_r2 = np.repeat(X[2::4], 2, axis=0)
# Change into r1, r2, r1, r2
interwoven = np.ravel(np.column_stack((d1_r1, d1_r2))).reshape(X.shape)
return interwoven
# Given Y, create X such that the instances now have doubling (r1->r1, r1->r2, r2->r1, r2->r2)
def Y_to_doubled(Y):
X = np.zeros(Y.shape)
X[0::4] = Y[0::4]
X[1::4] = Y[2::4]
X[2::4] = Y[1::4]
X[3::4] = Y[3::4]
return X
def train_naive_encoder(X_train, X_valid, Y_train, Y_valid, norms_X, norms_Y): # Dan: for untrained AE
inputs = X_train.shape[1]
#trains naive encoder model
# define the three layer of the encoder
input_layer = Input(shape=(inputs,))
encoded = Dense(256, activation='relu')(input_layer)
encoded = Dense(160, activation='relu', activity_regularizer=regularizers.l1(10e-7))(encoded)
encoded = Dense(32, activation='relu', activity_regularizer=regularizers.l1(10e-9))(encoded)
encoder = Model(input_layer, encoded)
# deinfe the three layers of of the decoder
encoded_input = Input(shape=(32,))
decoded = Dense(96, activation='relu', activity_regularizer=regularizers.l1(10e-7))(encoded_input)
decoded = Dense(256, activation='relu')(decoded)
decoded = Dense(Y_train.shape[1], activation="sigmoid")(decoded) # Dan: I changed this line!
decoder = Model(encoded_input, decoded)
# define total model - input, three encoder layers, three decoder layers, output.
model_input = Input(shape=(inputs,))
encoder_output = encoder(model_input)
decoder_output = decoder(encoder_output)
network = Model(model_input, decoder_output)
#define adaptive learning rate, number of epochs
lr_start = 0.01
lr_end = 0.001
epochs = 5000
decay = (lr_start - lr_end)/epochs
#define loss function (mean absolute error (mae)) and optimisation fucntion Adam.
adam = Adam(lr=lr_start, decay=decay)
network.compile(loss='mae', optimizer=adam, metrics=['mae', 'mse'])
batch_size = 128
# Fitting the data
network.fit(X_train, Y_train, epochs=epochs, batch_size=batch_size, shuffle=False, validation_data=(X_valid, Y_valid),
verbose=0,
callbacks=[EarlyStopping(monitor='val_loss', min_delta=0.00001, patience=100, verbose=0, mode='auto')])
recon_train = network.predict(X_train)
recon_valid = network.predict(X_valid)
X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid = _correct_data(X_train, X_valid, Y_train, Y_valid,
recon_train, recon_valid, norms_X,
norms_Y)
train_mae = np.average(np.absolute(Y_train - recon_train))
train_mse = np.average(np.square(Y_train - recon_train))
val_mae = np.average(np.absolute(Y_valid - recon_valid))
val_mse = np.average(np.square(Y_valid - recon_valid))
print(
'LOOV next result - train_mae: {:.4f} train_mse: {:.4f} val_mae: {:.4f} val_mse: {:.4f}'.format(
train_mae,
train_mse,
val_mae,
val_mse))
return X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid, np.array(
[train_mae, train_mse, val_mae, val_mse])
def train_mod_ autoencoders(X_train, X_valid, Y_train, Y_valid, norms_X, norms_Y):
# Create and train autoencoder for first domain
print("Training autoencoder on first domain")
d1_network, d1_encoder, d1_decoder = autoencoder_model(X_train.shape[1])
train_network(d1_network, X_train, X_valid, X_train, X_valid)
# Create and train autoencoder for second domain
print("Training autoencoder on second domain")
d2_network, d2_encoder, d2_decoder = autoencoder_model(Y_train.shape[1])
train_network(d2_network, Y_train, Y_valid, Y_train, Y_valid)
# Combine autoencoders and re-train network
print("Training combined network")
network = merge_autoencoders(X_train.shape[1], d1_encoder, d2_decoder)
train_network(network, X_train, X_valid, Y_train, Y_valid)
recon_train = network.predict(X_train)
recon_valid = network.predict(X_valid)
X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid = _correct_data(X_train, X_valid, Y_train, Y_valid,
recon_train, recon_valid, norms_X,
norms_Y)
train_mae = np.average(np.absolute(Y_train - recon_train))
train_mse = np.average(np.square(Y_train - recon_train))
val_mae = np.average(np.absolute(Y_valid - recon_valid))
val_mse = np.average(np.square(Y_valid - recon_valid))
return X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid, np.array(
[train_mae, train_mse, val_mae, val_mse])
# Train benchmarking machine learning methods (random regression forest, elastic net or kNN) and output predictions for the test data
# note results of elastic net is not included in the manuscript - while elastic net out preforms kNN it's predictions are not as good as the random regression forest.
def train_base_model(X_train, X_valid, Y_train, Y_valid, norms_X, norms_Y, model='rf'):
if model == 'rf': #random regression forest
predictor = RandomForestRegressor(max_features=0.3, n_estimators=200, n_jobs=3)
elif model == 'elastic': #elastic net (note the results for the elastic net in not included in the manuscript- it did not prefer as well as the random regression forest)
predictor = MultiTaskElasticNet(alpha=0.003, l1_ratio=0.7)
elif model == 'knn': #k-nearest neighbours.
predictor = KNeighborsRegressor(2, weights='distance')
else:
raise ValueError('{} is not a valid model!'.format(model))
predictor.fit(X_train, Y_train)
recon_train = predictor.predict(X_train)
recon_valid = predictor.predict(X_valid)
X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid = _correct_data(X_train, X_valid, Y_train, Y_valid,
recon_train, recon_valid, norms_X,
norms_Y)
train_mae = np.average(np.absolute(Y_train - recon_train))
train_mse = np.average(np.square(Y_train - recon_train))
val_mae = np.average(np.absolute(Y_valid - recon_valid))
val_mse = np.average(np.square(Y_valid - recon_valid))
return X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid, np.array(
[train_mae, train_mse, val_mae, val_mse])
def leave_one_out_evaluation(X, Y, compounds, model='autoencoder', x_vivo_arg=False, y_vivo_arg=False):
print("Performing leave-one-out evaluation with '{}' model".format(model))
#train the specified model using leave-one-compound-out cross validation
global x_vivo, y_vivo
x_vivo = x_vivo_arg
y_vivo = y_vivo_arg
total_errors = []
total_X_train = None
total_Y_train = None
total_X_valid = None
total_Y_valid = None
total_recon_train = None
total_recon_valid = None
total_train_compounds = None
total_valid_compounds = None
unique_compounds = np.unique(compounds)
counter = 0
for i, exclude_compound in enumerate(unique_compounds):
counter = counter + 1
print("Excluding compound", exclude_compound, "[", counter, "/", len(unique_compounds), "]")
X_train, X_valid, Y_train, Y_valid, norms_X, norms_Y, train_compounds, \
valid_compounds = split_train_test(X, Y, compounds, x_vivo, y_vivo, exclude_compound=exclude_compound)
if model == 'mod_autoencoder':
X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid, errors = train_mod_autoencoders(X_train, X_valid,
Y_train, Y_valid,
norms_X, norms_Y)
elif model == 'cnn':
X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid, errors = cnn_model(
X_train, X_valid,
Y_train, Y_valid,
norms_X, norms_Y)
elif model == 'naive_encoder':
X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid, errors = train_naive_encoder(
X_train, X_valid,
Y_train, Y_valid,
norms_X, norms_Y)
else:
X_train, X_valid, Y_train, Y_valid, recon_train, recon_valid, errors = train_base_model(X_train, X_valid,
Y_train, Y_valid,
norms_X, norms_Y,
model)
if i == 0:
total_X_train = X_train
total_Y_train = Y_train
total_X_valid = X_valid
total_Y_valid = Y_valid
total_recon_train = recon_train
total_recon_valid = recon_valid
total_train_compounds = train_compounds
total_valid_compounds = valid_compounds
else:
total_X_train = np.append(total_X_train, X_train, axis=0)
total_Y_train = np.append(total_Y_train, Y_train, axis=0)
total_X_valid = np.append(total_X_valid, X_valid, axis=0)
total_Y_valid = np.append(total_Y_valid, Y_valid, axis=0)
total_recon_train = np.append(total_recon_train, recon_train, axis=0)
total_recon_valid = np.append(total_recon_valid, recon_valid, axis=0)
total_train_compounds = np.append(total_train_compounds, train_compounds, axis=0)
total_valid_compounds = np.append(total_valid_compounds, valid_compounds, axis=0)
total_errors.append(errors)
total_errors = np.array(total_errors)
avg_errors = np.mean(total_errors, axis=0)
print("Average Errors:")
print("Training mae:{}, mse:{}".format(avg_errors[0], avg_errors[1]))
print("Validation mae:{}, mse:{}".format(avg_errors[2], avg_errors[3]))
store_mae = avg_errors[2] # Dan: I added this
print("Compounds sorted by validation MAE (compound, MAE):")
mae = total_errors[:, 2]
for i in np.argsort(mae):
print(unique_compounds[i], mae[i])
data = {
'X_train': total_X_train, 'Y_train': total_Y_train, 'recon_train': total_recon_train,
'X_valid': total_X_valid, 'Y_valid': total_Y_valid, 'recon_valid': total_recon_valid,
'train_compounds': total_train_compounds, 'valid_compounds': total_valid_compounds
}
return data, store_mae
def main():
tf.logging.set_verbosity(tf.logging.ERROR)
compound_list = CompoundLists.GENERAL_45
#define data domains note: X_type or input domain is always rat_vitro.
x_type = "rat_vitro"
y_type = "human_vitro" #change target domain here as desired "rat_vivo" or "human_vitro"
#load and parse data (extract time series of gene expression for the specfied compounds and genes in the specified domains)
global x_vivo, y_vivo
x_vivo = x_type == "rat_vivo"
y_vivo = y_type == "rat_vivo"
all_maes = []; #for storing results
for k in range(20,21): # number of genes in set (20, 35, 50, 65, 80)
for i in range(26,29): #identifier of gene sets
"""
Change location of input files below if desired:
- change 'Random' to 'Orth' for orthologs
- note that for size 20 there is (by definition) no 'Nested' folder!
- change domain name if desired
"""
#import gene set
file1 = "Data/RatInVitro/%d"%(k) + "/Nested/Random%d"%(k) + "/data_X%d"%(k) + "_%d"%(i) + "_nest.p"
file2 = "Data/HumanInVitro/%d"%(k) + "/Nested/Random%d"%(k) + "/data_%d"%(k) + "_%d"%(i) + "_human_nest.p"
print(file1)
print(file2)
X, _, gene_list_x, _ = pickle.load(open(file1, "rb"))
Y, data_compounds, gene_list_y,_ = pickle.load(open(file2, "rb"))
""" Dan: select desired method under 'model'
rf -> random regression forest
mod_autoencoder -> modified autoender
naive_encoder -> naive encoder model
cnn -> convolutional neural network
"""
#leave-one-compound-out cross validation for user specifed compounds and genes for the given model
data, mae = leave_one_out_evaluation(X, Y, data_compounds, model='rf',x_vivo_arg=x_vivo, y_vivo_arg=y_vivo)
all_maes.append(mae)
print("Here are all MAEs again:\n", all_maes)
""" Dan: this block can be uncommented to store the prediction values as a matlab workspace
filename = "predictions.mat"
print("filename", out_file_name)
data['genes'] = gene_list_y
data['genes_input'] = gene_list_x
sio.savemat(out_file_name, data)
"""
if __name__ == '__main__':
main()