-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathDRCRN_encoder.py
136 lines (108 loc) · 5.49 KB
/
DRCRN_encoder.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
# Original CRN: copyright of (c) 2020, Ioana Bica
# DR-CRN: copyright of (c) 2022, Jiebin Chu
import logging
import numpy as np
from DR_CRN_model import DR_CRN_Model
from utils.evaluation_utils import write_results_to_file, load_trained_model, get_processed_data
def fit_DRCRN_encoder(dataset_train, dataset_val, model_name, model_dir, hyperparams_file,
b_hyperparam_opt):
_, length, num_covariates = dataset_train['current_covariates'].shape
num_treatments = dataset_train['current_treatments'].shape[-1]
num_outputs = dataset_train['outputs'].shape[-1]
num_inputs = dataset_train['current_covariates'].shape[-1] + dataset_train['current_treatments'].shape[-1]
params = {'num_treatments': num_treatments,
'num_covariates': num_covariates,
'num_outputs': num_outputs,
'max_sequence_length': length,
'num_epochs': 100}
hyperparams = dict()
num_simulations = 50
best_validation_mse = 1000000
if b_hyperparam_opt:
logging.info("Performing hyperparameter optimization")
for simulation in range(num_simulations):
logging.info("Simulation {} out of {}".format(simulation + 1, num_simulations))
hyperparams['rnn_hidden_units'] = int(np.random.choice([0.5, 1.0, 2.0, 3.0, 4.0]) * num_inputs)
hyperparams['dr_size'] = int(np.random.choice([0.5, 1.0, 2.0, 3.0, 4.0]) * num_inputs)
hyperparams['fc_hidden_units'] = int(np.random.choice([0.5, 1.0, 2.0, 3.0, 4.0]) * (hyperparams['dr_size']))
hyperparams['learning_rate'] = np.random.choice([0.01, 0.001])
hyperparams['batch_size'] = np.random.choice([64, 128, 256])
hyperparams['rnn_keep_prob'] = np.random.choice([0.7, 0.8, 0.9])
logging.info("Current hyperparams used for training \n {}".format(hyperparams))
model = DR_CRN_Model(params, hyperparams)
model.train(dataset_train, dataset_val, model_name, model_dir)
validation_mse, _ = model.evaluate_predictions(dataset_val)
if (validation_mse < best_validation_mse):
logging.info(
"Updating best validation loss | Previous best validation loss: {} | Current best validation loss: {}".format(
best_validation_mse, validation_mse))
best_validation_mse = validation_mse
best_hyperparams = hyperparams.copy()
logging.info("Best hyperparams: \n {}".format(best_hyperparams))
write_results_to_file(hyperparams_file, best_hyperparams)
else:
logging.info("Using default hyperparameters")
# best_hyperparams = {
# 'rnn_hidden_units': 12,
# 'dr_size': 12,
# # 'dr_size': 24,
# # 'fc_hidden_units': 36,
# 'fc_hidden_units': 12,
# 'learning_rate': 0.01,
# # 'batch_size': 128,
# 'batch_size': 128,
# 'rnn_keep_prob': 0.9}
best_hyperparams = { #γ=5
'rnn_hidden_units': 12,
'dr_size': 12,
# 'dr_size': 24,
# 'fc_hidden_units': 36,
'fc_hidden_units': 12,
'learning_rate': 0.01,
# 'batch_size': 128,
'batch_size': 128,
'rnn_keep_prob': 0.9}
# best_hyperparams = { # γ=3
# 'rnn_hidden_units': 18,
# 'dr_size': 24,
# # 'dr_size': 24,
# # 'fc_hidden_units': 36,
# 'fc_hidden_units': 72,
# 'learning_rate': 0.001,
# # 'batch_size': 128,
# 'batch_size': 64,
# 'rnn_keep_prob': 0.9}
# best_hyperparams = { # γ=1
# 'rnn_hidden_units': 12,
# 'dr_size': 12,
# # 'dr_size': 24,
# # 'fc_hidden_units': 36,
# 'fc_hidden_units': 12,
# 'learning_rate': 0.01,
# # 'batch_size': 128,
# 'batch_size': 128,
# 'rnn_keep_prob': 0.9}
logging.info("Best hyperparams: \n {}".format(best_hyperparams))
write_results_to_file(hyperparams_file, best_hyperparams) # todo 这个文件的保存有点问题，改一下保存方法
model = DR_CRN_Model(params, best_hyperparams)
model.train(dataset_train, dataset_val, model_name, model_dir)#这里暂时取消训练，直接载入已训练模型
def DRCRN_encoder(pickle_map, models_dir,
encoder_model_name, encoder_hyperparams_file,
b_encoder_hyperparm_tuning):
training_data = pickle_map['training_data']
validation_data = pickle_map['validation_data']
test_data = pickle_map['test_data']
scaling_data = pickle_map['scaling_data']
training_processed = get_processed_data(training_data, scaling_data)
validation_processed = get_processed_data(validation_data, scaling_data)
test_processed = get_processed_data(test_data, scaling_data)
# if b_encoder_hyperparm_tuning:
fit_DRCRN_encoder(dataset_train=training_processed, dataset_val=validation_processed,
model_name=encoder_model_name, model_dir=models_dir,
hyperparams_file=encoder_hyperparams_file, b_hyperparam_opt=b_encoder_hyperparm_tuning)
# else:
# pass
encoder = load_trained_model(validation_processed, encoder_hyperparams_file, encoder_model_name, models_dir)
mean_mse, mse = encoder.evaluate_predictions(test_processed)
rmse = (np.sqrt(np.mean(mse))) / 1150 * 100 # Max tumour volume = 1150
return rmse