-
Notifications
You must be signed in to change notification settings - Fork 9
/
Copy pathtrain.py
183 lines (128 loc) · 7.03 KB
/
train.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
import numpy as np
import pickle
import matplotlib.pyplot as plt
import os
import tensorflow as tf
from tensorflow import keras
import tensorflow.keras.backend as K
import tensorflow.keras.preprocessing as prep
from sklearn.model_selection import train_test_split
from model import AugementedConvLSTM
import configparser
import h5py
config = configparser.ConfigParser()
config.read('config.ini')
DIR = config.get('Paths', 'dir')
os.environ["CUDA_VISIBLE_DEVICES"]="1"
def get_list_from_config(strr):
req_list = [int(i) for i in strr.split(',')]
return req_list
config = configparser.ConfigParser()
config.read('config.ini')
DIR = config.get('Paths', 'dir')
DIR_monsoon_gcm = config.get('Paths', 'processed_monsoon_gcm')
DIR_monsoon_observed = config.get('Paths', 'processed_monsoon_obs')
DIR_non_monsoon_gcm = config.get('Paths', 'processed_non_monsoon_gcm')
DIR_non_monsoon_observed = config.get('Paths', 'processed_non_monsoon_obs')
min_train_year = int(config.get('DataOptions', 'min_train_year'))
max_train_year = int(config.get('DataOptions', 'max_train_year'))
min_test_year = int(config.get('DataOptions', 'min_test_year'))
max_test_year = int(config.get('DataOptions', 'max_test_year'))
projection_dimensions = config.get('DataOptions', 'projection_dimensions')
channels = config.get('DataOptions', 'channels')
convlstm_kernels = config.get('ModelParams', 'convlstm_kernels')
convlstm_kernels = get_list_from_config(convlstm_kernels)
convlstm_kernel_sizes = config.get('ModelParams', 'convlstm_kernel_sizes')
convlstm_kernel_sizes = get_list_from_config(convlstm_kernel_sizes)
sr_block_kernels = config.get('ModelParams', 'sr_block_kernels')
sr_block_kernels = get_list_from_config(sr_block_kernels)
sr_block_kernel_sizes = config.get('ModelParams', 'sr_block_kernel_sizes')
sr_block_kernel_sizes = get_list_from_config(sr_block_kernel_sizes)
sr_block_depth = int(config.get('ModelParams', 'sr_block_depth'))
learning_rate_init = float(config.get('ModelParams', 'learning_rate_init'))
learning_rate_update_factor = float(config.get('ModelParams', 'learning_rate_update_factor'))
learning_rate_update_step = float(config.get('ModelParams', 'learning_rate_update_step'))
learning_rate_patience = float(config.get('ModelParams', 'learning_rate_patience'))
minimum_learning_rate = float(config.get('ModelParams', 'minimum_learning_rate'))
training_iters = int(config.get('ModelParams', 'training_iters'))
batch_size = int(config.get('ModelParams', 'batch_size'))
timesteps = int(config.get('ModelParams', 'timesteps'))
std_dev_observed=[]
def load_dataset(model_type):
if model_type == 'monsoon':
X = np.load(DIR_monsoon_gcm + 'X_low.npy')
Y = np.load(DIR_monsoon_observed+ 'Y_low.npy')
else:
X = np.load(DIR_non_monsoon_gcm + 'X_high.npy')
Y = np.load(DIR_non_monsoon_observed + 'Y_high.npy')
return X,Y
def normalize(data):
data = data - data.mean()
data = data / data.std()
return data
def set_data(X, Y,):
X_normalized = np.zeros((channels, np.max(X.shape), projection_dimensions[0], projection_dimensions[1]))
for i in range(7):
X_normalized[i,] = normalize(X[i,])
Y_normalized = normalize(Y)
print("Mean of GCM Data: ",X[0,].mean())
print("Variance of GCM Data: ",X[0,].std(),end="\n")
print("Mean of Obseved Data: ",Y.mean())
print("Variance of Obseved Data: ",Y.std(),end="\n")
std_observed = Y.std()
X = X_normalized.transpose(1,2,3,0)
Y = Y_normalized.reshape(-1,projection_dimensions[0], projection_dimensions[1], 1)
# print("X Shape: ", X.shape)
# print("Y Shape: ", Y.shape)
std_dev_observed.append(std_observed)
return X, Y, std_observed
def data_generator(X,Y):
total_years = max_test_year - min_train_year + 1
train_years = max_train_year - min_train_year + 1
n_days = np.max(X.shape)
train_days = int((n_days/total_years)*train_years)
train_x, train_y = X[:train_days], Y[:train_days]
test_x, test_y = X[train_days:], Y[train_days:]
time_steps = timesteps
batch_size = batch_size
train_generator = prep.sequence.TimeseriesGenerator(train_x, train_y.reshape(-1, projection_dimensions[0], projection_dimensions[1], 1),length=time_steps, batch_size=batch_size)
test_generator = prep.sequence.TimeseriesGenerator(test_x, test_y.reshape(-1, projection_dimensions[0], projection_dimensions[1], 1),length=time_steps, batch_size=batch_size)
return train_generator, test_generator
def root_mean_squared_error(y_true, y_pred):
return K.sqrt(K.mean(K.square(y_pred - y_true)))
def actual_rmse_loss(y_true, y_pred):
return K.sqrt(K.mean(K.square((y_pred - y_true)*std_dev_observed[0])))
def train(clstm_model, model_type, train_generator, test_generator):
adam = tf.keras.optimizers.Adam(lr=learning_rate_init)
clstm_model.compile(optimizer=adam, loss=root_mean_squared_error, metrics=[root_mean_squared_error, actual_rmse_loss])
checkpoint = tf.keras.callbacks.ModelCheckpoint(f"norm_clstm_{model_type}_prec_weights.h5", monitor='val_loss', verbose=1, save_best_only=True, mode='min')
tensorboard = tf.keras.callbacks.TensorBoard(log_dir=f"./Graphs/norm_csltm_india_{model_type}_prec_Graph", histogram_freq=0, write_graph=True, write_images=False)
termnan = tf.keras.callbacks.TerminateOnNaN()
reduce_lr = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=learning_rate_update_factor, patience=learning_rate_patience, min_delta=learning_rate_update_step, min_lr=minimum_learning_rate, verbose=1)
callbacks_list = [checkpoint,tensorboard, reduce_lr, termnan]
history = clstm_model.fit_generator(train_generator, callbacks=callbacks_list, epochs=training_iters, validation_data=test_generator ,verbose=1)
return history
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument("--mode [train/ infer]", type=str)
parser.add_argument("--model_type [monsoon/ non-monsoon][default: non-monsoon]", type=str, default='non-monsoon')
parser.add_argument("--batch_size [default: 15]", type=int, default=15)
parser.add_argument("--use_gpu [default: false]", type=bool, default=False)
args = parser.parse_args()
return args
if __name__ == "__main__":
args = get_args()
if args.use_gpu:
os.environ["CUDA_VISIBLE_DEVICES"]="1"
print("Using GPU")
else:
os.environ["CUDA_VISIBLE_DEVICES"]="-1"
print("Using CPU")
model_type = args.model_type
X,Y = load_dataset(model_type)
X, Y, std_observed = set_data(X,Y)
train_generator, test_generator = data_generator(X, Y)
Aug_ConvLSTM_model = AugementedConvLSTM
model = Aug_ConvLSTM_model.model(convlstm_kernels, convlstm_kernel_sizes, sr_block_kernels, sr_block_kernel_sizes, sr_block_depth)
history = train(model, model_type, train_generator, test_generator)
model.save_weights(f"epoch_{training_iters}_clstm_{model_type}_prec_weights.h5")