-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtf1.py
153 lines (110 loc) · 4.04 KB
/
tf1.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
import argparse
import math
import numpy as np
import pandas as pd
# Make NumPy printouts easier to read.
from cg_scheduler import NewtonLRScheduler
from clr_callback import CyclicLR, SGDRScheduler
np.set_printoptions(precision=3, suppress=True)
from tensorflow import keras
from tensorflow.keras import layers
import tensorflow.compat.v1 as tf
tf.disable_v2_behavior() # Enable tf v1 behavior as in v2 a lot have changed
# import dl_utils as utils
import newton_cg as es
print(tf.__version__)
parser = argparse.ArgumentParser(description='Keras regression case',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--optimizer', type=int, default=0, help='0 for ncg, 1 for adam, 2 for sgd')
parser.add_argument('--scheduler', type=int, default=0, help='1 for exponential, 2 for time-based, 3 for step-based, '
'0 for constant')
parser.add_argument('--lr', type=float, default=0.01, help='')
parser.add_argument('--epoch', type=int, default=50, help='')
args = parser.parse_args()
data = pd.read_csv("birth_rate.csv")
data.head()
def sigmoid(x):
z = np.exp(-x)
sig = 1 / (1 + z)
return sig
# Split data/labels
data_X = np.array(data['Birth rate'])
# data_X = np.array(range(10)) * 0.1 - 0.5
# np.array(data['Birth rate'])
# data_Y = sigmoid(2 * data_X + 0.2)
# np.array(data['Life expectancy'])
data_Y = np.array(data['Life expectancy'])
# print(data_X)
# print(data_Y)
# print(len(data))
# X = tf.placeholder(tf.float32, name='X')
# Y = tf.placeholder(tf.float32, name='Y')
# w = tf.get_variable('weight', initializer=tf.constant(0.1))
# b = tf.get_variable('bias', initializer=tf.constant(0.1))
# Y_hat = tf.add(tf.multiply(X,w), b)
# loss = (Y-Y_hat)*(Y-Y_hat) # tf.keras.losses.mean_squared_error(Y,Y_hat)
# loss = huber_loss(Y,Y_hat)
# model = keras.Model(inputs=[X], outputs=Y, name=f'model{i}')
# model = tf.keras.Sequential([
# X,
# layers.Dense(units=1)
# ])
model = tf.keras.Sequential()
model.add(tf.keras.layers.Dense(1, activation=tf.nn.sigmoid))
model.build((None, 1))
print(model.get_weights())
if args.optimizer < 1:
model.compile(optimizer=es.EHNewtonOptimizer(0.001), loss='mse')
elif args.optimizer < 2:
model.compile(optimizer=tf.train.AdamOptimizer(1), loss='mse')
else:
model.compile(optimizer=tf.train.GradientDescentOptimizer(1), loss='mse')
sess = tf.Session()
sess.run(tf.global_variables_initializer())
model.summary()
model.evaluate(data_X, data_Y)
# for manually changing lr
# for i in range(10):
# print(model.get_weights())
# # adapt lr
# model.fit(data_X, data_Y, epochs=1, callbacks=[lr_exp])
callbacks_array = []
if args.scheduler == 1:
callbacks_array.append(es.exp_lr)
elif args.scheduler == 2:
callbacks_array.append(es.time_lr)
elif args.scheduler == 3:
callbacks_array.append(es.step_lr)
else:
callbacks_array.append(es.const_lr)
print("Picked constant lr")
def scheduler(epoch, lr):
if epoch < 10:
return lr
else:
return lr - 0.01
nlr = NewtonLRScheduler(start_lr=1.0, drop=0.5, epochs_drop=10.0,
verbose=1, constant_epochs=5, schedule=scheduler)
clr_triangular = CyclicLR(mode='triangular2', base_lr=0.01, max_lr=1.0, step_size=26.)
lrc = tf.keras.callbacks.LearningRateScheduler(schedule=scheduler, verbose=1)
warm_clr = SGDRScheduler(min_lr=0.001, max_lr=1.0, steps_per_epoch=26, lr_decay=0.8, cycle_length=5, mult_factor=1.5)
# 190 training samples, 19 batch size => 10 Iteration per epoch, 30 epochs => 300 total iterations
# history = model.fit(data_X, data_Y, callbacks=[warm_clr], epochs=args.epoch, batch_size=19, verbose=1)
# h = clr_triangular.history
# lr = h['lr']
# iters = h['iterations']
#
# dict = {'lr': lr, 'iterations': iters}
#
df = pd.DataFrame(warm_clr.history)
# saving the dataframe
df.to_csv('warm.csv', index=False)
# losses = history.history['loss']
#
# print(lr)
# print(iters)
# print(losses)
# print(len(losses))
print(model.trainable_variables)
print("Weigths:")
print(model.get_weights())