-
Notifications
You must be signed in to change notification settings - Fork 0
/
stolera_scheduler.py
56 lines (46 loc) · 1.64 KB
/
stolera_scheduler.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
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import abc
import math
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.optimizers.schedules import LearningRateSchedule
from tensorflow.python import ops
from tensorflow.python.ops import math_ops
from tensorflow.python.util.tf_export import keras_export
import sys
@keras_export("keras.optimizers.schedules.Stolera")
class Stolera(LearningRateSchedule):
def __init__(
self,
initial_learning_rate,
sigma,
seed,
name=None):
super(Stolera, self).__init__()
self.initial_learning_rate = initial_learning_rate
self.sigma = sigma
self.seed = seed
self.name = name
tf.random.set_seed(seed)
def __call__(self, step):
with tf.name_scope(self.name or "Stolera") as name:
dtype = tf.dtypes.float32
initial_learning_rate = tf.convert_to_tensor(self.initial_learning_rate, dtype=dtype, name="initial_learning_rate")
sigma = math_ops.cast(self.sigma, dtype)
t_step = math_ops.cast(step, dtype)
# t_step = math_ops.multiply(t_step, t_step)
t_step = math_ops.add(t_step, tf.constant(1, dtype=dtype))
Z_t = tf.random.normal([1], mean=0.0, stddev=1.0, dtype=dtype)
term_a = math_ops.divide(Z_t[0], t_step)
term_b = math_ops.multiply(sigma, term_a)
term_c = math_ops.subtract(initial_learning_rate, term_b, name=name)
return term_c
def get_config(self):
return {
"initial_learning_rate": self.initial_learning_rate,
"sigma": self.sigma,
"seed": self.seed,
"name": self.name
}