forked from tensorflow/tensorflow
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathftrl.py
More file actions
134 lines (119 loc) · 5.52 KB
/
ftrl.py
File metadata and controls
134 lines (119 loc) · 5.52 KB
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
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Ftrl-proximal for TensorFlow."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import ops
from tensorflow.python.ops import math_ops
from tensorflow.python.training import optimizer
from tensorflow.python.training import training_ops
class FtrlOptimizer(optimizer.Optimizer):
"""Optimizer that implements the FTRL algorithm.
See this [paper](
https://www.eecs.tufts.edu/~dsculley/papers/ad-click-prediction.pdf).
@@__init__
"""
def __init__(self, learning_rate,
learning_rate_power=-0.5,
initial_accumulator_value=0.1,
l1_regularization_strength=0.0,
l2_regularization_strength=0.0,
use_locking=False, name="Ftrl"):
"""Construct a new FTRL optimizer.
Args:
learning_rate: A float value or a constant float `Tensor`.
learning_rate_power: A float value, must be less or equal to zero.
initial_accumulator_value: The starting value for accumulators.
Only positive values are allowed.
l1_regularization_strength: A float value, must be greater than or
equal to zero.
l2_regularization_strength: A float value, must be greater than or
equal to zero.
use_locking: If `True` use locks for update operations.
name: Optional name prefix for the operations created when applying
gradients. Defaults to "Ftrl".
Raises:
ValueError: If one of the arguments is invalid.
"""
super(FtrlOptimizer, self).__init__(use_locking, name)
if initial_accumulator_value <= 0.0:
raise ValueError("initial_accumulator_value %f needs to be positive" %
initial_accumulator_value)
if learning_rate_power > 0.0:
raise ValueError("learning_rate_power %f needs to be negative or zero" %
learning_rate_power)
if l1_regularization_strength < 0.0:
raise ValueError(
"l1_regularization_strength %f needs to be positive or zero" %
l1_regularization_strength)
if l2_regularization_strength < 0.0:
raise ValueError(
"l2_regularization_strength %f needs to be positive or zero" %
l2_regularization_strength)
self._learning_rate = learning_rate
self._learning_rate_power = learning_rate_power
self._initial_accumulator_value = initial_accumulator_value
self._l1_regularization_strength = l1_regularization_strength
self._l2_regularization_strength = l2_regularization_strength
self._learning_rate_tensor = None
self._learning_rate_power_tensor = None
self._l1_regularization_strength_tensor = None
self._l2_regularization_strength_tensor = None
def _create_slots(self, var_list):
# Create the "accum" and "linear" slots.
for v in var_list:
val = constant_op.constant(self._initial_accumulator_value,
dtype=v.dtype, shape=v.get_shape())
self._get_or_make_slot(v, val, "accum", self._name)
self._zeros_slot(v, "linear", self._name)
def _prepare(self):
self._learning_rate_tensor = ops.convert_to_tensor(
self._learning_rate,
name="learning_rate")
self._l1_regularization_strength_tensor = ops.convert_to_tensor(
self._l1_regularization_strength,
name="l1_regularization_strength")
self._l2_regularization_strength_tensor = ops.convert_to_tensor(
self._l2_regularization_strength,
name="l2_regularization_strength")
self._learning_rate_power_tensor = ops.convert_to_tensor(
self._learning_rate_power,
name="learning_rate_power")
def _apply_dense(self, grad, var):
accum = self.get_slot(var, "accum")
linear = self.get_slot(var, "linear")
return training_ops.apply_ftrl(
var, accum, linear, grad,
math_ops.cast(self._learning_rate_tensor, var.dtype.base_dtype),
math_ops.cast(self._l1_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._l2_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._learning_rate_power_tensor, var.dtype.base_dtype),
use_locking=self._use_locking)
def _apply_sparse(self, grad, var):
accum = self.get_slot(var, "accum")
linear = self.get_slot(var, "linear")
return training_ops.sparse_apply_ftrl(
var, accum, linear, grad.values, grad.indices,
math_ops.cast(self._learning_rate_tensor, var.dtype.base_dtype),
math_ops.cast(self._l1_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._l2_regularization_strength_tensor,
var.dtype.base_dtype),
math_ops.cast(self._learning_rate_power_tensor, var.dtype.base_dtype),
use_locking=self._use_locking)