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
|
# Copyright 2016 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.
# ==============================================================================
"""A collection of functions to be used as evaluation metrics."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.contrib import losses
from tensorflow.contrib.learn.python.learn.estimators import prediction_key
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import metrics
from tensorflow.python.ops import nn
INFERENCE_PROB_NAME = prediction_key.PredictionKey.PROBABILITIES
INFERENCE_PRED_NAME = prediction_key.PredictionKey.CLASSES
FEATURE_IMPORTANCE_NAME = 'global_feature_importance'
def _top_k_generator(k):
def _top_k(probabilities, targets):
targets = math_ops.to_int32(targets)
if targets.get_shape().ndims > 1:
targets = array_ops.squeeze(targets, axis=[1])
return metrics.mean(nn.in_top_k(probabilities, targets, k))
return _top_k
def _accuracy(predictions, targets, weights=None):
return metrics.accuracy(
labels=targets, predictions=predictions, weights=weights)
def _r2(probabilities, targets, weights=None):
targets = math_ops.to_float(targets)
y_mean = math_ops.reduce_mean(targets, 0)
squares_total = math_ops.reduce_sum(math_ops.square(targets - y_mean), 0)
squares_residuals = math_ops.reduce_sum(
math_ops.square(targets - probabilities), 0)
score = 1 - math_ops.reduce_sum(squares_residuals / squares_total)
return metrics.mean(score, weights=weights)
def _squeeze_and_onehot(targets, depth):
targets = array_ops.squeeze(targets, axis=[1])
return array_ops.one_hot(math_ops.to_int32(targets), depth)
def _sigmoid_entropy(probabilities, targets, weights=None):
return metrics.mean(
losses.sigmoid_cross_entropy(probabilities,
_squeeze_and_onehot(
targets,
array_ops.shape(probabilities)[1])),
weights=weights)
def _softmax_entropy(probabilities, targets, weights=None):
return metrics.mean(
losses.sparse_softmax_cross_entropy(probabilities,
math_ops.to_int32(targets)),
weights=weights)
def _predictions(predictions, unused_targets, **unused_kwargs):
return predictions
def _class_log_loss(probabilities, targets, weights=None):
return metrics.mean(
losses.log_loss(probabilities,
_squeeze_and_onehot(targets,
array_ops.shape(probabilities)[1])),
weights=weights)
def _precision(predictions, targets, weights=None):
return metrics.precision(
labels=targets, predictions=predictions, weights=weights)
def _precision_at_thresholds(predictions, targets, weights=None):
return metrics.precision_at_thresholds(
labels=targets,
predictions=array_ops.slice(predictions, [0, 1], [-1, 1]),
thresholds=np.arange(0, 1, 0.01, dtype=np.float32),
weights=weights)
def _recall(predictions, targets, weights=None):
return metrics.recall(
labels=targets, predictions=predictions, weights=weights)
def _recall_at_thresholds(predictions, targets, weights=None):
return metrics.recall_at_thresholds(
labels=targets,
predictions=array_ops.slice(predictions, [0, 1], [-1, 1]),
thresholds=np.arange(0, 1, 0.01, dtype=np.float32),
weights=weights)
def _auc(probs, targets, weights=None):
return metrics.auc(
labels=targets,
predictions=array_ops.slice(probs, [0, 1], [-1, 1]),
weights=weights)
_EVAL_METRICS = {
'auc': _auc,
'sigmoid_entropy': _sigmoid_entropy,
'softmax_entropy': _softmax_entropy,
'accuracy': _accuracy,
'r2': _r2,
'predictions': _predictions,
'classification_log_loss': _class_log_loss,
'precision': _precision,
'precision_at_thresholds': _precision_at_thresholds,
'recall': _recall,
'recall_at_thresholds': _recall_at_thresholds,
'top_5': _top_k_generator(5)
}
_PREDICTION_KEYS = {
'auc': INFERENCE_PROB_NAME,
'sigmoid_entropy': INFERENCE_PROB_NAME,
'softmax_entropy': INFERENCE_PROB_NAME,
'accuracy': INFERENCE_PRED_NAME,
'r2': prediction_key.PredictionKey.SCORES,
'predictions': INFERENCE_PRED_NAME,
'classification_log_loss': INFERENCE_PROB_NAME,
'precision': INFERENCE_PRED_NAME,
'precision_at_thresholds': INFERENCE_PROB_NAME,
'recall': INFERENCE_PRED_NAME,
'recall_at_thresholds': INFERENCE_PROB_NAME,
'top_5': INFERENCE_PROB_NAME
}
def get_metric(metric_name):
"""Given a metric name, return the corresponding metric function."""
return _EVAL_METRICS[metric_name]
def get_prediction_key(metric_name):
return _PREDICTION_KEYS[metric_name]
|
