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
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
|
// Copyright 2022 The Chromium Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "components/segmentation_platform/internal/post_processor/post_processor.h"
#include "base/time/time.h"
#include "components/segmentation_platform/internal/metadata/metadata_utils.h"
#include "base/check_op.h"
#include "base/logging.h"
#include "base/notreached.h"
#include "components/segmentation_platform/public/result.h"
namespace segmentation_platform {
namespace {
// UMA constants for various types of classifier outputs.
constexpr int kInvalidResult = -2;
constexpr int kNoWinningLabel = -1;
constexpr int kUnderflowBinIndex = -1;
bool IsValidResult(const proto::PredictionResult& prediction_result) {
if (metadata_utils::ValidateOutputConfig(prediction_result.output_config()) !=
metadata_utils::ValidationResult::kValidationSuccess) {
return false;
}
int output_length = 1;
const auto& predictor = prediction_result.output_config().predictor();
if (predictor.has_multi_class_classifier()) {
output_length = predictor.multi_class_classifier().class_labels_size();
int threshold_count = predictor.multi_class_classifier().class_thresholds_size();
if (threshold_count > 0 && output_length != threshold_count) {
return false;
}
}
if (predictor.has_generic_predictor()) {
output_length = predictor.generic_predictor().output_labels_size();
}
return prediction_result.result_size() > 0 &&
prediction_result.has_output_config() &&
prediction_result.result_size() == output_length;
}
bool IsScoreBelowMultiClassThreshold(
const proto::Predictor::MultiClassClassifier& multi_class_classifier,
const float class_score,
const int class_index) {
if (multi_class_classifier.class_thresholds_size() > 0) {
return class_score < multi_class_classifier.class_thresholds(class_index);
} else {
return class_score < multi_class_classifier.threshold();
}
}
} // namespace
// static
bool PostProcessor::IsClassificationResult(
const proto::PredictionResult& prediction_result) {
const proto::Predictor& predictor =
prediction_result.output_config().predictor();
switch (predictor.PredictorType_case()) {
case proto::Predictor::kBinaryClassifier:
case proto::Predictor::kMultiClassClassifier:
case proto::Predictor::kBinnedClassifier:
return true;
default:
return false;
}
}
std::vector<std::string> PostProcessor::GetClassifierResults(
const proto::PredictionResult& prediction_result) {
if (!IsValidResult(prediction_result)) {
return {};
}
const std::vector<float> model_scores(prediction_result.result().begin(),
prediction_result.result().end());
const proto::Predictor& predictor =
prediction_result.output_config().predictor();
switch (predictor.PredictorType_case()) {
case proto::Predictor::kBinaryClassifier:
return GetBinaryClassifierResults(model_scores,
predictor.binary_classifier());
case proto::Predictor::kMultiClassClassifier:
return GetMultiClassClassifierResults(model_scores,
predictor.multi_class_classifier());
case proto::Predictor::kBinnedClassifier:
return GetBinnedClassifierResults(model_scores,
predictor.binned_classifier());
default:
NOTREACHED();
}
}
std::vector<std::string> PostProcessor::GetBinaryClassifierResults(
const std::vector<float>& model_scores,
const proto::Predictor::BinaryClassifier& binary_classifier) const {
DCHECK_EQ(1u, model_scores.size());
const std::string& winning_label =
(model_scores[0] >= binary_classifier.threshold()
? binary_classifier.positive_label()
: binary_classifier.negative_label());
return std::vector<std::string>(1, winning_label);
}
std::vector<std::string> PostProcessor::GetMultiClassClassifierResults(
const std::vector<float>& model_scores,
const proto::Predictor::MultiClassClassifier& multi_class_classifier)
const {
DCHECK_EQ(static_cast<int>(model_scores.size()),
multi_class_classifier.class_labels_size());
CHECK(!(multi_class_classifier.has_threshold() &&
multi_class_classifier.class_thresholds_size() > 0))
<< "threshold and class_thresholds can't be both set at the same time";
if (multi_class_classifier.class_thresholds_size() > 0) {
CHECK_EQ(static_cast<int>(model_scores.size()),
multi_class_classifier.class_thresholds_size());
}
std::vector<std::pair<std::string, float>> labeled_results;
for (int index = 0; index < static_cast<int>(model_scores.size()); index++) {
if (!IsScoreBelowMultiClassThreshold(multi_class_classifier,
model_scores[index], index)) {
labeled_results.emplace_back(multi_class_classifier.class_labels(index),
model_scores[index]);
}
}
// Sort the labels in descending order of score.
std::stable_sort(labeled_results.begin(), labeled_results.end(),
[](const std::pair<std::string, float>& a,
const std::pair<std::string, float>& b) {
return a.second > b.second;
});
int elements_to_return =
std::min(multi_class_classifier.top_k_outputs(),
static_cast<int64_t>(labeled_results.size()));
std::vector<std::string> top_k_output_labels;
for (int index = 0; index < elements_to_return; index++) {
top_k_output_labels.emplace_back(labeled_results[index].first);
}
return top_k_output_labels;
}
std::vector<std::string> PostProcessor::GetBinnedClassifierResults(
const std::vector<float>& model_scores,
const proto::Predictor::BinnedClassifier& binned_classifier) const {
DCHECK_EQ(1u, model_scores.size());
DCHECK_LE(1, binned_classifier.bins_size());
std::string winning_bin_label = binned_classifier.underflow_label();
for (int index = 0; index < binned_classifier.bins_size(); index++) {
if (model_scores[0] >= binned_classifier.bins(index).min_range()) {
winning_bin_label = binned_classifier.bins(index).label();
}
}
return std::vector<std::string>(1, winning_bin_label);
}
ClassificationResult PostProcessor::GetPostProcessedClassificationResult(
const proto::PredictionResult& prediction_result,
PredictionStatus status) {
if (!IsValidResult(prediction_result)) {
// The post processing failed, mark the result as failure for the clients.
if (status == PredictionStatus::kSucceeded) {
status = PredictionStatus::kFailed;
}
return ClassificationResult(status);
}
ClassificationResult classification_result = ClassificationResult(status);
classification_result.ordered_labels =
GetClassifierResults(prediction_result);
return classification_result;
}
int PostProcessor::GetIndexOfTopLabel(
const proto::PredictionResult& prediction_result) {
if (!IsValidResult(prediction_result)) {
return kInvalidResult;
}
const std::vector<std::string>& result_labels =
GetClassifierResults(prediction_result);
if (result_labels.empty()) {
return kNoWinningLabel;
}
const std::string& top_label = result_labels[0];
const auto& predictor = prediction_result.output_config().predictor();
switch (predictor.PredictorType_case()) {
case proto::Predictor::kBinaryClassifier: {
bool bool_result =
(top_label == predictor.binary_classifier().positive_label());
return static_cast<int>(bool_result);
}
case proto::Predictor::kMultiClassClassifier: {
const auto& multi_class_classifier = predictor.multi_class_classifier();
for (int i = 0; i < multi_class_classifier.class_labels_size(); i++) {
if (top_label == multi_class_classifier.class_labels(i)) {
return i;
}
}
NOTREACHED();
}
case proto::Predictor::kBinnedClassifier: {
const auto& binned_classifier = predictor.binned_classifier();
if (top_label == binned_classifier.underflow_label()) {
return kUnderflowBinIndex;
}
for (int i = 0; i < binned_classifier.bins_size(); i++) {
if (top_label == binned_classifier.bins(i).label()) {
return i;
}
}
NOTREACHED();
}
default:
NOTREACHED();
}
}
base::TimeDelta PostProcessor::GetTTLForPredictedResult(
const proto::PredictionResult& prediction_result) {
std::vector<std::string> ordered_labels;
if (prediction_result.result_size() > 0 &&
prediction_result.has_output_config()) {
if (IsClassificationResult(prediction_result)) {
ordered_labels = GetClassifierResults(prediction_result);
}
if (!prediction_result.output_config().has_predicted_result_ttl()) {
LOG(ERROR) << "Prediction result has no `predicted_result_ttl` on its "
"`output_config`, returning empty TTL.";
return base::TimeDelta();
}
const auto& predicted_result_ttl =
prediction_result.output_config().predicted_result_ttl();
const auto& top_label_to_ttl_map =
predicted_result_ttl.top_label_to_ttl_map();
auto default_ttl = predicted_result_ttl.default_ttl();
const auto time_unit = predicted_result_ttl.time_unit();
if (ordered_labels.empty()) {
return default_ttl * metadata_utils::ConvertToTimeDelta(time_unit);
}
const auto iter = top_label_to_ttl_map.find(ordered_labels[0]);
int64_t ttl_to_use =
iter == top_label_to_ttl_map.end() ? default_ttl : iter->second;
return ttl_to_use * metadata_utils::ConvertToTimeDelta(time_unit);
}
return base::TimeDelta();
}
RawResult PostProcessor::GetRawResult(
const proto::PredictionResult& prediction_result,
PredictionStatus status) {
if (status != PredictionStatus::kSucceeded) {
return RawResult(status);
}
if (!IsValidResult(prediction_result)) {
return RawResult(PredictionStatus::kFailed);
}
RawResult result(status);
result.result = std::move(prediction_result);
return result;
}
} // namespace segmentation_platform
|
