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
|
// 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/execution/processing/uma_feature_processor.h"
#include "base/functional/bind.h"
#include "base/location.h"
#include "base/memory/weak_ptr.h"
#include "base/task/sequenced_task_runner.h"
#include "base/timer/elapsed_timer.h"
#include "components/segmentation_platform/internal/database/signal_database.h"
#include "components/segmentation_platform/internal/execution/processing/feature_aggregator.h"
#include "components/segmentation_platform/internal/execution/processing/feature_processor_state.h"
#include "components/segmentation_platform/internal/metadata/metadata_utils.h"
#include "components/segmentation_platform/internal/stats.h"
#include "components/segmentation_platform/public/proto/aggregation.pb.h"
#include "components/segmentation_platform/public/proto/model_metadata.pb.h"
#include "components/segmentation_platform/public/proto/types.pb.h"
namespace segmentation_platform::processing {
namespace {
const proto::UMAFeature& GetAsUMA(const Data& data) {
DCHECK(data.input_feature.has_value() || data.output_feature.has_value());
if (data.input_feature.has_value()) {
return data.input_feature->uma_feature();
}
return data.output_feature->uma_output().uma_feature();
}
} // namespace
UmaFeatureProcessor::UmaFeatureProcessor(
base::flat_map<FeatureIndex, Data>&& uma_features,
SignalDatabase* signal_database,
FeatureAggregator* feature_aggregator,
const base::Time prediction_time,
const base::Time observation_time,
const base::TimeDelta bucket_duration,
const SegmentId segment_id,
bool is_output)
: uma_features_(std::move(uma_features)),
signal_database_(signal_database),
feature_aggregator_(feature_aggregator),
prediction_time_(prediction_time),
observation_time_(observation_time),
bucket_duration_(bucket_duration),
segment_id_(segment_id),
is_output_(is_output) {}
UmaFeatureProcessor::~UmaFeatureProcessor() = default;
void UmaFeatureProcessor::Process(
std::unique_ptr<FeatureProcessorState> feature_processor_state,
QueryProcessorCallback callback) {
feature_processor_state_ = std::move(feature_processor_state);
callback_ = std::move(callback);
size_t max_bucket_count = 0;
for (const auto& feature : uma_features_) {
// Validate the proto::UMAFeature metadata.
const proto::UMAFeature& uma_feature = GetAsUMA(feature.second);
if (metadata_utils::ValidateMetadataUmaFeature(uma_feature) !=
metadata_utils::ValidationResult::kValidationSuccess) {
feature_processor_state_->SetError(
stats::FeatureProcessingError::kUmaValidationError);
base::SequencedTaskRunner::GetCurrentDefault()->PostTask(
FROM_HERE, base::BindOnce(std::move(callback_),
std::move(feature_processor_state_),
std::move(result_)));
return;
}
if (max_bucket_count < uma_feature.bucket_count()) {
max_bucket_count = uma_feature.bucket_count();
}
}
ProcessOnGotAllSamples(*signal_database_->GetAllSamples());
}
void UmaFeatureProcessor::GetStartAndEndTime(size_t bucket_count,
base::Time& start_time,
base::Time& end_time) const {
base::TimeDelta duration = bucket_duration_ * bucket_count;
if (is_output_) {
if (observation_time_ == base::Time()) {
start_time = prediction_time_ - duration;
end_time = prediction_time_;
} else if (observation_time_ - prediction_time_ > duration) {
start_time = observation_time_ - duration;
end_time = observation_time_;
} else {
start_time = prediction_time_;
end_time = observation_time_;
}
} else {
start_time = prediction_time_ - duration;
end_time = prediction_time_;
}
}
void UmaFeatureProcessor::ProcessOnGotAllSamples(
const std::vector<SignalDatabase::DbEntry>& samples) {
while (!uma_features_.empty()) {
if (feature_processor_state_->error()) {
break;
}
const auto& it = uma_features_.begin();
proto::UMAFeature next_feature = GetAsUMA(it->second);
FeatureIndex index = it->first;
uma_features_.erase(it);
ProcessSingleUmaFeature(samples, index, next_feature);
}
base::SequencedTaskRunner::GetCurrentDefault()->PostTask(
FROM_HERE,
base::BindOnce(std::move(callback_), std::move(feature_processor_state_),
std::move(result_)));
}
void UmaFeatureProcessor::ProcessSingleUmaFeature(
const std::vector<SignalDatabase::DbEntry>& samples,
FeatureIndex index,
const proto::UMAFeature& feature) {
// Enum histograms can optionally only accept some of the enum values.
// While the proto::UMAFeature is available, capture a vector of the
// accepted enum values. An empty vector is ignored (all values are
// considered accepted).
std::vector<int32_t> accepted_enum_ids{};
if (feature.type() == proto::SignalType::HISTOGRAM_ENUM) {
for (int i = 0; i < feature.enum_ids_size(); ++i) {
accepted_enum_ids.emplace_back(feature.enum_ids(i));
}
}
base::Time start_time;
base::Time end_time;
GetStartAndEndTime(feature.bucket_count(), start_time, end_time);
base::ElapsedTimer timer;
// We now have all the data required to process a single feature, so we can
// process it synchronously, and insert it into the
// FeatureProcessorState::input_tensor so we can later pass it to the ML model
// executor.
absl::optional<std::vector<float>> result = feature_aggregator_->Process(
feature.type(), feature.name_hash(), feature.aggregation(),
feature.bucket_count(), start_time, end_time, bucket_duration_,
accepted_enum_ids, samples);
// If no feature data is available, use the default values specified instead.
if (result.has_value()) {
const std::vector<float>& feature_data = result.value();
DCHECK_EQ(feature.tensor_length(), feature_data.size());
result_[index] =
std::vector<ProcessedValue>(feature_data.begin(), feature_data.end());
} else {
DCHECK_EQ(feature.tensor_length(),
static_cast<unsigned int>(feature.default_values_size()))
<< " Mismatch between expected value size and default value size for "
"UMA feature '"
<< feature.name()
<< "'. Did you forget to specify a default value for this feature?";
result_[index] = std::vector<ProcessedValue>(
feature.default_values().begin(), feature.default_values().end());
}
stats::RecordModelExecutionDurationFeatureProcessing(segment_id_,
timer.Elapsed());
}
} // namespace segmentation_platform::processing
|
