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
|
// 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.
syntax = "proto2";
option optimize_for = LITE_RUNTIME;
option java_package = "org.chromium.components.segmentation_platform.proto";
option java_outer_classname = "OutputConfigProto";
package segmentation_platform.proto;
// Used as time unit for rest of this proto.
enum TimeUnit {
UNKNOWN_TIME_UNIT = 0;
YEAR = 1;
MONTH = 2;
WEEK = 3;
DAY = 4;
HOUR = 5;
MINUTE = 6;
SECOND = 7;
}
// Defines what type of model is supplied. Results are based on which
// classifier/regressor the model is in.
message Predictor {
// A classifier to interpret model results as a boolean. The final result
// supplied to the client `ordered_lables` of length one, with either
// `positive_label` or `negative label`.
message BinaryClassifier {
// Model score below threshold is classified as `negative_label` else
// `positive_label`.
optional float threshold = 1;
// Labels for positive and negative outputs.
optional string positive_label = 2;
optional string negative_label = 3;
}
// A classifier to interpret model results as one of multiple classes. Each
// output of the model corresponds to one of the classes. The number of labels
// should be equal to the number of outputs from the model. The final result
// supplied to the client is list of `ordered_labels` of size `top_k_outputs`.
message MultiClassClassifier {
// Number of top results the client is interested in. Must be less than
// or equal to the number of labels specified.
optional int64 top_k_outputs = 1;
// Class labels associated with the outputs. The number of outputs from the
// model must be equal to the length of this list. Each label maps to the
// corresponding output index.
repeated string class_labels = 2;
// Threshold to get `class_labels` with score above it. Only labels with
// score above or equal to threshold are returned. Must not be used along
// with `class_thresholds`.
optional float threshold = 3;
// Per-class thresholds. Used to assign a different threshold to each class
// so that class_labels[n] is returned when outputs[n] >=
// class_thresholds[n]. Length of this list must be the same as the length
// of `class_labels`. Must not be used along with `threshold`.
repeated float class_thresholds = 4;
}
// A post-processor that converts a continuous model score into discrete
// ranges. The final result supplied to the client is one of the bin labels or
// `underflow_label`.
message BinnedClassifier {
// A bin represents a bucket containing entries.
message Bin {
// Represents minimum range for this bin. Entries are between the current
// bin `min_range` (inclusive) to the next bin `min_range` (exclusive).
// Range of bin : [min_range , next-bin min_range)
optional float min_range = 1;
// Signify what the bin corresponds to.
optional string label = 2;
}
// BinnedClassifier contains multiple bins arranged in ascending order of
// min_range of the bins. The result is `ordered_labels` of length 1 with
// selected bin's label.
repeated Bin bins = 1;
// In case the model score is less than the `min_range` of the first label,
// `underflow_label` is returned. If not supplied, then empty vector is
// returned.
optional string underflow_label = 2;
}
// Describes a regression model.
message Regressor {}
// Used when the model does not fit in any of the above descriptions. Contains
// basic description about the model output. Model result is vector of floats
// with labels.
message GenericPredictor {
// Label assigned to each corresponding output value returned by the model.
// The number of outputs from the model must be equal to the length of this
// list.
repeated string output_labels = 1;
}
// Specifies how to post process the raw model results based on model type,
// e.g. classification or regression.
oneof PredictorType {
BinaryClassifier binary_classifier = 1;
MultiClassClassifier multi_class_classifier = 2;
BinnedClassifier binned_classifier = 3;
Regressor regressor = 4;
GenericPredictor generic_predictor = 5;
}
}
// TTL for the result, unit in TimeUnit.
message PredictedResultTTL {
// TTL for the predicted result can be optionally configured based on the top
// result. For example, for a ClassificationResult with
// `ordered_labels` B, A, C, the TTL value of B in `top_label_to_ttl_map` will
// be used as the result TTL. This field is only valid for Classifier
// predictor (binary, multi class and binned).
map<string, uint64> top_label_to_ttl_map = 1;
// TTL when no other TTL is specified for a particular label.
optional uint64 default_ttl = 2;
// The time unit to be used for TTL.
optional TimeUnit time_unit = 3;
}
// Specified by client in the metadata on how to interpret the model results.
message OutputConfig {
// A post-processor associated with the model.
optional Predictor predictor = 1;
// TTL associated with the PredictionResult. Config defines when to refresh
// results.
optional PredictedResultTTL predicted_result_ttl = 2;
// Flag to ignore previous model TTL. If this is set to true, ttl is ignored
// and results are refreshed on model update. This is false by default. It
// doesn't have any effect on TTL.
optional bool ignore_previous_model_ttl = 3;
}
|
