-
Notifications
You must be signed in to change notification settings - Fork 276
/
slice_key_extractor.py
146 lines (125 loc) · 6.15 KB
/
slice_key_extractor.py
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
# Copyright 2018 Google LLC
#
# 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
#
# https://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.
"""Public API for extracting slice keys."""
import copy
from typing import List, Optional
import apache_beam as beam
from tensorflow_model_analysis import constants
from tensorflow_model_analysis import types
from tensorflow_model_analysis.extractors import extractor
from tensorflow_model_analysis.proto import config_pb2
from tensorflow_model_analysis.slicer import slicer_lib as slicer
from tensorflow_model_analysis.utils import util
SLICE_KEY_EXTRACTOR_STAGE_NAME = 'ExtractSliceKeys'
def SliceKeyExtractor(
slice_spec: Optional[List[slicer.SingleSliceSpec]] = None,
eval_config: Optional[config_pb2.EvalConfig] = None,
materialize: Optional[bool] = True) -> extractor.Extractor:
"""Creates an extractor for extracting slice keys.
The incoming Extracts must contain features stored under tfma.FEATURES_KEY
and optionally under tfma.TRANSFORMED_FEATURES.
The extractor's PTransform yields a copy of the Extracts input with an
additional extract pointing at the list of SliceKeyType values keyed by
tfma.SLICE_KEY_TYPES_KEY. If materialize is True then a materialized version
of the slice keys will be added under the key tfma.SLICE_KEYS_KEY.
Args:
slice_spec: Deprecated (use EvalConfig).
eval_config: Optional EvalConfig containing slicing_specs specifying the
slices to slice the data into. If slicing_specs are empty, defaults to
overall slice.
materialize: True to add MaterializedColumn entries for the slice keys.
Returns:
Extractor for slice keys.
"""
if slice_spec and eval_config:
raise ValueError('slice_spec is deprecated, only use eval_config')
if eval_config:
slice_spec = [
slicer.SingleSliceSpec(spec=spec) for spec in eval_config.slicing_specs
]
for cross_slice_spec in eval_config.cross_slicing_specs:
baseline_slice_spec = slicer.SingleSliceSpec(
spec=cross_slice_spec.baseline_spec)
if baseline_slice_spec not in slice_spec:
slice_spec.append(baseline_slice_spec)
for spec in cross_slice_spec.slicing_specs:
comparison_slice_spec = slicer.SingleSliceSpec(spec=spec)
if comparison_slice_spec not in slice_spec:
slice_spec.append(comparison_slice_spec)
if not slice_spec:
slice_spec = [slicer.SingleSliceSpec()]
return extractor.Extractor(
stage_name=SLICE_KEY_EXTRACTOR_STAGE_NAME,
ptransform=ExtractSliceKeys(slice_spec, eval_config, materialize))
@beam.typehints.with_input_types(types.Extracts, List[slicer.SingleSliceSpec])
@beam.typehints.with_output_types(types.Extracts)
class ExtractSliceKeysFn(beam.DoFn):
"""A DoFn that extracts slice keys that apply per example."""
def __init__(self, eval_config: Optional[config_pb2.EvalConfig],
materialize: bool):
self._eval_config = eval_config
self._materialize = materialize
self._duplicate_slice_keys_counter = beam.metrics.Metrics.counter(
constants.METRICS_NAMESPACE, 'num_examples_with_duplicate_slice_keys')
def process(self, element: types.Extracts,
slice_spec: List[slicer.SingleSliceSpec]) -> List[types.Extracts]:
# Slice on transformed features if available.
features_dicts = []
if (constants.TRANSFORMED_FEATURES_KEY in element and
element[constants.TRANSFORMED_FEATURES_KEY] is not None):
transformed_features = element[constants.TRANSFORMED_FEATURES_KEY]
# If only one model, the output is stored without keying on model name.
if not self._eval_config or len(self._eval_config.model_specs) == 1:
features_dicts.append(transformed_features)
else:
# Search for slices in each model's transformed features output.
for spec in self._eval_config.model_specs:
if spec.name in transformed_features:
features_dicts.append(transformed_features[spec.name])
# Search for slices first in transformed features (if any). If a match is
# not found there then search in raw features.
slice_keys = list(
slicer.get_slices_for_features_dicts(
features_dicts, util.get_features_from_extracts(element),
slice_spec))
# If SLICE_KEY_TYPES_KEY already exists, that means the
# SqlSliceKeyExtractor has generated some slice keys. We need to add
# them to current slice_keys list.
if (constants.SLICE_KEY_TYPES_KEY in element and
element[constants.SLICE_KEY_TYPES_KEY]):
slice_keys.extend(element[constants.SLICE_KEY_TYPES_KEY])
unique_slice_keys = list(set(slice_keys))
if len(slice_keys) != len(unique_slice_keys):
self._duplicate_slice_keys_counter.inc()
# Make a a shallow copy, so we don't mutate the original.
element_copy = copy.copy(element)
element_copy[constants.SLICE_KEY_TYPES_KEY] = (
slicer.slice_keys_to_numpy_array(unique_slice_keys))
# Add a list of stringified slice keys to be materialized to output table.
if self._materialize:
element_copy[constants.SLICE_KEYS_KEY] = types.MaterializedColumn(
name=constants.SLICE_KEYS_KEY,
value=(list(
slicer.stringify_slice_key(x).encode('utf-8')
for x in unique_slice_keys)))
return [element_copy]
@beam.ptransform_fn
@beam.typehints.with_input_types(types.Extracts)
@beam.typehints.with_output_types(types.Extracts)
def ExtractSliceKeys(extracts: beam.pvalue.PCollection,
slice_spec: List[slicer.SingleSliceSpec],
eval_config: Optional[config_pb2.EvalConfig] = None,
materialize: bool = True) -> beam.pvalue.PCollection:
return extracts | beam.ParDo(
ExtractSliceKeysFn(eval_config, materialize), slice_spec=slice_spec)