-
Notifications
You must be signed in to change notification settings - Fork 706
/
ai_platform_training_component.py
278 lines (248 loc) · 10.3 KB
/
ai_platform_training_component.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
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
277
278
# Lint as: python3
# Copyright 2020 Google LLC. 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.
"""Component that launches CAIP custom training job with flexible interface."""
from typing import Any, Dict, List, Optional, Text
from tfx.dsl.component.experimental import component_utils
from tfx.dsl.component.experimental import placeholders
from tfx.dsl.components.base import base_component
from tfx.dsl.components.base import executor_spec
from tfx.orchestration.kubeflow.v2.components.experimental import ai_platform_training_executor
from tfx.types import channel_utils
from tfx.types import component_spec
from tfx.utils import json_utils
def create_ai_platform_training(
name: Text,
project_id: Text,
region: Optional[Text] = None,
job_id: Optional[Text] = None,
image_uri: Optional[Text] = None,
args: Optional[List[placeholders.CommandlineArgumentType]] = None,
# TODO(jxzheng): support Python training spec
scale_tier: Optional[Text] = None,
training_input: Optional[Dict[Text, Any]] = None,
labels: Optional[Dict[Text, Text]] = None,
inputs: Dict[Text, Any] = None,
outputs: Dict[Text, Any] = None,
parameters: Dict[Text, Any] = None,
) -> base_component.BaseComponent:
"""Creates a pipeline step that launches a AIP training job.
The generated TFX component will have a component spec specified dynamically,
through inputs/outputs/parameters in the following format:
- inputs: A mapping from input name to the upstream channel connected. The
artifact type of the channel will be automatically inferred.
- outputs: A mapping from output name to the associated artifact type.
- parameters: A mapping from execution property names to its associated value.
Only primitive typed values are supported. Note that RuntimeParameter is
not supported yet.
For example:
```
create_ai_platform_training(
...
inputs: {
# Assuming there is an upstream node example_gen, with an output
# 'examples' of the type Examples.
'examples': example_gen.outputs['examples'],
},
outputs: {
'model': standard_artifacts.Model,
},
parameters: {
'n_steps': 100,
'optimizer': 'sgd',
}
...
)
```
will generate a component instance with a component spec equivalent to:
```
class MyComponentSpec(ComponentSpec):
INPUTS = {
'examples': ChannelParameter(type=standard_artifacts.Examples)
}
OUTPUTS = {
'model': ChannelParameter(type=standard_artifacts.Model)
}
PARAMETERS = {
'n_steps': ExecutionParameter(type=int),
'optimizer': ExecutionParameter(type=str)
}
```
with its input 'examples' is connected to the example_gen output, and
execution properties specified as 100 and 'sgd' respectively.
Example usage of the component:
```
# A single node training job.
my_train = create_ai_platform_training(
name='my_training_step',
project_id='my-project',
region='us-central1',
image_uri='gcr.io/my-project/caip-training-test:latest',
'args': [
'--examples',
placeholders.InputUriPlaceholder('examples'),
'--n-steps',
placeholders.InputValuePlaceholder('n_step'),
'--output-location',
placeholders.OutputUriPlaceholder('model')
]
scale_tier='BASIC_GPU',
inputs={'examples': example_gen.outputs['examples']},
outputs={
'model': standard_artifacts.Model
},
parameters={'n_step': 100}
)
# More complex setting can be expressed by providing training_input
# directly.
my_distributed_train = create_ai_platform_training(
name='my_training_step',
project_id='my-project',
training_input={
'scaleTier':
'CUSTOM',
'region':
'us-central1',
'masterType': 'n1-standard-8',
'masterConfig': {
'imageUri': 'gcr.io/my-project/my-dist-training:latest'
},
'workerType': 'n1-standard-8',
'workerCount': 8,
'workerConfig': {
'imageUri': 'gcr.io/my-project/my-dist-training:latest'
},
'args': [
'--examples',
placeholders.InputUriPlaceholder('examples'),
'--n-steps',
placeholders.InputValuePlaceholder('n_step'),
'--output-location',
placeholders.OutputUriPlaceholder('model')
]
},
inputs={'examples': example_gen.outputs['examples']},
outputs={'model': Model},
parameters={'n_step': 100}
)
```
Args:
name: name of the component. This is needed to construct the component spec
and component class dynamically as well.
project_id: the GCP project under which the AIP training job will be
running.
region: GCE region where the AIP training job will be running.
job_id: the unique ID of the job. Default to 'tfx_%Y%m%d%H%M%S'
image_uri: the GCR location of the container image, which will be used to
execute the training program. If the same field is specified in
training_input, the latter overrides image_uri.
args: command line arguments that will be passed into the training program.
Users can use placeholder semantics as in
tfx.dsl.component.experimental.container_component to wire the args with
component inputs/outputs/parameters.
scale_tier: Cloud ML resource requested by the job. See
https://cloud.google.com/ai-platform/training/docs/reference/rest/v1/projects.jobs#ScaleTier
training_input: full training job spec. This field overrides other
specifications if applicable. This field follows the
[TrainingInput](https://cloud.google.com/ai-platform/training/docs/reference/rest/v1/projects.jobs#traininginput)
schema.
labels: user-specified label attached to the job.
inputs: the dict of component inputs.
outputs: the dict of component outputs.
parameters: the dict of component parameters, aka, execution properties.
Returns:
A component instance that represents the AIP job in the DSL.
Raises:
ValueError: when image_uri is missing and masterConfig is not specified in
training_input, or when region is missing and training_input
does not provide region either.
TypeError: when non-primitive parameters are specified.
"""
training_input = training_input or {}
if scale_tier and not training_input.get('scale_tier'):
training_input['scaleTier'] = scale_tier
if not training_input.get('masterConfig'):
# If no replica config is specified, create a default one.
if not image_uri:
raise ValueError('image_uri is required when masterConfig is not '
'explicitly specified in training_input.')
training_input['masterConfig'] = {'imageUri': image_uri}
# Note: A custom entrypoint can be set to training_input['masterConfig']
# through key 'container_command'.
training_input['args'] = args
if not training_input.get('region'):
if not region:
raise ValueError('region is required when it is not set in '
'training_input.')
training_input['region'] = region
# Squash training_input, project, job_id, and labels into an exec property
# namely 'aip_training_config'.
aip_training_config = {
ai_platform_training_executor.PROJECT_CONFIG_KEY: project_id,
ai_platform_training_executor.TRAINING_INPUT_CONFIG_KEY: training_input,
ai_platform_training_executor.JOB_ID_CONFIG_KEY: job_id,
ai_platform_training_executor.LABELS_CONFIG_KEY: labels,
}
aip_training_config_str = json_utils.dumps(aip_training_config)
# Construct the component spec.
if inputs is None:
inputs = {}
if outputs is None:
outputs = {}
if parameters is None:
parameters = {}
input_channel_parameters = {}
output_channel_parameters = {}
output_channels = {}
execution_parameters = {
ai_platform_training_executor.CONFIG_KEY:
component_spec.ExecutionParameter(type=(str, Text))
}
for input_name, single_channel in inputs.items():
# Infer the type of input channels based on the channels passed in.
# TODO(b/155804245) Sanitize the names so that they're valid python names
input_channel_parameters[input_name] = (
component_spec.ChannelParameter(type=single_channel.type))
for output_name, channel_type in outputs.items():
# TODO(b/155804245) Sanitize the names so that they're valid python names
output_channel_parameters[output_name] = (
component_spec.ChannelParameter(type=channel_type))
artifact = channel_type()
channel = channel_utils.as_channel([artifact])
output_channels[output_name] = channel
# TODO(jxzheng): Support RuntimeParameter as parameters.
for param_name, single_parameter in parameters.items():
# Infer the type of parameters based on the parameters passed in.
# TODO(b/155804245) Sanitize the names so that they're valid python names
if not isinstance(single_parameter, (int, float, Text, bytes)):
raise TypeError(
'Parameter can only be int/float/str/bytes, got {}'.format(
type(single_parameter)))
execution_parameters[param_name] = (
component_spec.ExecutionParameter(type=type(single_parameter)))
default_init_args = {
**inputs,
**output_channels,
**parameters, ai_platform_training_executor.CONFIG_KEY:
aip_training_config_str
}
tfx_component_class = component_utils.create_tfx_component_class(
name=name,
tfx_executor_spec=executor_spec.ExecutorClassSpec(
ai_platform_training_executor.AiPlatformTrainingExecutor),
input_channel_parameters=input_channel_parameters,
output_channel_parameters=output_channel_parameters,
execution_parameters=execution_parameters,
default_init_args=default_init_args)
return tfx_component_class()