pipeline.append(trainer)
# Finally, call evaluate() on the workflow DAG outputs, This will
# automatically append Evaluators to compute metrics from the given
# SavedModel and 'eval' TF Examples.
self.evaluate(
pipeline,
examples=task.train_and_eval_examples,
model=trainer.outputs.model)
if __name__ == '__main__':
run_config = dict(
pipeline_name=config.PIPELINE_NAME + '_openML',
data_dir=config.OTHER_DOWNLOAD_DIR,
)
if config.USE_KUBEFLOW:
# We need the string "KubeflowDagRunner" in this file to appease the
# validator used in `tfx create pipeline`.
# Validator: https://github.com/tensorflow/tfx/blob/v0.22.0/tfx/tools/cli/handler/base_handler.py#L105
nitroml.main(
pipeline_root=config.PIPELINE_ROOT,
tfx_runner=nitroml.get_default_kubeflow_dag_runner(),
**run_config)
else:
# This example has not been tested with engines other than Kubeflow.
nitroml.main(**run_config)
examples=task.train_and_eval_examples,
preprocessor=ad.BasicPreprocessor()))
# Define a Trainer to train our model on the given task.
trainer = self.add(
at.AutoTrainer(
problem_statement=task.problem_statement,
transformed_examples=autodata.outputs.transformed_examples,
transform_graph=autodata.outputs.transform_graph,
schema=autodata.outputs.schema,
train_steps=1000,
eval_steps=500,
enable_tuning=enable_tuning))
# Finally, call evaluate() on the workflow DAG outputs. This will
# automatically append Evaluators to compute metrics from the given
# SavedModel and 'eval' TF Examples.
self.evaluate(task=task, model=trainer.outputs.model)
if __name__ == '__main__':
if config.USE_KUBEFLOW:
# We need the string "KubeflowDagRunner" in this file to appease the
# validator used in `tfx create pipeline`.
# Validator: https://github.com/tensorflow/tfx/blob/v0.22.0/tfx/tools/cli/handler/base_handler.py#L105
nitroml.main(pipeline_name=config.PIPELINE_NAME + '_titanic',
pipeline_root=config.PIPELINE_ROOT,
data_dir=config.TF_DOWNLOAD_DIR,
tfx_runner=nitroml.get_default_kubeflow_dag_runner())
else:
nitroml.main()
task_pipeline = task.components + autodata.components + test_meta_components + [
trainer
]
# Finally, call evaluate() on the workflow DAG outputs, This will
# automatically append Evaluators to compute metrics from the given
# SavedModel and 'eval' TF Examples.ss
self.evaluate(task_pipeline,
examples=task.train_and_eval_examples,
model=trainer.outputs.model)
if __name__ == '__main__':
metalearning_algorithm = 'nearest_neighbor'
run_config = dict(pipeline_name=f'metalearning_{metalearning_algorithm}',
data_dir=config.OTHER_DOWNLOAD_DIR,
algorithm=metalearning_algorithm)
if config.USE_KUBEFLOW:
# We need the string "KubeflowDagRunner" in this file to appease the
# validator used in `tfx create pipeline`.
# Validator: https://github.com/tensorflow/tfx/blob/v0.22.0/tfx/tools/cli/handler/base_handler.py#L105
nitroml.main(pipeline_root=os.path.join(config.PIPELINE_ROOT,
run_config['pipeline_name']),
tfx_runner=nitroml.get_default_kubeflow_dag_runner(),
**run_config)
else:
# This example has not been tested with engines other than Kubeflow.
nitroml.main(**run_config)