def preprocessing_fn(self, inputs: Dict):
"""
Standard preprocessing function.
Args:
inputs: Dict mapping features to their respective data.
Returns:
output: Dict mapping transformed features to their respective
transformed data.
"""
schema = infer_schema(inputs)
output = {}
for key, value in inputs.items():
if key not in self.f_dict:
self.f_dict[key] = self.f_d_dict[schema[key]]
value = self.apply_filling(value, self.f_dict[key])
if key in self.features or key in self.labels:
if key not in self.t_dict:
self.t_dict[key] = self.t_d_dict[schema[key]]
result = self.apply_transform(key, value, self.t_dict[key])
result = tf.cast(result, dtype=tf.float32)
if key in self.features:
output[naming_utils.transformed_feature_name(key)] = result
if key in self.labels:
output[naming_utils.transformed_label_name(key)] = result
output[key] = value
return output
}]
}
}))
training_pipeline.add_trainer(
TFFeedForwardTrainer(loss='binary_crossentropy',
last_activation='sigmoid',
output_units=1,
metrics=['accuracy'],
epochs=20))
# Add an evaluator
training_pipeline.add_evaluator(
TFMAEvaluator(slices=[['has_diabetes']],
metrics={
transformed_label_name('has_diabetes'):
['binary_crossentropy', 'binary_accuracy']
}))
# Add the deployer
training_pipeline.add_deployment(
GCAIPDeployer(
project_id=GCP_PROJECT,
model_name=MODEL_NAME,
))
# Run the pipeline
training_pipeline.run()
# Another way to do is is to create a DeploymentPipeline.
# Uncomment to create the model via this pipeline
],
labels=['has_diabetes'],
overwrite={
'has_diabetes': {
'transform': [{
'method': 'no_transform',
'parameters': {}
}]
}
}))
# Add a trainer
training_pipeline.add_trainer(MyScikitTrainer(
C=0.8,
kernel='rbf',
))
# Add an evaluator
label_name = naming_utils.transformed_label_name('has_diabetes')
training_pipeline.add_evaluator(
AgnosticEvaluator(prediction_key=naming_utils.output_name(label_name),
label_key=label_name,
slices=[['has_diabetes']],
metrics=['mean_squared_error']))
# Run the pipeline locally
training_pipeline.run()
# Evaluate
training_pipeline.evaluate()
}
}))
# Add a trainer
training_pipeline.add_trainer(
TFFeedForwardTrainer(loss='binary_crossentropy',
last_activation='sigmoid',
output_units=1,
metrics=['accuracy'],
epochs=20))
# Add an evaluator
training_pipeline.add_evaluator(
TFMAEvaluator(slices=[['has_diabetes']],
metrics={
transformed_label_name('has_diabetes'):
['binary_crossentropy', 'binary_accuracy']
}))
# Run the pipeline locally
training_pipeline.run()
# Run inference
model_uri = training_pipeline.get_model_uri()
infer_pipeline = BatchInferencePipeline(model_uri=model_uri, )
infer_pipeline.add_datasource(ds)
infer_pipeline.add_infer_step(
TensorflowInferrer(labels=[transformed_label_name('has_diabetes')]))
infer_pipeline.run()
df = infer_pipeline.get_predictions()
print(df.head())