def main():
"""Trains a model locally to test get_model()."""
train_x, train_y, eval_x, eval_y = load_data()
train_y, eval_y = [np.ravel(x) for x in [train_y, eval_y]]
params = argparse.Namespace(C=1.0)
model = get_model(params)
model.fit(train_x, train_y)
score = model.score(eval_x, eval_y)
print(score)
def main():
config = "config.yaml"
model = TFModel(config)
model.generate_files()
_upload_data_to_gcs(model)
pipeline = KfpPipeline(model)
# preprocess and upload dataset to expected location.
load_data(model.data["train"], model.data["evaluation"])
# define pipeline structure
p = pipeline.add_train_component()
pipeline.add_deploy_component(parent=p)
pipeline.add_predict_component(parent=p)
pipeline.print_structure()
pipeline.generate_pipeline()
# Create batch prediction data in GCS.
pred_input = [{
"age": 0.02599666,
"workclass": 6,
"education_num": 1.1365801,
"marital_status": 4,
"occupation": 0,
"relationship": 1,
"race": 4,
"capital_gain": 0.14693314,
"capital_loss": -0.21713187,
"hours_per_week": -0.034039237,
"native_country": 38,
"income_bracket": 0,
}]
_upload_input_data_to_gcs(model, pred_input)
# Run the pipeline.
# pylint: disable=import-outside-toplevel
from orchestration import pipeline as kfp_pipeline
kfp_pipeline.main()
def _upload_data_to_gcs(model):
"""Calls the preprocessing fn which uploads train/eval data to GCS."""
load_data(model.data["train"], model.data["evaluation"])
def _upload_data_to_gcs(model):
load_data(model.data["train"], model.data["evaluation"])