Vertex AI Pipelines에서 커스텀 ML 모델 학습

이 튜토리얼에서는 Vertex AI Pipelines를 사용하여 엔드 투 엔드 ML 워크플로를 실행하는 방법을 보여주며, 여기에는 다음 작업이 포함됩니다.

  • 데이터를 가져오고 변환합니다.
  • 선택한 ML 프레임워크를 사용하여 모델을 학습합니다.
  • 학습된 모델을 Vertex AI Model Registry로 가져옵니다.
  • 선택사항: Vertex AI Prediction으로 온라인 서빙을 위한 모델을 배포합니다.

시작하기 전에

  1. Google Cloud 프로젝트 및 개발 환경 설정의 1~3 태스크를 완료했는지 확인합니다.

  2. Python용 Vertex AI SDK 및 Kubeflow Pipelines SDK를 설치합니다.

    python3 -m pip install "kfp<2.0.0" "google-cloud-aiplatform>=1.16.0" --upgrade --quiet

ML 모델 학습 파이프라인 실행

다음 탭에서 학습 목표 및 ML 프레임워크를 선택하여 환경에서 실행할 수 있는 샘플 코드를 가져옵니다. 이 샘플 코드는 다음을 수행합니다.

  • 파이프라인 빌딩 블록으로 사용할 구성요소 저장소에서 구성요소를 로드합니다.
  • 구성요소 태스크를 만들고 인수를 사용하여 구성요소 태스크 간에 데이터를 전달하여 파이프라인을 구성합니다.
  • Vertex AI Pipelines에서 실행할 파이프라인을 제출합니다. Vertex AI Pipelines 가격 책정을 참조하세요.

코드를 개발 환경에 복사하고 실행합니다.

테이블 형식 분류

TensorFlow

# python3 -m pip install "kfp<2.0.0" "google-cloud-aiplatform>=1.16.0" --upgrade --quiet
from kfp import components

# %% Loading components
download_from_gcs_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/storage/download/component.yaml")
select_columns_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Select_columns/in_CSV_format/component.yaml")
fill_all_missing_values_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Fill_all_missing_values/in_CSV_format/component.yaml")
binarize_column_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Binarize_column/in_CSV_format/component.yaml")
split_rows_into_subsets_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/dataset_manipulation/Split_rows_into_subsets/in_CSV/component.yaml")
create_fully_connected_tensorflow_network_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/tensorflow/Create_fully_connected_network/component.yaml")
train_model_using_Keras_on_CSV_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/tensorflow/Train_model_using_Keras/on_CSV/component.yaml")
predict_with_TensorFlow_model_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/tensorflow/Predict/on_CSV/component.yaml")
upload_Tensorflow_model_to_Google_Cloud_Vertex_AI_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Upload_Tensorflow_model/component.yaml")
deploy_model_to_endpoint_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Deploy_to_endpoint/component.yaml")

# %% Pipeline definition
def train_tabular_classification_model_using_TensorFlow_pipeline():
    dataset_gcs_uri = "gs://ml-pipeline-dataset/Chicago_taxi_trips/chicago_taxi_trips_2019-01-01_-_2019-02-01_limit=10000.csv"
    feature_columns = ["trip_seconds", "trip_miles", "pickup_community_area", "dropoff_community_area", "fare", "tolls", "extras"]  # Excluded "trip_total"
    label_column = "tips"
    training_set_fraction = 0.8
    # Deploying the model might incur additional costs over time
    deploy_model = False

    classification_label_column = "class"
    all_columns = [label_column] + feature_columns

    dataset = download_from_gcs_op(
        gcs_path=dataset_gcs_uri
    ).outputs["Data"]

    dataset = select_columns_using_Pandas_on_CSV_data_op(
        table=dataset,
        column_names=all_columns,
    ).outputs["transformed_table"]

    dataset = fill_all_missing_values_using_Pandas_on_CSV_data_op(
        table=dataset,
        replacement_value="0",
        # # Optional:
        # column_names=None,  # =[...]
    ).outputs["transformed_table"]

    classification_dataset = binarize_column_using_Pandas_on_CSV_data_op(
        table=dataset,
        column_name=label_column,
        predicate=" > 0",
        new_column_name=classification_label_column,
    ).outputs["transformed_table"]

    split_task = split_rows_into_subsets_op(
        table=classification_dataset,
        fraction_1=training_set_fraction,
    )
    classification_training_data = split_task.outputs["split_1"]
    classification_testing_data = split_task.outputs["split_2"]

    network = create_fully_connected_tensorflow_network_op(
        input_size=len(feature_columns),
        # Optional:
        hidden_layer_sizes=[10],
        activation_name="elu",
        output_activation_name="sigmoid",
        # output_size=1,
    ).outputs["model"]

    model = train_model_using_Keras_on_CSV_op(
        training_data=classification_training_data,
        model=network,
        label_column_name=classification_label_column,
        # Optional:
        loss_function_name="binary_crossentropy",
        number_of_epochs=10,
        #learning_rate=0.1,
        #optimizer_name="Adadelta",
        #optimizer_parameters={},
        #batch_size=32,
        #metric_names=["mean_absolute_error"],
        #random_seed=0,
    ).outputs["trained_model"]

    predictions = predict_with_TensorFlow_model_on_CSV_data_op(
        dataset=classification_testing_data,
        model=model,
        # label_column_name needs to be set when doing prediction on a dataset that has labels
        label_column_name=classification_label_column,
        # Optional:
        # batch_size=1000,
    ).outputs["predictions"]

    vertex_model_name = upload_Tensorflow_model_to_Google_Cloud_Vertex_AI_op(
        model=model,
    ).outputs["model_name"]

    # Deploying the model might incur additional costs over time
    if deploy_model:
        vertex_endpoint_name = deploy_model_to_endpoint_op(
            model_name=vertex_model_name,
        ).outputs["endpoint_name"]

pipeline_func = train_tabular_classification_model_using_TensorFlow_pipeline

# %% Pipeline submission
if __name__ == '__main__':
    from google.cloud import aiplatform
    aiplatform.PipelineJob.from_pipeline_func(pipeline_func=pipeline_func).submit()

PyTorch

# python3 -m pip install "kfp<2.0.0" "google-cloud-aiplatform>=1.16.0" --upgrade --quiet
from kfp import components

# %% Loading components
download_from_gcs_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/storage/download/component.yaml")
select_columns_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Select_columns/in_CSV_format/component.yaml")
fill_all_missing_values_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Fill_all_missing_values/in_CSV_format/component.yaml")
binarize_column_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Binarize_column/in_CSV_format/component.yaml")
create_fully_connected_pytorch_network_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/PyTorch/Create_fully_connected_network/component.yaml")
train_pytorch_model_from_csv_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/PyTorch/Train_PyTorch_model/from_CSV/component.yaml")
create_pytorch_model_archive_with_base_handler_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/PyTorch/Create_PyTorch_Model_Archive/with_base_handler/component.yaml")
upload_PyTorch_model_archive_to_Google_Cloud_Vertex_AI_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Upload_PyTorch_model_archive/component.yaml")
deploy_model_to_endpoint_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Deploy_to_endpoint/component.yaml")

# %% Pipeline definition
def train_tabular_classification_model_using_PyTorch_pipeline():
    dataset_gcs_uri = "gs://ml-pipeline-dataset/Chicago_taxi_trips/chicago_taxi_trips_2019-01-01_-_2019-02-01_limit=10000.csv"
    feature_columns = ["trip_seconds", "trip_miles", "pickup_community_area", "dropoff_community_area", "fare", "tolls", "extras"]  # Excluded "trip_total"
    label_column = "tips"
    # Deploying the model might incur additional costs over time
    deploy_model = False

    classification_label_column = "class"
    all_columns = [label_column] + feature_columns

    training_data = download_from_gcs_op(
        gcs_path=dataset_gcs_uri
    ).outputs["Data"]

    training_data = select_columns_using_Pandas_on_CSV_data_op(
        table=training_data,
        column_names=all_columns,
    ).outputs["transformed_table"]

    # Cleaning the NaN values.
    training_data = fill_all_missing_values_using_Pandas_on_CSV_data_op(
        table=training_data,
        replacement_value="0",
        #replacement_type_name="float",
    ).outputs["transformed_table"]

    classification_training_data = binarize_column_using_Pandas_on_CSV_data_op(
        table=training_data,
        column_name=label_column,
        predicate=" > 0",
        new_column_name=classification_label_column,
    ).outputs["transformed_table"]

    network = create_fully_connected_pytorch_network_op(
        input_size=len(feature_columns),
        # Optional:
        hidden_layer_sizes=[10],
        activation_name="elu",
        output_activation_name="sigmoid",
        # output_size=1,
    ).outputs["model"]

    model = train_pytorch_model_from_csv_op(
        model=network,
        training_data=classification_training_data,
        label_column_name=classification_label_column,
        loss_function_name="binary_cross_entropy",
        # Optional:
        #number_of_epochs=1,
        #learning_rate=0.1,
        #optimizer_name="Adadelta",
        #optimizer_parameters={},
        #batch_size=32,
        #batch_log_interval=100,
        #random_seed=0,
    ).outputs["trained_model"]

    model_archive = create_pytorch_model_archive_with_base_handler_op(
        model=model,
        # Optional:
        # model_name="model",
        # model_version="1.0",
    ).outputs["Model archive"]

    vertex_model_name = upload_PyTorch_model_archive_to_Google_Cloud_Vertex_AI_op(
        model_archive=model_archive,
    ).outputs["model_name"]

    # Deploying the model might incur additional costs over time
    if deploy_model:
        vertex_endpoint_name = deploy_model_to_endpoint_op(
            model_name=vertex_model_name,
        ).outputs["endpoint_name"]

pipeline_func=train_tabular_classification_model_using_PyTorch_pipeline

# %% Pipeline submission
if __name__ == '__main__':
    from google.cloud import aiplatform
    aiplatform.PipelineJob.from_pipeline_func(pipeline_func=pipeline_func).submit()

XGBoost

# python3 -m pip install "kfp<2.0.0" "google-cloud-aiplatform>=1.16.0" --upgrade --quiet
from kfp import components

# %% Loading components
download_from_gcs_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/storage/download/component.yaml")
select_columns_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Select_columns/in_CSV_format/component.yaml")
fill_all_missing_values_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Fill_all_missing_values/in_CSV_format/component.yaml")
binarize_column_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Binarize_column/in_CSV_format/component.yaml")
split_rows_into_subsets_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/dataset_manipulation/Split_rows_into_subsets/in_CSV/component.yaml")
train_XGBoost_model_on_CSV_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/XGBoost/Train/component.yaml")
xgboost_predict_on_CSV_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/XGBoost/Predict/component.yaml")
upload_XGBoost_model_to_Google_Cloud_Vertex_AI_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Upload_XGBoost_model/component.yaml")
deploy_model_to_endpoint_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Deploy_to_endpoint/component.yaml")

# %% Pipeline definition
def train_tabular_classification_model_using_XGBoost_pipeline():
    dataset_gcs_uri = "gs://ml-pipeline-dataset/Chicago_taxi_trips/chicago_taxi_trips_2019-01-01_-_2019-02-01_limit=10000.csv"
    feature_columns = ["trip_seconds", "trip_miles", "pickup_community_area", "dropoff_community_area", "fare", "tolls", "extras"]  # Excluded "trip_total"
    label_column = "tips"
    training_set_fraction = 0.8
    # Deploying the model might incur additional costs over time
    deploy_model = False

    classification_label_column = "class"
    all_columns = [label_column] + feature_columns

    dataset = download_from_gcs_op(
        gcs_path=dataset_gcs_uri
    ).outputs["Data"]

    dataset = select_columns_using_Pandas_on_CSV_data_op(
        table=dataset,
        column_names=all_columns,
    ).outputs["transformed_table"]

    dataset = fill_all_missing_values_using_Pandas_on_CSV_data_op(
        table=dataset,
        replacement_value="0",
        # # Optional:
        # column_names=None,  # =[...]
    ).outputs["transformed_table"]

    classification_dataset = binarize_column_using_Pandas_on_CSV_data_op(
        table=dataset,
        column_name=label_column,
        predicate="> 0",
        new_column_name=classification_label_column,
    ).outputs["transformed_table"]

    split_task = split_rows_into_subsets_op(
        table=classification_dataset,
        fraction_1=training_set_fraction,
    )
    classification_training_data = split_task.outputs["split_1"]
    classification_testing_data = split_task.outputs["split_2"]

    model = train_XGBoost_model_on_CSV_op(
        training_data=classification_training_data,
        label_column_name=classification_label_column,
        objective="binary:logistic",
        # Optional:
        #starting_model=None,
        #num_iterations=10,
        #booster_params={},
        #booster="gbtree",
        #learning_rate=0.3,
        #min_split_loss=0,
        #max_depth=6,
    ).outputs["model"]

    # Predicting on the testing data
    predictions = xgboost_predict_on_CSV_op(
        data=classification_testing_data,
        model=model,
        # label_column needs to be set when doing prediction on a dataset that has labels
        label_column_name=classification_label_column,
    ).outputs["predictions"]

    vertex_model_name = upload_XGBoost_model_to_Google_Cloud_Vertex_AI_op(
        model=model,
    ).outputs["model_name"]

    # Deploying the model might incur additional costs over time
    if deploy_model:
        vertex_endpoint_name = deploy_model_to_endpoint_op(
            model_name=vertex_model_name,
        ).outputs["endpoint_name"]

pipeline_func = train_tabular_classification_model_using_XGBoost_pipeline

# %% Pipeline submission
if __name__ == '__main__':
    from google.cloud import aiplatform
    aiplatform.PipelineJob.from_pipeline_func(pipeline_func=pipeline_func).submit()

scikit-learn

# python3 -m pip install "kfp<2.0.0" "google-cloud-aiplatform>=1.16.0" --upgrade --quiet
from kfp import components

# %% Loading components
download_from_gcs_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/storage/download/component.yaml")
select_columns_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Select_columns/in_CSV_format/component.yaml")
fill_all_missing_values_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Fill_all_missing_values/in_CSV_format/component.yaml")
binarize_column_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Binarize_column/in_CSV_format/component.yaml")
train_logistic_regression_model_using_scikit_learn_from_CSV_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/1f5cf6e06409b704064b2086c0a705e4e6b4fcde/community-content/pipeline_components/ML_frameworks/Scikit_learn/Train_logistic_regression_model/from_CSV/component.yaml")
upload_Scikit_learn_pickle_model_to_Google_Cloud_Vertex_AI_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Upload_Scikit-learn_pickle_model/component.yaml")
deploy_model_to_endpoint_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Deploy_to_endpoint/component.yaml")

# %% Pipeline definition
def train_tabular_classification_logistic_regression_model_using_Scikit_learn_pipeline():
    dataset_gcs_uri = "gs://ml-pipeline-dataset/Chicago_taxi_trips/chicago_taxi_trips_2019-01-01_-_2019-02-01_limit=10000.csv"
    feature_columns = ["trip_seconds", "trip_miles", "pickup_community_area", "dropoff_community_area", "fare", "tolls", "extras"]  # Excluded "trip_total"
    label_column = "tips"
    # Deploying the model might incur additional costs over time
    deploy_model = False

    classification_label_column = "class"
    all_columns = [label_column] + feature_columns

    training_data = download_from_gcs_op(
        gcs_path=dataset_gcs_uri
    ).outputs["Data"]

    training_data = select_columns_using_Pandas_on_CSV_data_op(
        table=training_data,
        column_names=all_columns,
    ).outputs["transformed_table"]

    # Cleaning the NaN values.
    training_data = fill_all_missing_values_using_Pandas_on_CSV_data_op(
        table=training_data,
        replacement_value="0",
        #replacement_type_name="float",
    ).outputs["transformed_table"]

    classification_training_data = binarize_column_using_Pandas_on_CSV_data_op(
        table=training_data,
        column_name=label_column,
        predicate="> 0",
        new_column_name=classification_label_column,
    ).outputs["transformed_table"]

    model = train_logistic_regression_model_using_scikit_learn_from_CSV_op(
        dataset=classification_training_data,
        label_column_name=classification_label_column,
        # Optional:
        #penalty="l2",
        #solver="lbfgs",
        #max_iterations=100,
        #multi_class_mode="auto",
        #random_seed=0,
    ).outputs["model"]

    vertex_model_name = upload_Scikit_learn_pickle_model_to_Google_Cloud_Vertex_AI_op(
        model=model,
    ).outputs["model_name"]

    # Deploying the model might incur additional costs over time
    if deploy_model:
        sklearn_vertex_endpoint_name = deploy_model_to_endpoint_op(
            model_name=vertex_model_name,
        ).outputs["endpoint_name"]

pipeline_func = train_tabular_classification_logistic_regression_model_using_Scikit_learn_pipeline

# %% Pipeline submission
if __name__ == '__main__':
    from google.cloud import aiplatform
    aiplatform.PipelineJob.from_pipeline_func(pipeline_func=pipeline_func).submit()

테이블 형식 회귀

TensorFlow

# python3 -m pip install "kfp<2.0.0" "google-cloud-aiplatform>=1.16.0" --upgrade --quiet
from kfp import components

# %% Loading components
download_from_gcs_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/storage/download/component.yaml")
select_columns_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Select_columns/in_CSV_format/component.yaml")
fill_all_missing_values_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Fill_all_missing_values/in_CSV_format/component.yaml")
split_rows_into_subsets_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/dataset_manipulation/Split_rows_into_subsets/in_CSV/component.yaml")
create_fully_connected_tensorflow_network_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/tensorflow/Create_fully_connected_network/component.yaml")
train_model_using_Keras_on_CSV_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/tensorflow/Train_model_using_Keras/on_CSV/component.yaml")
predict_with_TensorFlow_model_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/tensorflow/Predict/on_CSV/component.yaml")
upload_Tensorflow_model_to_Google_Cloud_Vertex_AI_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Upload_Tensorflow_model/component.yaml")
deploy_model_to_endpoint_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Deploy_to_endpoint/component.yaml")

# %% Pipeline definition
def train_tabular_regression_model_using_Tensorflow_pipeline():
    dataset_gcs_uri = "gs://ml-pipeline-dataset/Chicago_taxi_trips/chicago_taxi_trips_2019-01-01_-_2019-02-01_limit=10000.csv"
    feature_columns = ["trip_seconds", "trip_miles", "pickup_community_area", "dropoff_community_area", "fare", "tolls", "extras"]  # Excluded "trip_total"
    label_column = "tips"
    training_set_fraction = 0.8
    # Deploying the model might incur additional costs over time
    deploy_model = False

    all_columns = [label_column] + feature_columns

    dataset = download_from_gcs_op(
        gcs_path=dataset_gcs_uri
    ).outputs["Data"]

    dataset = select_columns_using_Pandas_on_CSV_data_op(
        table=dataset,
        column_names=all_columns,
    ).outputs["transformed_table"]

    dataset = fill_all_missing_values_using_Pandas_on_CSV_data_op(
        table=dataset,
        replacement_value="0",
        # # Optional:
        # column_names=None,  # =[...]
    ).outputs["transformed_table"]

    split_task = split_rows_into_subsets_op(
        table=dataset,
        fraction_1=training_set_fraction,
    )
    training_data = split_task.outputs["split_1"]
    testing_data = split_task.outputs["split_2"]

    network = create_fully_connected_tensorflow_network_op(
        input_size=len(feature_columns),
        # Optional:
        hidden_layer_sizes=[10],
        activation_name="elu",
        # output_activation_name=None,
        # output_size=1,
    ).outputs["model"]

    model = train_model_using_Keras_on_CSV_op(
        training_data=training_data,
        model=network,
        label_column_name=label_column,
        # Optional:
        #loss_function_name="mean_squared_error",
        number_of_epochs=10,
        #learning_rate=0.1,
        #optimizer_name="Adadelta",
        #optimizer_parameters={},
        #batch_size=32,
        metric_names=["mean_absolute_error"],
        #random_seed=0,
    ).outputs["trained_model"]

    predictions = predict_with_TensorFlow_model_on_CSV_data_op(
        dataset=testing_data,
        model=model,
        # label_column_name needs to be set when doing prediction on a dataset that has labels
        label_column_name=label_column,
        # Optional:
        # batch_size=1000,
    ).outputs["predictions"]

    vertex_model_name = upload_Tensorflow_model_to_Google_Cloud_Vertex_AI_op(
        model=model,
    ).outputs["model_name"]

    # Deploying the model might incur additional costs over time
    if deploy_model:
        vertex_endpoint_name = deploy_model_to_endpoint_op(
            model_name=vertex_model_name,
        ).outputs["endpoint_name"]

pipeline_func=train_tabular_regression_model_using_Tensorflow_pipeline

# %% Pipeline submission
if __name__ == '__main__':
    from google.cloud import aiplatform
    aiplatform.PipelineJob.from_pipeline_func(pipeline_func=pipeline_func).submit()

PyTorch

# python3 -m pip install "kfp<2.0.0" "google-cloud-aiplatform>=1.16.0" --upgrade --quiet
from kfp import components

# %% Loading components
download_from_gcs_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/storage/download/component.yaml")
select_columns_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Select_columns/in_CSV_format/component.yaml")
fill_all_missing_values_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Fill_all_missing_values/in_CSV_format/component.yaml")
create_fully_connected_pytorch_network_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/PyTorch/Create_fully_connected_network/component.yaml")
train_pytorch_model_from_csv_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/PyTorch/Train_PyTorch_model/from_CSV/component.yaml")
create_pytorch_model_archive_with_base_handler_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/PyTorch/Create_PyTorch_Model_Archive/with_base_handler/component.yaml")
upload_PyTorch_model_archive_to_Google_Cloud_Vertex_AI_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Upload_PyTorch_model_archive/component.yaml")
deploy_model_to_endpoint_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Deploy_to_endpoint/component.yaml")

# %% Pipeline definition
def train_tabular_regression_model_using_PyTorch_pipeline():
    dataset_gcs_uri = "gs://ml-pipeline-dataset/Chicago_taxi_trips/chicago_taxi_trips_2019-01-01_-_2019-02-01_limit=10000.csv"
    feature_columns = ["trip_seconds", "trip_miles", "pickup_community_area", "dropoff_community_area", "fare", "tolls", "extras"]  # Excluded "trip_total"
    label_column = "tips"
    all_columns = [label_column] + feature_columns
    # Deploying the model might incur additional costs over time
    deploy_model = False

    training_data = download_from_gcs_op(
        gcs_path=dataset_gcs_uri
    ).outputs["Data"]

    training_data = select_columns_using_Pandas_on_CSV_data_op(
        table=training_data,
        column_names=all_columns,
    ).outputs["transformed_table"]

    # Cleaning the NaN values.
    training_data = fill_all_missing_values_using_Pandas_on_CSV_data_op(
        table=training_data,
        replacement_value="0",
        #replacement_type_name="float",
    ).outputs["transformed_table"]

    network = create_fully_connected_pytorch_network_op(
        input_size=len(feature_columns),
        # Optional:
        hidden_layer_sizes=[10],
        activation_name="elu",
        # output_activation_name=None,
        # output_size=1,
    ).outputs["model"]

    model = train_pytorch_model_from_csv_op(
        model=network,
        training_data=training_data,
        label_column_name=label_column,
        # Optional:
        #loss_function_name="mse_loss",
        #number_of_epochs=1,
        #learning_rate=0.1,
        #optimizer_name="Adadelta",
        #optimizer_parameters={},
        #batch_size=32,
        #batch_log_interval=100,
        #random_seed=0,
    ).outputs["trained_model"]

    model_archive = create_pytorch_model_archive_with_base_handler_op(
        model=model,
        # Optional:
        # model_name="model",
        # model_version="1.0",
    ).outputs["Model archive"]

    vertex_model_name = upload_PyTorch_model_archive_to_Google_Cloud_Vertex_AI_op(
        model_archive=model_archive,
    ).outputs["model_name"]

    # Deploying the model might incur additional costs over time
    if deploy_model:
        vertex_endpoint_name = deploy_model_to_endpoint_op(
            model_name=vertex_model_name,
        ).outputs["endpoint_name"]

pipeline_func=train_tabular_regression_model_using_PyTorch_pipeline

# %% Pipeline submission
if __name__ == '__main__':
    from google.cloud import aiplatform
    aiplatform.PipelineJob.from_pipeline_func(pipeline_func=pipeline_func).submit()

XGBoost

# python3 -m pip install "kfp<2.0.0" "google-cloud-aiplatform>=1.16.0" --upgrade --quiet
from kfp import components

# %% Loading components
download_from_gcs_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/storage/download/component.yaml")
select_columns_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Select_columns/in_CSV_format/component.yaml")
fill_all_missing_values_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Fill_all_missing_values/in_CSV_format/component.yaml")
split_rows_into_subsets_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/dataset_manipulation/Split_rows_into_subsets/in_CSV/component.yaml")
train_XGBoost_model_on_CSV_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/XGBoost/Train/component.yaml")
xgboost_predict_on_CSV_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/XGBoost/Predict/component.yaml")
upload_XGBoost_model_to_Google_Cloud_Vertex_AI_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Upload_XGBoost_model/component.yaml")
deploy_model_to_endpoint_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Deploy_to_endpoint/component.yaml")

# %% Pipeline definition
def train_tabular_regression_model_using_XGBoost_pipeline():
    dataset_gcs_uri = "gs://ml-pipeline-dataset/Chicago_taxi_trips/chicago_taxi_trips_2019-01-01_-_2019-02-01_limit=10000.csv"
    feature_columns = ["trip_seconds", "trip_miles", "pickup_community_area", "dropoff_community_area", "fare", "tolls", "extras"]  # Excluded "trip_total"
    label_column = "tips"
    training_set_fraction = 0.8
    # Deploying the model might incur additional costs over time
    deploy_model = False

    all_columns = [label_column] + feature_columns

    dataset = download_from_gcs_op(
        gcs_path=dataset_gcs_uri
    ).outputs["Data"]

    dataset = select_columns_using_Pandas_on_CSV_data_op(
        table=dataset,
        column_names=all_columns,
    ).outputs["transformed_table"]

    dataset = fill_all_missing_values_using_Pandas_on_CSV_data_op(
        table=dataset,
        replacement_value="0",
        # # Optional:
        # column_names=None,  # =[...]
    ).outputs["transformed_table"]

    split_task = split_rows_into_subsets_op(
        table=dataset,
        fraction_1=training_set_fraction,
    )
    training_data = split_task.outputs["split_1"]
    testing_data = split_task.outputs["split_2"]

    model = train_XGBoost_model_on_CSV_op(
        training_data=training_data,
        label_column_name=label_column,
        # Optional:
        #starting_model=None,
        #num_iterations=10,
        #booster_params={},
        #objective="reg:squarederror",
        #booster="gbtree",
        #learning_rate=0.3,
        #min_split_loss=0,
        #max_depth=6,
    ).outputs["model"]

    # Predicting on the testing data
    predictions = xgboost_predict_on_CSV_op(
        data=testing_data,
        model=model,
        # label_column needs to be set when doing prediction on a dataset that has labels
        label_column_name=label_column,
    ).outputs["predictions"]

    vertex_model_name = upload_XGBoost_model_to_Google_Cloud_Vertex_AI_op(
        model=model,
    ).outputs["model_name"]

    # Deploying the model might incur additional costs over time
    if deploy_model:
        vertex_endpoint_name = deploy_model_to_endpoint_op(
            model_name=vertex_model_name,
        ).outputs["endpoint_name"]

pipeline_func = train_tabular_regression_model_using_XGBoost_pipeline

# %% Pipeline submission
if __name__ == '__main__':
    from google.cloud import aiplatform
    aiplatform.PipelineJob.from_pipeline_func(pipeline_func=pipeline_func).submit()

scikit-learn

# python3 -m pip install "kfp<2.0.0" "google-cloud-aiplatform>=1.16.0" --upgrade --quiet
from kfp import components

# %% Loading components
download_from_gcs_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/storage/download/component.yaml")
select_columns_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Select_columns/in_CSV_format/component.yaml")
fill_all_missing_values_using_Pandas_on_CSV_data_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/pandas/Fill_all_missing_values/in_CSV_format/component.yaml")
train_linear_regression_model_using_scikit_learn_from_CSV_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/1f5cf6e06409b704064b2086c0a705e4e6b4fcde/community-content/pipeline_components/ML_frameworks/Scikit_learn/Train_linear_regression_model/from_CSV/component.yaml")
upload_Scikit_learn_pickle_model_to_Google_Cloud_Vertex_AI_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Upload_Scikit-learn_pickle_model/component.yaml")
deploy_model_to_endpoint_op = components.load_component_from_url("http://raw.githubusercontent.com/GoogleCloudPlatform/vertex-ai-samples/399405402d95f4a011e2d2e967c96f8508ba5688/community-content/pipeline_components/google-cloud/Vertex_AI/Models/Deploy_to_endpoint/component.yaml")

# %% Pipeline definition
def train_tabular_regression_linear_model_using_Scikit_learn_pipeline():
    dataset_gcs_uri = "gs://ml-pipeline-dataset/Chicago_taxi_trips/chicago_taxi_trips_2019-01-01_-_2019-02-01_limit=10000.csv"
    feature_columns = ["trip_seconds", "trip_miles", "pickup_community_area", "dropoff_community_area", "fare", "tolls", "extras"]  # Excluded "trip_total"
    label_column = "tips"
    all_columns = [label_column] + feature_columns
    # Deploying the model might incur additional costs over time
    deploy_model = False

    training_data = download_from_gcs_op(
        gcs_path=dataset_gcs_uri
    ).outputs["Data"]

    training_data = select_columns_using_Pandas_on_CSV_data_op(
        table=training_data,
        column_names=all_columns,
    ).outputs["transformed_table"]

    # Cleaning the NaN values.
    training_data = fill_all_missing_values_using_Pandas_on_CSV_data_op(
        table=training_data,
        replacement_value="0",
        #replacement_type_name="float",
    ).outputs["transformed_table"]

    model = train_linear_regression_model_using_scikit_learn_from_CSV_op(
        dataset=training_data,
        label_column_name=label_column,
    ).outputs["model"]

    vertex_model_name = upload_Scikit_learn_pickle_model_to_Google_Cloud_Vertex_AI_op(
        model=model,
    ).outputs["model_name"]

    # Deploying the model might incur additional costs over time
    if deploy_model:
        sklearn_vertex_endpoint_name = deploy_model_to_endpoint_op(
            model_name=vertex_model_name,
        ).outputs["endpoint_name"]

pipeline_func = train_tabular_regression_linear_model_using_Scikit_learn_pipeline

# %% Pipeline submission
if __name__ == '__main__':
    from google.cloud import aiplatform
    aiplatform.PipelineJob.from_pipeline_func(pipeline_func=pipeline_func).submit()

제공되는 코드 샘플에 대해서는 다음 사항에 유의하세요.

  • Kubeflow 파이프라인은 Python 함수로 정의됩니다.
  • 파이프라인의 워크플로 단계는 Kubeflow 파이프라인 구성요소를 사용하여 생성됩니다. 구성요소 출력을 다른 구성요소의 입력으로 사용하여 파이프라인의 워크플로를 그래프로 정의합니다. 예를 들어 fill_all_missing_values_using_Pandas_on_CSV_data_op 구성요소 태스크는 select_columns_using_Pandas_on_CSV_data_op 구성요소 태스크의 transformed_table 출력에 따라 달라집니다.
  • Python용 Vertex AI SDK를 사용하여 Vertex AI Pipelines에서 파이프라인 실행을 만듭니다.

파이프라인 모니터링

Google Cloud console의 Vertex AI 섹션에서 파이프라인 페이지로 이동하여 실행 탭을 엽니다.

파이프라인 실행으로 이동

다음 단계