GSP-354 : Building and Deploying Machine Learning Solutions with Vertex AI

GSP-354 : Building and Deploying Machine Learning Solutions with Vertex AI

Overview

Task - 3: Import dataset

TODO: fill in PROJECT_ID

1PROJECT_ID = "<PROJECT_ID>"

TODO: Create a globally unique Google Cloud Storage bucket for artifact storage.

1GCS_BUCKET = f"gs://<PROJECT_ID>-vertex-challenge-lab"

Task - 4: Build and train model

TODO: Add a hub.KerasLayer for BERT text preprocessing using the hparams dict.

Name the layer 'preprocessing' and store in the variable preprocessor.

1preprocessor = hub.KerasLayer(hparams['tfhub-bert-preprocessor'],name='preprocessing')

TODO: Add a trainable hub.KerasLayer for BERT text encoding using the hparams dict.

Name the layer 'BERT_encoder' and store in the variable encoder.

1encoder = hub.KerasLayer(hparams['tfhub-bert-encoder'], trainable=True, name='BERT_encoder')

TODO: Save your BERT sentiment classifier locally.

Hint: Save it to './bert-sentiment-classifier-local'. Note the key name in model.save().

1"model-dir": "./bert-sentiment-classifier-local"

Task 5: Create artifact registry for custom container images

TODO: create a Docker Artifact Registry using the gcloud CLI. Note the required respository-format and location flags.

1!gcloud artifacts repositories create {ARTIFACT_REGISTRY} \
2--repository-format=docker \
3--location={REGION} \
4--description="Artifact registry for ML custom training images for sentiment classification"

TODO: use Cloud Build to build and submit your custom model container to your Artifact Registry.

Hint: make sure the config flag is pointed at {MODEL_DIR}/cloudbuild.yaml defined above and you include your model directory.

1!gcloud builds submit {MODEL_DIR} --timeout=20m --config {MODEL_DIR}/cloudbuild.yaml

Task 6: Define a pipeline using the KFP V2 SDK

TODO: change this to your name.

1USER = "qwiklabsdemo"

TODO: fill in the remaining arguments from the pipeline constructor.

1display_name=display_name,
2    container_uri=container_uri,
3    model_serving_container_image_uri=model_serving_container_image_uri,
4    base_output_dir=GCS_BASE_OUTPUT_DIR,

TODO: Generate online predictions using your Vertex Endpoint.

1endpoint = vertexai.Endpoint(
2endpoint_name=ENDPOINT_NAME,
3project=PROJECT_ID,
4location=REGION
5)

TODO: write a movie review to test your model e.g. "The Dark Knight is the best Batman movie!"

1test_review = "The Dark Knight is the best Batman movie!"

TODO: use your Endpoint to return prediction for your test_review.

1prediction = endpoint.predict([test_review])

Congratulations, you're all done with the lab 😄