Skip to content

Latest commit

 

History

History

tfx-caip-tf23

README.md

Continuous training with TFX and Cloud AI Platform

This series of hands on labs guides you through the process of implementing a TensorFlow Extended (TFX) continuous training pipeline that automates training and deployment of a TensorFlow 2.3 model.

The below diagram represents the workflow orchestrated by the pipeline.

TFX_CAIP.

  1. Training data in the CSV format is ingested from a GCS location using CsvExampleGen. The URI to the data root is passed as a runtime parameter. The CsvExampleGen component splits the source data into training and evaluation splits and converts the data into the TFRecords format.
  2. The StatisticsGen component generates statistics for both splits.
  3. The SchemaGen component autogenerates a schema . This is done for data validation and anomaly detection. The pipeline uses a curated schema imported by the ImportedNode component.
  4. The ImporterNode component is used to bring the curated schema file into the pipeline. The location of the schema file is passed as a runtime parameter.
  5. The ExampleValidator component validates the generated examples against the imported schema
  6. The Transform component preprocess the data to the format required by the Trainer component. It also saves the preprocessing TensorFlow graph for consistent feature engineering at training and serving time.
  7. The Trainer starts an AI Platform Training job. The AI Platform Training job is configured for training in a custom container.
  8. The Tuner component in the example pipeline tunes model hyperparameters using CloudTuner (KerasTuner instance) and AI Platform Vizier as a back-end. It can added and removed from the pipeline using the enable_tuning environment variable set in the notebook or in the pipeline code. When included in the pipeline, it ouputs a "best_hyperparameter" artifact directly into the Trainer. When excluded hyperparameters are drawn from the defaults set in the pipeline code.
  9. The ResolverNode component retrieves the best performing model from the previous runs and passed it to the Evaluator to be used as a baseline during model validation.
  10. The Evaluator component evaluates the trained model against the eval split and validates against the baseline model from the ResolverNode. If the new model exceeds validation thresholds it is marked as "blessed".
  11. The InfraValidator component validates the model serving infrastructure and provides a "infra_blessing" that the model can be loaded and queried for predictions.
  12. If the new model is blessed by the Evaluator and InfraValidator, the Pusher deploys the model to AI Platform Prediction.

The ML model utilized in the labs is a multi-class classifier that predicts the type of forest cover from cartographic data. The model is trained on the Covertype Data Set dataset.

Preparing the lab environment

You will use the lab environment configured as on the below diagram:

Lab env

The core services in the environment are:

  • ML experimentation and development - AI Platform Notebooks
  • Scalable, serverless model training - AI Platform Training
  • Parallelized and distributed model hyperparameter tuning - AI Platform Vizier
  • Scalable, serverless model serving - AI Platform Prediction
  • Machine learning pipelines - AI Platform Pipelines
  • Distributed data processing - Cloud Dataflow
  • Artifact store - Google Cloud Storage
  • CI/CD tooling - Cloud Build

In this environment, all services are provisioned in the same Google Cloud Project.

Enabling Cloud Services

To enable Cloud Services utilized in the lab environment:

  1. Launch Cloud Shell
  2. Set your project ID
PROJECT_ID=[YOUR PROJECT ID]

gcloud config set project $PROJECT_ID
  1. Use gcloud to enable the services
gcloud services enable \
cloudbuild.googleapis.com \
container.googleapis.com \
cloudresourcemanager.googleapis.com \
iam.googleapis.com \
containerregistry.googleapis.com \
containeranalysis.googleapis.com \
ml.googleapis.com \
dataflow.googleapis.com
  1. The Cloud Build service account needs the Editor permissions in your GCP project to upload the pipeline package to an AI Platform Pipelines instance.
PROJECT_NUMBER=$(gcloud projects describe $PROJECT_ID --format="value(projectNumber)")
CLOUD_BUILD_SERVICE_ACCOUNT="${PROJECT_NUMBER}@cloudbuild.gserviceaccount.com"
gcloud projects add-iam-policy-binding $PROJECT_ID \
  --member serviceAccount:$CLOUD_BUILD_SERVICE_ACCOUNT \
  --role roles/editor
  1. Create a custom service account to give CAIP training job access to AI Platform Vizier service for pipeline hyperparameter tuning.
SERVICE_ACCOUNT_ID=tfx-tuner-caip-service-account
gcloud iam service-accounts create $SERVICE_ACCOUNT_ID  \
    --description="A custom service account for CAIP training job to access AI Platform Vizier service for pipeline hyperparameter tuning" \
    --display-name="TFX Tuner CAIP Vizier"
  1. Grant your custom service account access to the ml.admin role to create and manage AI Platform Vizier studies.
gcloud projects add-iam-policy-binding $PROJECT_ID \
    --member=serviceAccount:${SERVICE_ACCOUNT_ID}@${PROJECT_ID}.iam.gserviceaccount.com \
    --role=roles/ml.admin
  1. Grant your custom service account access to the storage.objectAdmin role for artifact access and temporary tuning file storage.
gcloud projects add-iam-policy-binding $PROJECT_ID \
    --member=serviceAccount:${SERVICE_ACCOUNT_ID}@${PROJECT_ID}.iam.gserviceaccount.com \
    --role=roles/storage.objectAdmin
  1. Grant your project's AI Platform Google-managed service account the iam.serviceAccountAdmin role for your new custom service account. To do so, use the gcloud tool to run the following command:
gcloud iam service-accounts add-iam-policy-binding \
  --role=roles/iam.serviceAccountAdmin \
  --member=serviceAccount:service-${PROJECT_NUMBER}@cloud-ml.google.com.iam.gserviceaccount.com \
  ${SERVICE_ACCOUNT_ID}@${PROJECT_ID}.iam.gserviceaccount.com

Creating an instance of AI Platform Pipelines

The core component of the lab environment is AI Platform Pipelines. To create an instance of AI Platform Pipelines follow the Setting up AI Platform Pipelines how-to guide. Make sure to enable the access to https://www.googleapis.com/auth/cloud-platform when creating a GKE cluster.

Creating an instance of AI Platform Notebooks

An instance of AI Platform Notebooks is used as a primary experimentation/development workbench.

To provision the instance follow the Create an new notebook instance setup guide. Use the TensorFlow Enterprise 2.3 no-GPU image. Leave all other settings at their default values.

After the instance is created, you can connect to JupyterLab IDE by clicking the OPEN JUPYTERLAB link in the AI Platform Notebooks Console.

In the JupyterLab, open a terminal and clone this repository in the home folder.

cd
git clone https://github.com/GoogleCloudPlatform/mlops-on-gcp.git

From the mlops-labs/workshops/tfx-caip-tf23 folder execute the install.sh script to install TFX and KFP SDKs.

cd mlops-on-gcp/workshops/tfx-caip-tf23
./install.sh

Summary of lab exercises

Lab-01 - TFX Components walk-through

In this lab, you will walk through the configuration and execution of the key TFX components. The primary goal of the lab is to get a high level understanding of the function and usage of each of the components. You will work in an AI Platform Notebooks instance using the components in an interactive mode.

Lab-02 - Orchestrating model training and deployment with TFX and Cloud AI Platform

In this lab you will develop, deploy and run a TFX pipeline that runs on Google Cloud.

Lab-03 - CI/CD for a TFX pipeline

In this lab you will walk through authoring of a Cloud Build CI/CD workflow that automatically builds and deploys a TFX pipeline. You will also integrate your workflow with GitHub by setting up a trigger that starts the workflow when a new tag is applied to the GitHub repo hosting the pipeline's code.

Lab-04 - ML Metadata

In this lab, you will explore ML metadata and ML artifacts created by TFX pipeline runs.