StatMike
StatMike
  • 12
  • 143 759
End-to-End: Pipeline Orchestration (KFP) - BigQuery (BQML) Model For Endpoint Update [notebook 03C]
In [notebook 03a] we trained a model using BigQuery ML (BQML). In [notebook 03b] we used Vertex AI to upload the BQML model and deploy it to a live endpoint for online predictions. In this [notebook 03c] we will build a Kubeflow Pipeline (KFP) to conditionally replace the model on the online endpoint with a better model. Vertex AI Pipelines is a service that runs Kubeflow pipelines as a managed service where we don’t have to worry about infrastructure. This makes it incredibly straight-forward to build a pipelines to orchestrate all the steps of a challenger model scenario:
➡️ Train a Challenger Model
➡️ Retrieve evaluation metrics for the Challenger Model
➡️ Retrieve evaluation metrics for the current model
➡️ Compare the models
➡️ Evaluate if the Challenger model is better than the current model
➡️ If better, update the online endpoint
➡️➡️ Deploy the challenger model
➡️➡️ Move all traffic to the challenger model
➡️➡️ Undeploy the previous model
GitHub Repository: github.com/statmike/vertex-ai-mlops
The Notebook followed in this video is an older version - link for the version in the video: github.com/statmike/vertex-ai-mlops/blob/fd442b458c710a0a7afdc41bae690d2a3282e93c/03c%20-%20BQML%20%2B%20Vertex%20AI%20%3E%20Pipelines%20-%20automated%20pipelines%20for%20updating%20models.ipynb
An updated version of the notebook can be found here: github.com/statmike/vertex-ai-mlops/blob/main/03%20-%20BigQuery%20ML%20(BQML)/03Tools%20-%20Pipelines%20Example%202.ipynb
Timeline:
0:00 - Introduction
0:44 - Overview
2:20 - Start Walkthrough
11:18 - [Notebook Section] Setup
13:30 - [Notebook Section] Custom Components
23:07 - [Notebook Section] Pipeline Definition
27:55 - [Notebook Section] Compile Pipeline
29:30 - [Notebook Section] Create Vertex AI Pipeline Job
30:08 - Review Pipeline Run with Console
49:50 - Re-run Pipeline (Endpoint Update)
58:20 - Q&A: When would I want to run a pipeline like this?
1:03:35 - [Notebook Section] Prediction
1:10:50 - Wrap-up
Переглядів: 3 304

Відео

End-to-End: ML with TensorFlow in Jupyter with Tensorflow I/O BigQuery Reader [notebook 05]1:15:11End-to-End: ML with TensorFlow in Jupyter with Tensorflow I/O BigQuery Reader [notebook 05]
End-to-End: ML with TensorFlow in Jupyter with Tensorflow I/O BigQuery Reader [notebook 05]
Переглядів 3,9 тис.2 роки тому
An end-to-end workflow using a Jupyter Notebook hosted by Vertex AI Workbench to train an ML model with TensorFlow within the notebook. Training data is read using the TensorFlow I/O reader for BigQuery. Many deep explanations along the way including using Tensorboard to evaluate the model training. The final model is deployed to a Vertex AI Endpoint and online predictions are demonstrated usin...
Part 2 - End-To-End: Pipeline Orchestration (KFP) - AutoML in Vertex AI for ML Ops [notebook 02c]10:28Part 2 - End-To-End: Pipeline Orchestration (KFP) - AutoML in Vertex AI for ML Ops [notebook 02c]
Part 2 - End-To-End: Pipeline Orchestration (KFP) - AutoML in Vertex AI for ML Ops [notebook 02c]
Переглядів 2 тис.2 роки тому
Today we revisit a previous video and add a section for model evaluation using the Vertex AI API. See how easy it is to retrieve a vast array of evaluation metrics for AutoML models. The original video: ua-cam.com/video/1gHJgY7AXAs/v-deo.html An end-to-end workflow using Pipelines within Vertex AI on Google Cloud Platform. We will use AutoML to train a machine learning model. A walkthrough of b...
Part 2 - End-To-End: Interactive Code (Python) - AutoML in Vertex AI for ML Ops [notebook 02b]9:26Part 2 - End-To-End: Interactive Code (Python) - AutoML in Vertex AI for ML Ops [notebook 02b]
Part 2 - End-To-End: Interactive Code (Python) - AutoML in Vertex AI for ML Ops [notebook 02b]
Переглядів 2,6 тис.2 роки тому
Today we revisit a previous video and add a section for model evaluation using the Vertex AI API. See how easy it is to retrieve a vast array of evaluation metrics for AutoML models. The original video: ua-cam.com/video/GOxHYfCLc6U/v-deo.html An end-to-end workflow using Python clients for Vertex AI on Google Cloud Platform. We will use AutoML to train a machine learning model. A walkthrough of...
ML with SQL in BigQuery to Online Predictions in Vertex AI for ML Operations [notebook 03b]32:07ML with SQL in BigQuery to Online Predictions in Vertex AI for ML Operations [notebook 03b]
ML with SQL in BigQuery to Online Predictions in Vertex AI for ML Operations [notebook 03b]
Переглядів 2,8 тис.2 роки тому
An end-to-end workflow using the Python clients for Vertex AI on Google Cloud Platform. We export a model created with BigQuery ML and use it for online predictions in Vertex AI. This video follows the notebook 03b - Vertex AI BigQuery Machine Learning (BQML) - Online Predictions with BQML Models. The Notebook followed in this video is an older version - link for the version in the video: githu...
End-To-End: ML with SQL in BigQuery (BQML) [notebook 03a]54:07End-To-End: ML with SQL in BigQuery (BQML) [notebook 03a]
End-To-End: ML with SQL in BigQuery (BQML) [notebook 03a]
Переглядів 6 тис.2 роки тому
An end-to-end workflow using the Python client for BigQuery on Google Cloud Platform. We use BigQuery ML to train a model using SQL! A walkthrough of all the steps from connecting to data sources, training a model, evaluating the final model, and requesting predictions from multiple clients. A few deep dives along the way including model explainability! This video follows the notebook 03a - Big...
End-To-End: Pipeline Orchestration (KFP) - AutoML in Vertex AI for ML Operations [notebook 02c]57:49End-To-End: Pipeline Orchestration (KFP) - AutoML in Vertex AI for ML Operations [notebook 02c]
End-To-End: Pipeline Orchestration (KFP) - AutoML in Vertex AI for ML Operations [notebook 02c]
Переглядів 10 тис.2 роки тому
An end-to-end workflow using Pipelines within Vertex AI on Google Cloud Platform. We will use AutoML to train a machine learning model. A walkthrough of building a repeatable pipeline to orchestrate all the steps from connecting to data sources, training a model, evaluating the final model, deploying to an online endpoint and requesting predictions from multiple clients. A few deep dives along ...
End-To-End: Interactive Code (Python) - AutoML in Vertex AI for ML Operations [notebook 02b]53:33End-To-End: Interactive Code (Python) - AutoML in Vertex AI for ML Operations [notebook 02b]
End-To-End: Interactive Code (Python) - AutoML in Vertex AI for ML Operations [notebook 02b]
Переглядів 8 тис.2 роки тому
An end-to-end workflow using Python clients for Vertex AI on Google Cloud Platform. We will use AutoML to train a machine learning model. A walkthrough of all the steps from connecting to data sources, training a model, evaluating the final model, deploying to an online endpoint and requesting predictions from multiple clients. A few deep dives along the way including model explainability! This...
End-To-End: No Code - AutoML in Vertex AI for ML Operations [notebook 02a]1:00:14End-To-End: No Code - AutoML in Vertex AI for ML Operations [notebook 02a]
End-To-End: No Code - AutoML in Vertex AI for ML Operations [notebook 02a]
Переглядів 27 тис.2 роки тому
An end-to-end workflow completely within the Vertex AI interface in the Google Cloud Console. We will use AutoML to train a machine learning model. A walkthrough of all the steps from connecting to data sources, training a model, evaluating the final model, deploying to an online endpoint and requesting predictions from multiple clients. A few deep dives along the way! This video follows the no...
Introduction - Vertex AI for ML Operations14:48Introduction - Vertex AI for ML Operations
Introduction - Vertex AI for ML Operations
Переглядів 42 тис.2 роки тому
Introduction to the playlist of end-to-end workflow walkthroughs for machine learning operations using Google Cloud Platform’s Vertex AI. GitHub Repository: github.com/statmike/vertex-ai-mlops Timeline: 0:00 - Introduction 3:25 - The GitHub Repository 4:40 - Walkthrough List of Workflows 6:40 - Q&A - What is not covered 7:30 - Q&A - Are all the videos needed? 8:12 - Q&A - How do I learn ML? 12:...
Environment Setup - Vertex AI for ML Operations [notebook 00]24:22Environment Setup - Vertex AI for ML Operations [notebook 00]
Environment Setup - Vertex AI for ML Operations [notebook 00]
Переглядів 24 тис.2 роки тому
A walkthrough of creating a Google Cloud Platform project and setting up the environment for this series of end-to-end workflows. This video follows the readme and the first notebook [00 - Environment Setup] in the repository. GitHub Repository: github.com/statmike/vertex-ai-mlops The Notebook followed in this video: github.com/statmike/vertex-ai-mlops/blob/main/00 - Setup/00 - Environment Setu...
Data Source - Vertex AI for ML Operations [notebook 01]37:11Data Source - Vertex AI for ML Operations [notebook 01]
Data Source - Vertex AI for ML Operations [notebook 01]
Переглядів 13 тис.2 роки тому
A walkthrough of creating the data source for this project using BigQuery. We will import, review, and prepare the data for use in machine learning workflows. This video follows the notebook [01 - BigQuery - Table Data Source] in the repository. GitHub Repository: github.com/statmike/vertex-ai-mlops The Notebook followed in this video: github.com/statmike/vertex-ai-mlops/blob/main/01 - Data Sou...