Introduction to Google Cloud Platform VertexAI AutoML
What is Vertex AI?
Vertex AI is Google’s latest AI service that allows all levels of data scientists to build high-accuracy models. Vertex AI is a one-stop service for building AI solutions. In Vertex AI, a data scientist can Ingest data → Analyze Imported Datasets → Transform and Select Features → Model Training → Model Evaluation → Deployment → Submit Prediction Requests. In terms of AI solutions, Vertex AI provides model training in Tabular Data, Image Data, Video Data, and Text Data. Here’s a summary of what Vertex AI is capable of…
AutoML or Custom Training
After preparing your datasets, you are prompted with three model training options. These include, AutoML, AutoML Edge, and Custom Model.
For both AutoML and AutoML Edge, Vertex AI will proactively seek the best model for you. The only thing that you would have to provide is the budget hour. The budget hour is a parameter that tells Vertex AI to search different models and parameters within the given hour limit. For example, a budget hour of 1 means to search the best model within an hour and tear down all resources right after.
On the other hand, training custom models can be done in two ways. Your first option is to create Jupyter Notebooks (Workbench Tab) in the Vertex AI console and start writing code that imports data from Google Cloud Storage and outputs models to a designated location in Google Cloud Storage. The other option is to make your training code into a Docker Image and push that Docker Image into Google’s Container Registry.
Use AutoML and AutoML Edge when your
1. Input data is simple (one data type representation)
2. Knowledge of building model is limited
3. Goal is to come up with a prototype or proof of concept (Test if your data with a model can provide insightful results)
Use Custom Model Training when your
1. Input data is complex (one or more data type representations)
2. Model Architecture is complex (GANs or Transformers)
3. Hardware requires a specific model type or need a specific model library (TensorFlow or Pytorch)
4. Goal is to improve existing models
5. Project needs more control over your model’s training process
Additional Resources For Learning Vertex AI
To facilitate your journey in learning Vertex AI, here are some of the online resources that I find helpful and informative. Additionally, I have included my repository where you can find some sample notebooks that I used for requesting predictions from Vertex AI.
Google Cloud Tech (YouTube Channel)
This is an official website where you can find lots of introductions to Vertex AI’s functionalities and concept explanations. I think this should be the first stop before jumping into documentation.
All Vertex AI code Samples (Official Documentation)
This is where you find the components and sample codes for you to build models, request predictions, and cancel training jobs. This should be your reference site when you forget how to do a certain step or want to accomplish a certain task.
Pre-built Docker Container Image (Official Documentation)
Finally, here is a list of containers that are provided by Google. If you are training your model and need a docker image, this is where you find the resource.
I hope anyone who happens to read this article can try out Vertex AI. I believe that you will be impressed by Google’s AI service and start your Cloud Computing Journey using the Google Cloud Platform.