Vertex AI
Vertex AI is a kind of platform of Google Cloud Platform(GCP). It supports following functions.
- Upload dataset.
- Give a label.
- Store frequently used features in feature store and use while training.
- Automatic hyperparemeter tuning.
- Searching model architecture(AutoML).
- Train model in GPU, TPU server.
- Manage trained models.
- Batch prediction.
There are also many additional functions.
Follow the following steps when using Vertex AI to distribute model and make a prediction.
- Authorization.
- Make GCS bucket to save SavedModel. Set project id, bucket name, and location(region).
- Upload directory to a new bucket. In GCS, file is called ‘blob’. The blob is stored withut directory, so use ‘/’ to make it look like the directory exists.
- Use multithreading to speed up.
- Inform Vertex AI about the model. Use ‘google-cloud-aiplatform’ library. Make new Vertex AI model by assigning name, GCS path, and container.
- Make endpoint(a point that client application connects when accessing to a service) to distribute model.
- Make a prediction. Transform image to python list first.
- Remove endpoint to prevent additional cost.
If you need to do a number of predictions, request Vertex AI to do predictions instead of repeatedly calling prediction service.
For specific implementation, see Go for Codes.
All images, except those with separate source indications, are excerpted from lecture materials.
댓글남기기