Image creation – Vertex AI Workbench and Custom Model Training-1

Now that we have completed the creation of Dockerfile and Python code for the model building. We can proceed with image creation.

Step 1: Image creation using docker build

Type the following commands as shown in Figure 4.19 for image creation:

IMAGE_URL=”gcr.io/$PROJECT_ID/price:v1”

docker build ./ -t $IMAGE_URL

Figure 4.19: Image creation process

1. Type the commands in the Terminal.

It will take a few minutes to build the image from the Dockerfile. Once the build is complete, success message will be displayed in the terminal as shown in Figure 4.20:

Figure 4.20: Image creation complete

1. Success message after completion of image creation.

Step 2: Checking the docker image

After creation of the image, let us check if we can run the python code from the image created. Type the command given below as shown in Figure 4.21:

docker run $IMAGE_URL

Figure 4.21: Test run of python code on container

1. Type the command.

Python code has run successfully as shown in the following figure:

Figure 4.22: Successful run of code

Pushing image to container registry

Now that we have created the image and tested if the code runs with proper output, we can now push the image to the container registry.

Step 1: Image push command

Type the following command in the Terminal as shown in Figure 4.23:

docker push $IMAGE_URL

Figure 4.23: Image to project container registry

1. Type the command.
2. Image successfully pushed to the container registry.

Step 2: Check container registry

Open the container registry from the navigation menu of the platform (it will be listed under CI/CD). Alternatively, type container registry in the search box of the platform to open it. Landing page of container registry is as shown in Figure 4.24:

Figure 4.24: Created price folder in container registry

1. A folder called price will be created in the container registry (Refer to the image creation where we have specified price to be the name and v1 to be the tag). Click on price.

Figure 4.25: Created image pushed to container registry

1. Image will be available inside the folder and under tag v1.

Submitting the custom model training job

For the model to get trained, it picks the data from the .csv which is uploaded on the cloud storage bucket. By default, workbench resources will have a read and write access to the cloud storage. However, when we submit the job for training, training code will execute on one or more Google Cloud-managed VM instances. These VM will not have access to the cloud storage buckets and hence we will use the service account. Follow these steps to submit the training job:

Step 1: Training module of vertex AI

Open the training module of the vertex AI as shown in Figure 4.26:

Figure 4.26: Training module

1. Select training module.
2. Select custom jobs.
3. Click CREATE.