Digital Product School AI Challenge

First make an account on Google Cloud Platform https://cloud.google.com/. Then navigate to console tab on the upper right corner.

Initiate

We start by creating a new project

Now we open the Command line Interface and Enable our GCloud apis.

gcloud services enable compute.googleapis.com         \
containerregistry.googleapis.com  \
aiplatform.googleapis.com

Now we create our bucket for this project.

BUCKET_NAME=gs://$GOOGLE_CLOUD_PROJECT-bucket
gsutil mb -l us-central1 $BUCKET_NAME

here is my bucket name.

Training

Now we create a directory mpg/ now in this directory we make the Dockerfile mpg/Dockerfile and one more directory mpg/trainer with a file mpg/trainer/train.py.

We would need replace the bucket name in the training file

sed -i "s|BUCKET_NAME|$BUCKET_NAME|g" trainer/train.py

Once you have made the appropriate files with the appropriate content we move on to making a container image. Set the env variable

IMAGE_URI="gcr.io/$GOOGLE_CLOUD_PROJECT/mpg:v1" 

Run the build command

docker build ./ -t $IMAGE_URI

Now push the Image to Registry

docker push $IMAGE_URI

Here you can see the image in the container registry.

Lets train the model for our problem using Vertex AI Training.

Now once the training is done lets deploy the model. First we install pip3 install google-cloud-aiplatform --upgrade --user.

Deployment

• make sure you are in the mpg directory

Create a new file mpg/deploy.py and then execute

python3 deploy.py | tee deploy-output.txt 

First it will import the model.

Then it will create an end point of the model.

Now we can observe the endpoint has been created successfully.

Testing

For testing the api create a mpg/predct.py. replace the end point name with your endpoint name, you can get it from

ENDPOINT=$(cat deploy-output.txt | sed -nre 's:.*Resource name\: (.*):\1:p' | tail -1)
sed -i "s|ENDPOINT_STRING|$ENDPOINT|g" predict.py

Here are my testing outputs.

python3 predict.py