paul-octopus-python

You can use this Python code to construct your solution to predict the results of the 2022 World Cup.

This is a Python webapp, and you will need to deploy it in a cloud provider. In this example we will use the Azure Container Instances to run the webapp and generate the predictions.

Requirements

• Docker
• Python

Azure Container Registry (ACR)

In this example we will use the Azure Container Registry to push our Docker image containing the webapp.

The provided Azure environment already have an ACR instance named pauloctopus. You can use it, or you can create your own instance of ACR.

To start using the ACR to push images from your local environment, you will first need to authenticate.

docker login pauloctopus.azurecr.io

The username and password to authenticate can be found in the Azure Portal. Search for the Azure Container Registry, go into the pauloctopus instance, and click in the Access keys blade. There you will find the username and password to authenticate.

Configuring your development environment

This solution needs to access an Azure Storage Account to download the csv files with data. You will need to provide an environment variable with the Storage Account connection string.

You can find the connection string in the Azure Portal, by accessing the pauloctopus2022 Storage Account and clicking on the Access keys blade.

Show and copy one of the connection strings, and set the following environment variable inside your Python IDE.

AZURE_STORAGE_CONNECTION_STRING=DefaultEndpointsProtocol=https;AccountName=pauloctopus2022;AccountKey=xxxxxxx;EndpointSuffix=core.windows.net

Implementing your code to predict matches

Feel free to change or adapt anything in this solution.

If you just want to write a function to predict a match, based on the teams that are playing, you just need to create a new python file inside predictor folder, and extend the AbstractPredictor class.

One dummy predictor called OneZeroPredictor is included as an example.

Remember that you will need to use at lease one of the dataset files provided inside the Storage Account.

Building and pushing your docker image

When you have your predictor implemented, you can build the docker image using the provided Dockerfile.

Include your login in the image, to avoid conflicts with others that are sharing the same environment.

docker build -t pauloctopus.azurecr.io/<login>-predictor .
docker push pauloctopus.azurecr.io/<login>-predictor

This will push your predictor WebApp to ACR, and it will be available to deploy using the Azure Container Instances.

Deploy in Azure Container Instances (ACI)

Azure Container Instances is an easy way to deploy your WebApp in the Azure platform. You can use the Azure Portal, search for Azure Container Instances, and create a new app there.

Basics

On the first step, you will need to fill the container name with your app name (remember to include your login to avoid conflict). Select the option to use the image from ACR, and then select the image that you just pushed on the previous step.

Networking

On the networking tab, you can choose a part of the DNS name to access your webapp after the deployment. In this example we are using "gbim-predictor" as part of the DNS.

IMPORTANT: You will also need to change the exposed port to be 8080. This is the port that we are serving the app.

Advanced

On the advanced tab, you will need to configure the AZURE_STORAGE_CONNECTION_STRING environment variable. This is the same variable that you configured for your development environment. This is the connection string that will allow your app to download csv files from the storage account.

Review and create

After configuring the container instance, you will review and create it.

Wait for the creation of the Container Instance. You can explore the portal, check the logs and make sure that the app is running.

Accessing your App and generating the Predictions

After the deployment of your app, you can check the Overview tab and get the full DNS where it is available.

You can now access the app from your browser (don't forget to put the 8080 port at the URL).

Running a predictor class

The way that this solution is implemented, you can pass the name of your predictor in the URL like the following example:

http://gbim-predictor.caeja3fbbzdfb0gz.brazilsouth.azurecontainer.io:8080/predict/OneZeroPredictor

In this example, it will instantiate the OneZeroPredictor class and call the method to make predictions. The browser will download the CSV file called predictions.csv.

This file that you just download contains the predictions from your algorith and should be ready to submit.