MLOps Approach for Fraud Detection Example

Title:	Fraud Detection MLOps
Goal:	Implement MLOps using the Fraud Detection example.
Output:	Learn how to use Data Science Pipelines and Inference.
Timing:	1h
Notes:	We'll start with RHOAI installed.

---

TODO fix use crt

Prerequisites:

As part of the demo, you will have to do some changes and commits. So it is important that you fork the repository and clone it in your local.

git clone https://github.com/alpha-hack-program/fraud-detection-mlops.git

Steps to Run the Demo

• Update GitHub Credentials

Before proceeding with the demo, you need to update the GitHub credentials in the gitops/onboard-datascience/app-onboard-datascience.yaml file. These credentials are required for the pipeline to interact with your GitHub repository.

1. Open the file gitops/onboard-datascience/app-onboard-datascience.yaml.
2. Locate the following section:

   parameters:
   - name: "pipeline.github.token"
     value: "changeme_token"
   - name: "pipeline.github.user"
     value: "changeme_user"

• Set Up the Data Science Environment

Apply the following manifest to configure the data science environment:

oc apply -f gitops/onboard-datascience/app-onboard-datascience.yaml

• Prepare the Training Data (v1.csv)

Create a file named 'v1.csv' containing the data required for model training in data folder. you can rename 'card_transdata_sort.csv'. Ensure the file adheres to the expected format.

• Commit the Training Data

Commit and push the v1.csv file to the Git repository to trigger the model training pipeline:

git add data/v1.csv
git commit -m "Add data v1.csv for model training"
git push origin main

• Deploy the Inference Service

Apply the following manifest to deploy the inference service:

oc apply -f gitops/inference/app-inference.yaml

• Test the Inference Service

Once the model is trained and deployed, test the inference service by sending test data to the service endpoint.

host=<YOUR_HOST>
url="https://fraudinference-fraud.apps.$host/v2/models/fraudinference/versions/1/infer" 
data='{
        "id" : "42",
        "inputs": [
                    {
                        "name": "dense_input",
                        "shape": [1, 5],
                        "datatype": "FP32",
                        "data": [0.3111400080477545, 1.9459399775518593, 1.0, 0.0, 0.0]
                    }
                ]
        }'
curl -k -X POST "$url" -H "Content-Type: application/json" -d "$data"

• Update the Model with New Data

To train a new version of the model. Create a file named 'v2.csv' containing the data required for model training in data folder. you can rename 'card_transdata_full.csv'. Ensure the file adheres to the expected format.

Commit and push the v2.csv file to the Git repository to trigger the model training pipeline again:

git pull
git add data/v2.csv
git commit -m "Add data v2.csv for model training"
git push origin main

This will trigger the pipeline to train and deploy the updated model.

• Inference model v2

Update the version field in the gitops/onboard-datascience/values-datasciene.yaml file to 2 to use the new model version:

# filepath: gitops/inference/values-inference.yaml
inference:
  dataconnections: dataconnection-model
  version: 2
multimodel: false

Commit and push the file to the Git repository to inference model v2:

git add gitops/inference/values-inference.yaml
git commit -m "Inference model v2"
git push origin main

• Test the Inference Service with the Updated Model

After the updated model is deployed, test the inference service again to validate the new model's performance.

host=<YOUR_HOST>
url="https://fraudinference-fraud.apps.$host/v2/models/fraudinference/versions/1/infer" 
data='{
        "id" : "42",
        "inputs": [
                    {
                        "name": "dense_input",
                        "shape": [1, 5],
                        "datatype": "FP32",
                        "data": [0.3111400080477545, 1.9459399775518593, 1.0, 0.0, 0.0]
                    }
                ]
        }'
curl -k -X POST "$url" -H "Content-Type: application/json" -d "$data"