ML Ops using AWS Fraud detector

The Basics

Lets get AWS Fraud detector up and running in a single account first using a Notebook!

This section demonstrates how to train and deploy Amazon fraud detector using Boto3.

Prerequisites

1. Python 3.6.5 - For more details, see https://www.python.org/downloads/release/python-365/
2. Setup Virtual env - For more details, see https://docs.python.org/3.6/tutorial/venv.html

Local Setup

1. Once you have activated your virtual env, install dependencies in the command line

   pip install -r ./src/requirements.txt 
   pip install -r ./tests/requirements.txt
   pip install -r ./requirements_notebook.txt
   

2. To check if the install & setup works, run the tests in the command line

   export PYTHONPATH=./src
   pytest 
   

Run

See sample code in notebook ReadmeFraudDetector.ipynb. To run on the cli see samples below

Data Transformation

1. The scaffolding for data transformation code is in src/main_demo_data_transformer.py. To run transformation with sample data

export PYTHONPATH=./src
# get help
python ./src/main_demo_data_transformer.py -h

# sample run
python ./src/main_demo_data_transformer.py --s3desturi s3://mybucket/fraud-demo/

Training

1. The scaffolding for training code is in src/main_demo_fraud_detector_train.py. To run training with sample data that we created using the transformation step above. Note: A prerequisite is the the roles need to be set up for Fraud detector to run. See infra/README.md on how to set deploy fraud detector role

export PYTHONPATH=./src
# get help
python ./src/main_demo_fraud_detector_train.py -h

# sample run
python ./src/main_demo_fraud_detector_train.py --s3uri s3://mybucket/fraud-demo/train.csv --role <roleArnAssumedByFraudDetectorToAccessS3data> --sampledata "<sample_training_data>"

Deploy Model

1. The scaffolding for deployment code is in src/main_demo_fraud_detector_deploy.py. To run deploy the model that we created using the training step above

export PYTHONPATH=./src
# get help
python ./src/main_demo_fraud_detector_deploy.py -h

# sample run
python ./src/main_demo_fraud_detector_deploy.py --model sample_model_name --modelVersion 1.0 --detector demo

UnDeploy Model

1. The scaffolding for undeployment code is in src/main_demo_fraud_detector_undeploy.py. To undeploy a model sample_model_name, version 1.0 and detector demo

export PYTHONPATH=./src
# get help
python ./src/main_demo_fraud_detector_undeploy.py -h

# sample run
python ./src/main_demo_fraud_detector_undeploy.py --model sample_model_name --modelVersion 1.0 --detector demo

MLOps and Multiaccount deployment using CDK

To create a multi-account Codepipeline workflow to deploy Fraud Detector, See ./infra/README.md

Useful resources

• AWS Fraud Detector github samples

Security

See CONTRIBUTING for more information.

License

This library is licensed under the MIT-0 License. See the LICENSE file.