guglielmosanchini/AI_Workflow_Capstone_Project

IBM AI Workflow Capston Project

IBM AI Workflow Capstone Project

https://github.com/aavail/ai-workflow-capstone

Tests

To run all tests at once:

python -m pytest 

To run specific tests:

python -m pytest tests/{filename}

where {filename} can be test_api.py, test_logger.py or test_model.py.

Scripts

The scripts package contains modules to download and clean data, to transform data and build the model(s), to build loggers and to monitor the model in production.

Data

The data folder contains:

• some binary files used to plot;
• the folder where models and model configs are saved;
• the folder where training and production datasets are saved.

Docker

To build a Docker image:

docker build -t image_name .

To see Docker images:

docker images

To run it:

docker run -p 4000:8080 image_name

then go to http://0.0.0.0:4000/.

Update training and production data

To update training and production data, run:

export PYTHONPATH="."
python scripts/utils.py

The new training and production dataframes will be saved in ./data/datasets.

Monitor production performance

export PYTHONPATH="."
python scripts/monitoring.py

The score of the model on production data will be printed, and some plots shown.

Additional info

What could have been done more:

• perform extensive gridsearch for country-specific models
• add confidence intervals estimation to provide a risk metric.
• use a non supervised approachs.
• add more tests.
• take security into account, e.g. using The Adversarial Robustness Toolbox.
• add periodic automatic retraining.