Machine Learning Model Deployment with FastAPI and Heroku

Census project starter kit

Census project: Predict whether income exceeds $50K/yr based on census data. Also known as Adult dataset.

Data has been downloaded from the Udacity nd0821-c3 project starter kit

The UC Irvine Machine Learning Repository is where you can find information on the original dataset

Environment

• Create your conda environment:

  $ conda create --name <your environment name> --file requirements.txt
  $ conda env create --file conda.yaml
  
  $ conda activate <your environment name>
  

Data cleaning:

Data cleaning can be performed by using the Jupyter notebook "Census_Clean_Data.ipynb". The notebook also provides a good overview of the data. Data cleaning will also be provided via Python file: "/ml/data_cleaning.py"

Sanity check:

$ python -m tests.sanitycheck

Answer path question with "tests.api_tests.py" as test file for a check of functionality to meet course specifications

Training the model:

$ python -m ml.train_model

After the model has been trained successfully, the following files will be saved:

Metrics will be written to "/artifacts/slice_output.txt"

Model will be saved to file "/artifacts/model.joblib"

Encoder will be saved to "/artifacts/encoder.joblib"

Label binarizer will be saved to "/artifacts/lb.joblib"

The output will be shown on screen and also be saved in "/logs/census.log"

Census API tests:

Start the uvicorn server with:

$ uvicorn main:app --reload

The server is then accessible via: "http://127.0.0.1:8000"

Documents can be found here: "http://127.0.0.1:8000/docs"

FastAPI tests can be performed by using the Jupyter notebook "Census_Tests_API.ipynb".

Pytest:

Pytest will run all tests in the tests folder and can be executed via:

$ pytest -vv

Model card

Find detailed information in the "model_card.md"

GithubActions

If changes have been made, github actions is called.

Continuous Deployment to Heroku

The Heroku app can be tested by using the Jupyter notebook "Census_Test_Heroku.ipynb".

The app deployed on Heroku can be accessed at: "https://census-salaries-d3e2956470bf.herokuapp.com/"