Made With ML

     

If you need refresh yourself on ML algorithms, check out our ML Foundations repository (🔥  Among the top ML repositories on GitHub)

📆  New lessons every month!
Subscribe for our monthly updates on new content.

Directory structure

app/
├── api.py        - FastAPI app
└── cli.py        - CLI app
├── schemas.py    - API model schemas
tagifai/
├── config.py     - configuration setup
├── data.py       - data processing components
├── eval.py       - evaluation components
├── main.py       - training/optimization pipelines
├── models.py     - model architectures
├── predict.py    - inference components
├── train.py      - training components
└── utils.py      - supplementary utilities

Documentation for this application can be found here.

Workflows

Use existing model

1. Set up environment.

   bash
   export venv_name="venv"
   make venv name=${venv_name} env="dev"
   source ${venv_name}/bin/activate
   

2. Pull latest model.

   bash
   dvc pull
   

3. Run Application

   bash
   make app env="dev"
   

   You can interact with the API directly or explore via the generated documentation at http://0.0.0.0:5000/docs.

Update model (CI/CD)

Coming soon after CI/CD lesson where the entire application will be retrained and deployed when we push new data (or trigger manual reoptimization/training). The deployed model, with performance comparisons to previously deployed versions, will be ready on a PR to push to the main branch.

Update model (manual)

1. Set up the development environment.

   bash
   export venv_name="venv"
   make venv name=${venv_name} env="dev"
   source ${venv_name}/bin/activate
   

2. Pull versioned data and model artifacts.

   bash
   dvc pull
   

3. Optimize using distributions specified in

   tagifai.main.objective

   . This also writes the best model's params to config/params.json

   bash
   tagifai optimize \
   --params-fp config/params.json \
   --study-name optimization \
   --num-trials 100
   

   We'll cover how to train using compute instances on the cloud from Amazon Web Services (AWS) or Google Cloud Platforms (GCP) in later lessons. But in the meantime, if you don't have access to GPUs, check out the optimize.ipynb notebook for how to train on Colab and transfer to local. We essentially run optimization, then train the best model to download and transfer it's artifacts.

4. Train a model (and save all it's artifacts) using params from config/params.json and publish metrics to model/performance.json. You can view the entire run's details inside

   experiments/{experiment_id}/{run_id}

   or via the API (

   GET

   /runs/{run_id}).

   bash
   tagifai train-model \
   --params-fp config/params.json \
   -- model-dir model \
   --experiment-name best \
   --run-name model
   

5. Predict tags for an input sentence. It'll use the best model saved from

   train-model

   but you can also specify a

   run-id

   to choose a specific model.

   bash
   tagifai predict-tags --text "Transfer learning with BERT"  # test with CLI app
   make app env="dev"  # run API and test as well
   

6. View improvements Once you're done training the best model using the current data version, best hyperparameters, etc., we can view performance difference.

   bash
   tagifai diff --commit-a workspace --commit-b HEAD
   

7. Push versioned data and model artifacts.

   make dvc
   

8. Commit to git This will clean and update versioned assets (data, experiments), run tests, styling, etc.

   bash
   git add .
   git commit -m ""
   git push origin main
   

Commands

Docker

make docker  # docker build -t tagifai:latest -f Dockerfile .
             # docker run -p 5000:5000 --name tagifai tagifai:latest

Application

make app  # uvicorn app.api:app --host 0.0.0.0 --port 5000 --reload --reload-dir tagifai --reload-dir app
make app-prod  # gunicorn -c config/gunicorn.py -k uvicorn.workers.UvicornWorker app.api:app

Streamlit dashboard

make streamlit  # streamlit run streamlit/app.py

MLFlow

make mlflow  # mlflow server -h 0.0.0.0 -p 5000 --backend-store-uri stores/model/

Mkdocs

make docs  # python -m mkdocs serve

Testing

make great-expectations  # great_expectations checkpoint run [projects, tags]
make test  # pytest --cov tagifai --cov app --cov-report html
make test-non-training  # pytest -m "not training"

Start Jupyterlab

python -m ipykernel install --user --name=tagifai
jupyter labextension install @jupyter-widgets/jupyterlab-manager
jupyter labextension install @jupyterlab/toc
jupyter lab

You can also run all notebooks on Google Colab.

FAQ

Why is this free?

While this content is for everyone, it's especially targeted towards people who don't have as much opportunity to learn. I firmly believe that creativity and intelligence are randomly distributed but opportunity is siloed. I want to enable more people to create and contribute to innovation.

Who is the author?

• I've deployed large scale ML systems at Apple as well as smaller systems with constraints at startups and want to share the common principles I've learned along the way.
• I created Made With ML so that the community can explore, learn and build ML and I learned how to build it into an end-to-end product that's currently used by over 20K monthly active users.
• Connect with me on Twitter and LinkedIn

@article{madewithml,
    title  = "Applied ML - Made With ML",
    author = "Goku Mohandas",
    url    = "https://madewithml.com/courses/mlops/"
    year   = "2021",
}