Introduction to MLFlow

MLflow is an open source platform to manage the ML lifecycle, including experimentation, reproducibility, deployment, and a central model registry.

Why MLFlow?

MLflow lets you train, reuse, and deploy models with any library and package them into reproducible steps that other data scientists can use as a “black box,” without even having to know which library you are using.

Tutorial

<aside> 💡 We use mlflow autolog method to enables automatic logging from TensorFlow to MLflow.

</aside>

• Framework:

  1. Tensorflow / Keras (Can be deploy directly in PrimeHub deploy function.)
  2. Pytorch (Can be deploy directly in PrimeHub deploy function.)
  3. Scikit-learn (Can be deploy directly in PrimeHub deploy function.)
  4. XGBoost/LightGBM/Spark etc. (Need to write model server code additionally.)

• Add sample MLFlow autolog code to log our parameters, metrics and models automatically.

  import mlflow
  mlflow.set_experiment("<Experiment-Name>")
  mlflow.<framework>.autolog(log_models=True)
  
  # framework = tensorflow/keras/pytorch/Scikit-learn/xgboost/lightgbm/...
  # framework list: <https://mlflow.org/docs/latest/tracking.html#automatic-logging>
  

• After we finish running the code, we will get the record in MLFlow platform.

Example

Case : COVID-19 Face Mask Detection - MobileNetV2 (Kaggle)

We have a training model code in kaggle:

COVID-19 Face Mask Detection

• Notebook images:

  TensorFlow 2.4.1 (Python 3.7)

  [infuseai/docker-stacks:tensorflow-notebook-v2-4-1-dbdcead1]

• Step 1: Download data and upload to jupyterlab.

• Step 2: Unzip dataset.

• Step 3: Download code, upload to jupyterlab and modify data path.

• Step 4: We only need to create a cell at the begin of notebook and add the following code into cell.

  # Code Meaning:
  # Import MLFlow package and set the new experiment call "mask-detection".
  # Then, use MLFlow "autolog" method to record the Parameters, Metrics and Artifacts.
  # When we run this code, we do not get any log or information.
  # However, after we finish training model, you can see the job in MLFLow platform.
  
  import mlflow
  mlflow.set_experiment("mask-detection")
  mlflow.tensorflow.autolog(log_models=True)
  

• Step 5: We can train the model and get results from MLFlow to register the model.