Managing ML-lifecycle with MLflow

Installation

0. Clone this repository using git or download using Github interface

1. Install miniconda distribution for python 3.7 -> link

2. Open Anaconda Prompt (Miniconda 3) navigate to the tutorial folder and install requirements:

   • Optional: you can create separate conda environment for this tutorial from environment file:

     conda env create -f conda.yaml
     

     conda activate mlflow-tutorial
     

   • Installing requirements from requirements.txt file:

     pip install -r requirements.txt -f https://download.pytorch.org/whl/torch_stable.html
     

3. Launch Jupyter Lab from from the tutorial directory:

   jupyter lab
   

4. Launch MLflow server:

   mlflow server --backend-store-uri="sqlite:///C:\\path\\to\\project_folder\\backend\\mlflow_data.db" 
                 --default-artifact-root="file:///C:\\path\\to\\project_folder\\artifact_store\\"
   

   • Parameters:

     • backend-store-uri - URI to which to persist experiment and run data (sqlite database in our case).
     • default-artifact-root - Local or S3 URI to store artifacts, for new experiments (local folder in our case).

   • Hint: you can launch MLflow server also from terminal inside Jupyter Lab interface.

5. Make sure that you can access mlflow server and jupyter notebook from your browser :

   • mlflow server : http://localhost:5000/
   • jupyter notebook: http://localhost:8888/

Note: There could be some minor changes in the dependencies and files, thus make sure to double-check prior to the event. Note: Tutorial functionality was tested on Windows 10.

Congrats! You are ready good to go!