This repo is an adaptation of the official pytorch examples of mnist training.
Upon this repository we add a layer for tracking hyperparameters and metrics with
a tool called mlflow.
This is part of a presentation of this tool in a meetup in Toulouse in 2019.
To reproduce the experiment, just run the train_mniyst.sh. which will run 5 times the training with different hyperparamters. Don't forget to install the dependencies (torch ... etc).
To start the mlflow dashboard, just run mlflow server where the mlruns folder resides.
Author: Hacene Karrad, Computer Vision Engineer at Delair.