Demonstration of Training Pipeline for AGC Demo Day #2
This demo provides an example of how mlflow
can be used during training in the AGC context. A simple signal/background classification is used here for ease of demonstration. Some physics details may not be completely realistic here since this is mainly meant to serve as a demonstration of the training mechanics.
More information on mlflow
can be found here.
Additionally, xgboost
and sklearn
are used here, but the approach can be modified to most popular machine learning libraries, including tensorflow
and pytorch
.
The hyperparameter optimization workflow structure is similar to the one described in this article: https://medium.com/@chiefhustler/hyperparameter-tuning-with-dask-distributed-and-mlflow-ca6a4a275a2e