Newer
Older
As a scientist using machine learning packages, you will want to compare the
performance of different models and metrics for the same data. There are a
myriad of techniques, models and metrics out there. Nowadays,
many popular machine learning packages out there, for example `scikit-learn`
or `keras` have user-friendly interfaces in Python (or other languages).
When you need to test out several models and metrics at the same time, code
repetition and unnecessary verbosity sometimes creeps in to your projects.
The Renku MLS plug-in is designed to help you benchmark your machine learning
models easier, all whilst interfacing with the classic Renku command-line.
At present, the plug-in is compatible with `keras`, `scikit-learn` and `XGBoost`
with classification tasks and the following metrics: `accuracy_score`,
`roc_auc_score` and `f1_score`.
### A sneak peek
![renku mls leaderboard](images/sneak-peak.gif)
The dataset used in this project contains numeric data with the target variable
the `label` column. The `src/train.py` file is a training script for the data
and it uses command-line arguments to avoid code repetition. The `export()`
function exposes the model to the plug-in.
The `notebooks/renku-runs.ipynb` shows the example tasks run in this sample
project.