trackable

A minimalistic machine learning model tracker and reporting tool

• Documentation: https://MillenniumForce.github.io/trackable
• GitHub: https://github.com/MillenniumForce/trackable
• PyPI: https://pypi.org/project/trackable/
• Free software: MIT

trackable is a package focussed on users already familiar with machine learning in Python and aims to:

1. Provide a minimal model tracking tool with no frills
2. An intuitive and lightweight api

Installation

The latest released version can be installed from PyPI using:

# pip
pip install trackable

Features

To start using trackable import the main reporting functionality via:

from trackable import Report

It's simple to start using the package. The example below (although simplistic) shows how easy it is to pick up the api:

from sklearn.datasets import make_classification
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score, f1_score, roc_auc_score
from trackable import Report

X, y = make_classification()

lr = LogisticRegression().fit(X, y)
rf = RandomForestClassifier().fit(X, y)

# Instantiate the report...
report = Report(X, y, metrics = [accuracy_score, f1_score, roc_auc_score])

# Add models...
report.add_model(lr)
report.add_model(rf)

# Generate the report...
report.generate()

Credits

This package was created with Cookiecutter and the waynerv/cookiecutter-pypackage project template.