This Lightweight, Python library ‘Steppy’ can be used for fast and reproducible data science/machine learning experimentation

‘Steppy’ is a light python-3 library that can be used for fast and reproducible data science/machine learning experimentation. It reduces the burden of data scientists from software development issues. The minimal interface does not impose constraints for ‘Steppy’; instead, it enables clean machine learning pipeline design.

‘Steppy’ solves some of the data science project problems with the help of minimal interface for building machine learning pipelines. It uses two simple abstractions: Step and Tranformer

Step: It is the execution wrapper over the transformer. Example: Checking intermediate results.

Tranformer: It represents the computation step and performs operation on data. Mostly, Transformers are neural networks, machine learning algorithms..

Installation

Steppy requires python3.5 or above.

pip3 install steppy

Getting started with steps (Code Source: https://github.com/neptune-ml/steppy-examples/blob/master/tutorials/1-getting-started.ipynb)

This notebook shows how to create steps, fit them to data, transform new data and take advantage of persistence

%load_ext autoreload
%autoreload 2

import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

from steppy.base import Step, BaseTransformer

EXPERIMENT_DIR = './ex1'

import shutil

# By default pipelines will try to load previously trained models so we delete the cache to ba sure we're starting from scratch
shutil.rmtree(EXPERIMENT_DIR, ignore_errors=True)

Github: https://github.com/neptune-ml/steppy

Tutorial notebooks (their repository):

• ▶️ Getting started
• ▶️Steps with multiple inputs
• ▶️ Advanced adapters
• ▶️ Caching and persistance
• ▶️ Steppy with Keras