In the course of the project, data scientist faces two problems:
Steppy address both problems by introducing two simple abstractions:
Tranformer. We consider it minimal interface for building machine learning pipelines.
Stepis a wrapper over the transformer and handles multiple aspects of the execution of the pipeline, such as saving intermediate results (if needed), checkpointing the model during training and much more.
Tranformerin turn, is purely computational, data scientist-defined piece that takes an input data and produces some output data. Typical Transformers are neural network, machine learning algorithms and pre- or post-processing routines.
python3.5 or above.
pip3 install steppy
(you probably want to install it in your virtualenv)
Please send us your ideas on how to improve steppy library! We are looking for your comments here: Feature requests.
⏩ At this point steppy is early-stage library heavily tested on multiple machine learning challenges (data-science-bowl, toxic-comment-classification-challenge, mapping-challenge) and educational projects (minerva-advanced-data-scientific-training).
⏩ We are developing steppy towards practical tool for data scientists who can run their experiments easily and change their pipelines with just few manipulations in the code.
We are also building steppy-toolkit, a collection of high quality implementations of the top deep learning architectures -> all of them with the same, intuitive interface.
You are welcome to contribute to the Steppy library. Please check CONTRIBUTING for more information.
Steppy is MIT-licensed.