serotiny

While going about the work of building deep learning projects, several simultaneous problems seemed to emerge:

How do we reuse as much work from previous projects as possible, and focus on building the part of the project that makes it distinct?
How can we automate the generation of new models that are based on existing models, but vary in a crucial yet non-trivial way?
When generating a multiplicity of related models, how can we keep all of the results, predictions, and analyses straight?
How can the results from any number of trainings and predictions be compared and integrated in an insightful yet generally applicable way?

Serotiny arose from the need to address these issues and convert the complexity of deep learning projects into something simple, reproducible, configurable, and automatable at scale.

Serotiny is still a work-in-progress, but as we go along the solutions to these problems become more clear. Maybe you've run into similar situations? We'd love to hear from you.

Overview

serotiny is a framework and set of tools to structure, configure and drive deep learning projects, developed with the intention of streamlining the lifecycle of deep learning projects at Allen Institute for Cell Science.

It achieves this goal by:

Standardizing the structure of DL projects
Relying on the modularity afforded by this standard structure to make DL projects highly configurable, using hydra as the framework for configuration
Making it easy to adopt best-practices and latest-developments in DL infrastructure by tightly integrating with
- Pytorch Lightning for neural net training/testing/prediction
- MLFlow for experiment tracking and artifact management

In doing so, DL projects become reproducible, easy to collaborate on and can benefit from general and powerful tooling.

Getting started

For more information, check our documentation, or jump straight into our getting started page, and learn how training a DL model can be as simple as:

$ serotiny train data=my_dataset model=my_model

Authors

Guilherme Pires @colobas
Ryan Spangler @prismofeverything
Ritvik Vasan @ritvikvasan
Caleb Chan @calebium
Theo Knijnenburg @tknijnen
Nick Gomez @gomeznick86

Citing

If you find serotiny useful, please cite this repository as:

Serotiny Authors (2022). Serotiny: a framework of tools to structure, configure and drive deep learning projects [Computer software]. GitHub. https://github.com/AllenCellModeling/serotiny
Free software: BSD-3-Clause

Metadata

Release files for serotiny 0.0.0.dev20220504

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for serotiny 0.0.0.dev20220504
File	Size	Uploaded
serotiny-0.0.0.dev20220504.tar.gz	43.4 kB	May 4, 2022	Details

Built distribution (wheel)

Table of built distributions (wheels) for serotiny 0.0.0.dev20220504
File	Interpreter	ABI	Platform	Reset
serotiny-0.0.0.dev20220504-py3-none-any.whl 66.0 kB May 4, 2022	Python 3	none	any	Details

Total release size: 109.4 kB

Release files / serotiny-0.0.0.dev20220504.tar.gz

Download URL	serotiny-0.0.0.dev20220504.tar.gz
Size	43.4 kB
Tags	Source
SHA-256 checksum How to use checksums	`13d7925ad7ea0fb827efcacf43be570753fb916cec2e9200dec9ec39a52b5557`
BLAKE2b-256 checksum How to use checksums	`d5850e69072dd95e3ec2f9825bc45fa1bb3e40ff66e764bc63163282048a45df`
Upload date	May 4, 2022
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/4.0.0 CPython/3.9.12`

Release files / serotiny-0.0.0.dev20220504-py3-none-any.whl

Download URL	serotiny-0.0.0.dev20220504-py3-none-any.whl
Size	66.0 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`c2213b3a92e3f4ac5d918787e37907094194b44f4f828978948b523ca4523065`
BLAKE2b-256 checksum How to use checksums	`abb27bcb0a30b89900facdf0731b73ccfb041f6c76bcdaecc27275d68d8bb41e`
Upload date	May 4, 2022
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/4.0.0 CPython/3.9.12`