<figure class="image"><img src="https://storage.yandexcloud.net/datasouls-ods/files/406cf2b9/2006.catalyst-intro.png" /></figure>

<p>Catalyst is <a href="https://twitter.com/PyTorch">PyTorch</a> framework for Deep Learning research and development. You get a training loop with metrics, model checkpointing, advanced logging and distributed training support without the boilerplate.</p>

<p>Part of <a href="https://pytorch.org/ecosystem/">PyTorch Ecosystem</a>. Part of <a href="https://docs.google.com/presentation/d/1D-yhVOg6OXzjo9K_-IS5vSHLPIUxp1PEkFGnpRcNCNU/edit?usp=sharing">Catalyst Ecosystem</a>:</p>

<ul>
	<li><a href="https://github.com/catalyst-team/alchemy">Alchemy</a> - Experiments logging &amp; visualization</li>
	<li><a href="https://github.com/catalyst-team/catalyst">Catalyst</a> - Accelerated Deep Learning Research and Development</li>
	<li><a href="https://github.com/catalyst-team/reaction">Reaction</a> - Convenient Deep Learning models serving</li>
</ul>

<h2><strong>Getting started</strong></h2>

<pre>
<code class="language-plaintext">pip install -U catalyst</code></pre>

<figure class="image"><img src="https://storage.yandexcloud.net/datasouls-ods/files/5e871d83/2006.code3.png" />
<figcaption>Catalyst pipeline example,<br />
&nbsp;with 30 lines of code you get model training, inference and tracing support</figcaption>
</figure>

<p><a href="https://github.com/catalyst-team/catalyst#minimal-examples">More minimal examples here.</a></p>

<h2>Overview</h2>

<ul>
	<li>Catalyst is pure PyTorch framework without any custom modules.</li>
	<li>With Catalyst you get the same PyTorch pipelines but in much more readable and fully reproducible way.</li>
	<li>Catalyst is easily extendable with Runner and Callback abstractions.</li>
	<li><strong>You focus on key parts of your deep learning R&amp;D.</strong></li>
</ul>

<p><a href="https://github.com/catalyst-team/catalyst/tree/master/tests">Catalyst is rigorously tested</a> and supports extra subpackages for <a href="https://github.com/catalyst-team/catalyst#installation">Computer Vision</a> and <a href="https://github.com/catalyst-team/catalyst#installation">Natural Language Processing</a>.&nbsp;</p>

<p><a href="https://github.com/orgs/catalyst-team/people">Catalyst.Team</a> and <a href="https://github.com/catalyst-team/catalyst/graphs/contributors">Catalyst.Contributors</a> consist of industrial and academia researches, kaggle masters and professional software engineers. You can check out Catalyst-based kaggle competitions, pipelines and projects <a href="https://github.com/catalyst-team/catalyst#catalyst">here</a>.</p>

<h3><strong>Tutorials</strong></h3>

<ul>
	<li><a href="https://github.com/catalyst-team/catalyst/blob/master/examples/notebooks/customizing_what_happens_in_train.ipynb">Customizing what happens in train</a></li>
	<li><a href="https://github.com/catalyst-team/catalyst/blob/master/examples/notebooks/demo.ipynb">Demo with minimal examples</a> for CV, NLP, GANs and RecSys</li>
	<li>Detailed <a href="https://github.com/catalyst-team/catalyst/blob/master/examples/notebooks/classification-tutorial.ipynb">classification tutorial</a></li>
	<li>Advanced <a href="https://github.com/catalyst-team/catalyst/blob/master/examples/notebooks/segmentation-tutorial.ipynb">segmentation tutorial</a></li>
	<li>Comprehensive <a href="https://github.com/catalyst-team/classification">classification pipeline</a></li>
	<li>Binary and semantic <a href="https://github.com/catalyst-team/segmentation">segmentation pipeline</a></li>
	<li><a href="https://evilmartians.com/chronicles/beyond-fashion-deep-learning-with-catalyst">Beyond fashion: Deep Learning with Catalyst (Config API)</a></li>
	<li><a href="https://github.com/Bekovmi/Segmentation_tutorial">Tutorial from Notebook API to Config API (RU)</a></li>
</ul>

<h2><strong>More than a framework &mdash; Ecosystem</strong></h2>

<figure class="image"><img src="https://storage.yandexcloud.net/datasouls-ods/files/76e76e07/2006.catalyst.ecosystem.png" />
<figcaption>Combining human creativity with machine&rsquo;s computational power</figcaption>
</figure>

<h3><strong>Catalyst.Ecosystem manifesto</strong></h3>

<p>Catalyst was designed by researchers for researchers,</p>

<ul>
	<li>to get the foundation for DL &amp; RL research, which would be well tested and verified</li>
	<li>to generalize and develop best practices that work with all DL areas</li>
	<li>to think less about tech-side and focus on research hypothesis and their value</li>
	<li>to accelerate your research</li>
</ul>

<p>Catalyst.Ecosystem,</p>

<ul>
	<li>combines research best practices and helps knowledge sharing</li>
	<li>connects DL/RL with Software Engineer for maximum performance</li>
	<li>allows quickly and efficiently validate your hypotheses</li>
</ul>

<p>Catalyst main principles,</p>

<ul>
	<li>open &mdash; it&rsquo;s fully Open Source Ecosystem</li>
	<li>equivalently &mdash; everyone can contribute and propose new ideas</li>
	<li>expertise &mdash; we are gathering top deep learning knowledge into one place</li>
	<li>high-performance &mdash; we are developing the ecosystem with software engineering best practices</li>
</ul>

<h3><strong>Open Source Research Community</strong></h3>

<p>Last but not least, <a href="https://github.com/catalyst-team/catalyst#trusted-by">Catalyst.Friends</a> community connects different startups, companies and research labs all over the world. For example, Catalyst.Team is working with Translational Research in Neuroimaging and Data Science (<a href="http://trendscenter.org/">TReNDS</a>) on DL <a href="https://github.com/catalyst-team/neuro">brain image analysis</a>. Feel free to <a href="mailto:catalyst.team.core@gmail.com">write us</a>, if you are also interested in such collaborations.</p>

<p>&nbsp;</p>

<p>Slack ODS.ai:&nbsp;<a href="https://opendatascience.slack.com/archives/CGK4KQBHD">#tool_catalyst</a></p>

Artem Zolkin

Sergey Kolesnikov

- bagxi -

PyTorch framework for accelerated DL/RL research and development

CatalystFounded 5 years ago

Catalyst
Founded 5 years ago