Qt Contributors Summit 2019 Program/Qt Machine Learning and Math

Introduction

• Slides:
• The Python modules that are used in ML have C++ implementations under the hood. Here are some examples:

 * TensorFLow
 * Keras
 * Scikit-learn
 * Theano
 * PyTorch

• In DS the main Python modules are:

 * NumPy
 * Pandas

• Data science work consists in Data Extracting, transforming it, loading it and cleaning it, and just then they apply ML techniques;
• Math in C++:

 * cmath
 * Armadillo
 * We don't have math modules in Qt because it's hard to have something so specific;

Thoughts on how Qt and Qt for Python fit in this market

• Make the interface interaction with Qt and these libraries

 * The point is not reimplementing data frames or numpy stuff, we want to add support to it on PySide, having widgets ready to use;

• PyTorch offers some support that'd make it easy to integrate with Qt;
• Create training interfaces [...];
• Show insides of a module like TensorFlow it's doing in its website;
• Important to ask ourselves what are we bringing to the table, because there is a lot of things that are already out there doing a pretty good job;

 * It's hard to get in to a market that's already established;

• Offer a tool that allows data scientists to interact with the data and modify it;
• If we had a better graphical display system in Qt it'd be easier to implement this kind of thing. Maybe the answer is to fix QtChart, or maybe the answer is to kill QtChart;

 * We could provide building blocks to people to build their own dashboard;
   * It'd help a lot the DS people to visualize their data, helping them cleaning it;
 * It probably makes more sense to just use the tools available in Python and forget about the C++;