DiGyt’s Blog

Entropy-based measures in Machine Learning

2023-02-26T00:00:00+00:00

Introducing: ASRpy

2022-07-02T00:00:00+00:00

Last Year, when I started comparing different EEG cleaning algorithms (see this blog post), I was looking for a valid and trustworthy Python implementation of the original algorithm for Artifact Subspace Reconstruction.

Artifact Subspace Reconstruction is a promising EEG cleaning algorithm, which is gaining more and more popularity due to its convincing cleaning results. It was originally developed for an EEGLAB extension in MATLAB, but there was no perfectly equivalent version in Python. Following preceding work done by Nicholas Barascud, and the original implementation by Christian Kothe and Scott Makeig, I started to create a Python version of Artifact Subspace Reconstruction that would be perfectly equivalent with the original MATLAB implementation. After a lot of work and comparisons I succeded in doing so, producing an Artifact Subspace Reconstruction which was only differing from the MATLAB version by external factors, such as slightly differing Eigenspace solver algorithms used in numpy and MATLAB. However, all other operations were perfectly replicating the MATLAB code.

The result can be found in my ASRpy repository. Information on the algorithm, installation and examples can be found there.

IF you want to know how to apply ASRpy to your own Python EEG data (either with MNE-Python or with numpy arrays), you can go through this quick tutorial:

Introducing: CatEyes

2021-11-18T00:00:00+00:00

Over the last years, I have been working a lot with classification of eye tracking data.

I found that there are tons of different gaze classification available out there, but many of them took a terrible amount of installation, implementation, and data wrangling work until ready for application. If you had to apply gaze classification at one point in your career, you might know what I’m talking about. As this problem deterred me and many of my lab members from actually using these algorithms, I decided to unite all these different algorithms in one toolbox, and unite all their functionality into one simple and easily applicable workflow.

The result of this work is the CatEyes Toolbox. It does not allow you to test different gaze classification algorithms with very simple and standardized functions, it also provides some tools to easily handle and visualize eye tracking data, especially with respect to classification.

If this idea caught your attention, feel free to check out the repository, where you can find everything you need to know about application, installation, as well as practical examples on how to directly apply gaze classification to your specific eye tracking project. Probably the best way to get an idea of how CatEyes works, is by using our CatEyes minimal example. You can inspect the intrdocutory example below, or click on the “Open in Colab” button, to run it in Google Colab’s hosted runtime.

Of course this is by far not everything you can do with CatEyes. If you wonder what other functionality CatEyes provides, you can check out the repository, or have a look at the CatEyes documentation.

Automated EEG Cleaning - A Comparison

2021-06-14T00:00:00+00:00

Example: Automated EEG Cleaning

2021-06-07T00:00:00+00:00

Will there be snow on Christmas?

2021-02-09T00:00:00+00:00

What is noise? (Baby don’t hurt me)

2021-02-08T00:00:00+00:00

Texture Generation with Spatial GANs

2021-01-28T00:00:00+00:00

Last year, I teamed up with Lucas Feldmann and Josefine Zerbe to replicate the SGAN network designed by Jetchev, Bergmann & Vollgraf (2016) from Zalando Research. Their SGAN model provides a nice way to generate non-repeating patterns of any texture. In the following ipynb notebook, we describe the model and training results on different textures. After that, we provide another notebook including a Tensorflow implementation of the original SGAN architecture as well as the training procedure that we used to generate our textures. See for yourself and feel free trying out to train the network on your own texture images:

Probabilistic Poker Bots

2020-12-14T00:00:00+00:00

Large graphs and small worlds

2020-12-06T00:00:00+00:00

Making Sense of networks can be hard. In some cases, using graph theory can help to understand certain properties of the underlying network, even though the network itself is too complex to understand. In the following script, you can investigate Python implementations for a number of graph theory measures which we used to analyze brain connectivity patterns.