workshop-moabb-bci-graz-2019's Introduction

10:00 - 10:30 : Classification methods in BCI

Marco's presentation

10:30 - 11:00 : MOABB Project

Motivation for creating/using MOABB
Present MNE
Present Scikit-learn (not all published estimators are available in there)
PyRiemann is an example of DYI classifier with scikit-learn template

11:00 - 12:00 : Installation and basic setup

An e-mail to all people participating in the workshop with the instructions to install MOABB, scikit-learn, MNE, pyriemann, etc., etc. (which datasets to download as well)
For those that are not comfortable in installing by themselves, we will propose a virtual machine with EVERYTHING installed and downloaded. The user will have less margin of control, but we will be sure that it works.

13:30 - 14:15 : Hands-on: easy benchmark

Part I : Basic concepts on Machine Learning classification, suppose we already have the trials (no mention about MOABB) Dataset downloaded by hand and loaded via scipy.io.loadmat Dataset BNCI2014001 ?

Part II : Create a scikit-learn notebook with step-by-step for classification

Show how to use MOABB for downloading, filtering, epoching data
Analyze the signals without classification (MOABB can do this)
Use scikit-learn/pyriemann in a step-by-step way for classification
- Explain cross-validation procedure ? KFold ?
- Do the steps one by one or use a make_pipeline
Use moabb's evalutation procedure with a pipeline created above

14:15-15:00 : Hands-on: write your own classifier ! Write your own dataset !

What are the methods and paradigm for creating a classifier/estimator on scikit-learn (fit, predict, etc)
How to create a dataset in MOABB
Small local "competition" for testing different pipelines and getting scores on certain datasets

Recommend Projects