jesus-333 / fbcsp-python

Python implemementation of the FBCSP algorithm

Hi! I'm Alberto Zancanaro, a PhD student from the University of Padua

Contacs

Currently working

• 🌱 Artificial intelligence for greenhouses (PhD main topic)
• 🌐 Age of Information (et similia) for network optimization

Past projects (University)

• 🧠 Variational Autoencoder (VAE) applied to EEG classification (Post graduation research)
• 🧠 Convolutional neural network applied to EEG classification [1] (Master's thesis)
• 🧠 Machine learning applied to EEG classification (CSP and FBCSP algorithm) (Master's thesis)
• 👋 Augmented reality in unity with hand tracking (Course project)
• 👋 Pytorch hand tracking (Course project)
• 💊 11 C-PIB tracer mathematical modelling (Bachelor's thesis)

Free time projects

• Google-Hash-Code-2019: Python solver for the Google Hash Code 2019
• Fractals generator in Python
• Cellular Automata in Python
• Arctic Circle Theorem visualization in Unity
• Evolution Simulator in Unity
• Solar System Simulation in Unity
• Neovim config (link)

Project's repository

💻 Technology	🚀 Projects

Pubblications

Click to expand!

• [1] A. Zancanaro, G. Cisotto, J.R.Paulo, G. Pires, and U. J. Nunes, “CNN-based approaches for cross-subject classification in motor imagery:From the state-of-the-art to Dynamicnet,” in 2021 IEEE Conferenceon Computational Intelligence in Bioinformatics and ComputationalBiology (CIBCB), 2021, pp. 1–7
• [2] A. Zancanaro, G. Cisotto, L. Badia, "Modeling Value of Information in Remote Sensing from Correlated Sources" in 2022 IEEE Mediterranean Communication and Computer Networking Conference (MedComNet)
• [3] A. Zancanaro, G. Cisotto, L. Badia, "Challenges of the Age of Information Paradigm for Metrology in Cyberphysical Ecosystems" in 2022 IEEE Metrology for Living Environment (MetroLivEnv)
• A. Zancanaro, G. Cisotto, D. Tegegn, L. Badia, S. L. Manzoni, I. Reguzzoni, E. Lotti, I. Zoppis, "Variational Autoencoder for Early Stress Detection in Smart Agriculture: A Pilot Study" in 2022 IEEE Metrology for Agriculture and Forestry
• Yeah very short... I hope to enlarge it in the future 😂

Stats

fbcsp-python's Issues

Custom Filter Bands( freqs_band) Issue with Fix

The default freqs_band argument choice FBCSP_4 is np.linspace(4, 40, 10) while I want to pass a custom one np.linspace(8, 32, 7) but your code gives a bug associated to this assertion (freqs_band == None)

Your Code

Frequencies band
         if(freqs_band == None): self.freqs = np.linspace(4, 40, 10)
         else: self.freqs =  freqs_band

My Fix

        if isinstance(freqs_band, np.ndarray):
            self.freqs = freqs_band
        elif(freqs_band == None):
            self.freqs = np.linspace(4, 40, 10)
        else:
            raise ValueError('freqs_band must be a Numpy Array')

Btw your code is really helpful in my project and luckily I found it. Thanks for making it opensource

Bug with custom classifier

If you uncomment below, and run your test_file_multi.py, you will know:
FBCSP_multi_clf = FBCSP_Multiclass(trials_dict, fs, classifier = SVC(kernel = 'rbf', probability = True), print_var = print_var)

Implementation Issue with MIBIF Feature Extraction

Given the paired nature of CSP, in the original FBCSP paper, the number of features per classifier in the overall one versus rest classifier can vary from 4 to 8 total, however, with the way the FBCSP_V4 class is set up only 8 features can be selected (based on the n_features parameter), but implementation wise this is not considered, If we run into a scenario say where we have six features selected, then an error will be thrown because of a shape mismatch between the classifier during training and when it is used for evaluation of new trials.

I have included a demonstration of this issue in my fork of your repo.

Best,
Tekin