Giter Site home page Giter Site logo

bci_chips's Introduction

BCI_chips

Optimization of a space representation for a better SVM classification

The main file is main_all_sub_test_multi_chan.m

All these optimization methods are undertaken under the leave-one-out meaning that for each optimization method, the training set is every signal unless one and we test the signal which is left. When you make a loop over every possible training set, you can then make statistics on the accuracy of the process.

For more information about the topic please read the PhD thesis of Xavier Artusi.

Once it is opened and launched you will trigger this sucession of function:

wavelet transform with classic db4 wavelet (without any optimization)

if dwt_without_optim = 1 in the main file, you will run:

  • programme_test_descripteur_xval.m
    * calc_feature (used to have a first set of coefficient computed on db4) * svm_learning (used to train the set of coefficient computed on db4) * test_class_ovr (used to compute class labeling on the test set) * function_acp (let us see in a better way of understanding the axes and sample influence on the results)

wavelet transform with error probability criterion estimation

if optim_pce = 1 in the main file, you will run:

  • programme_test_descripteur_xval.m
    • create_dic (computes every set of classical wavelet coefficients on a parameterized wavelet over a theta)
    • calc_feature (used to have a first set of coefficient computed on db4, in order to compare the effectiveness of our parameterization)
    • optim_wavelet_multi_cv (gives the best theta and its coefficients according to the pce criterion)
    • svm_learning (used to train the set of coefficient computed on a theta optimal)
    • test_class_ovr (used to compute class labeling on the test set)
    • function_acp (let us see in a better way of understanding the axes and sample influence on the results)

wavelet transform with Fisher criterion estimation

if optim_fisher = 1 in the main file, you will run:

  • programme_test_descripteur_xval.m
    • create_dic (computes every set of classical wavelet coefficients on a parameterized wavelet over a theta)
    • calc_feature (used to have a first set of coefficient computed on db4, in order to compare the effectiveness of our parameterization)
    • optim_wavelet_multi_cv (gives the best theta and its coefficients according to the fisher criterion)
    • svm_learning (used to train the set of coefficient computed on a certain theta)
    • test_class_ovr (used to compute class labeling on the test set)
    • function_acp (let us see in a better way of understanding the axes and sample influence on the results)

wavelet packet transform: the best basis adventure

if optim_basis = 1 in the main file, you will run:

  • programme_test_descripteur_xval.m
    • set_wp_marg(computes all the mavelet packet marginals)
    • best_basis search (search what will be the best basis among all possible basis from the total set of marginals)
      • calc_fisher_wp_features (computes all the additive criteria used to choose the best basis)
      • find_best_basis_fisher (analyzes the criteria to find the best basis and plot the final tree with the chosen coefficients)
    • svm_learning (used to train the set of marginals chosen thanks to the previous function)
    • test_class_ovr (used to compute class labeling on the test set)
    • function_acp (let us see in a better way of understanding the axes and sample influence on the results)

wavelet packet transform optimization + PCA pre "lifting"

if optim_basis_acp = 1 in the main file, you will run:

  • programme_test_descripteur_xval.m
    • set_wp_marg(computes all the mavelet packet marginals)
    • best_basis search (search what will be the best basis among all possible basis from the total set of marginals)
      • calc_fisher_wp_features (computes all the additive criteria used to choose the best basis)
      • find_best_basis_fisher (analyzes the criteria to find the best basis and plot the final tree with the chosen coefficients)
    • function_acp (used to have less axes in the space representation in order to undertake a more efficient svm classification)
    • svm_learning (used to train the set of marginals chosen thanks to the previous function)
    • test_class_ovr (used to compute class labeling on the test set)
    • function_acp (let us see in a better way of understanding the axes and sample influence on the results)

hybrid method

if optim_algo = 1 in the main file, you will run:

NOTHING, this method needs to be finished. The principle should be first to find an optimized mother wavelet on the standard wavelet coefficients and then to find a better basis to use this wavelet and then to compute a new optimized mother wavelet over this new basis to increase the classification results and looping this two steps till a certain criterion value.

bci_chips's People

Watchers

 avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.