Skip to content

Using HappyFeat

Feature Selection & Training the Classifier (right panel)

This step allows to train the classification algorithm, using sub-sets of sensors and frequencies.

First, select the relevant sensor/frequency pair(s):

  • Use this format: Electrode;Frequency separated with a semicolon (;). Example: C4;13
  • A frequency range can be entered, using a colon symbol(:) as such: C4;13:16
  • If you want to use more than one feature, click Add Feature to add a field.
  • You can remove the last field using Remove Last Feat
  • Alternatively, you can automatically fill these fields by using the Automatic Feat. Selection mechanism.
  • Note: If you have setup your workspace to allow for using two metrics, you can choose to use training features for only one of the metrics or for both metrics

Note

These features can be automatically entered using the "AutoFeat" mechanism. Note that you can still add/remove training features after using this mechanism.

  • The number of k-fold partitions to use for the cross-validation step can be set.

The list of runs available for training is updated with runs which have undergone feature extraction in this session (see paragraph on sessions in workspaces)

Select the run(s) you want to use for training in the list. Trials from all selected runs are considered ("concatenated") for the following training attempt.

Click on "Train classifier". After processing, a detailed report of accuracies is provided in a pop-up window, and a shorter summary is made available in the bottom part of this panel. Results from all training attempts of the current session are listed. This process can be run as many times as needed.

Note

For a given session, each training attempt is numbered, and the computed classifier weights can be found in <happyfeat_install>/<currentWorkspace>/sessions/<sessionNb>/train/classifier-weights-<nb>.xml

An Online/Testing scenario can be generated by selecting a training attempt in the list, and clicking on "Use selected classifier (Online scen.)". This will update the scenario sc3-online.xml (in <happyfeat_install>/<currentWorkspace>/) with the corresponding channel/frequency features and the trained classifier's weights.

Combination training

See dedicated page.

Using a trained classifier

See dedicated page.