README.md

Time-warping time-frequency maps

Motivation

TODO

Workflow

1. Load the raw data and create epochs around events of interest. Make sure that epochs completely cover the longest epoch in the data. Since MNE supports only constant-length epochs, shorter epochs will contain irrelevant data at the end. This is fine, because the next steps will take care of this issue. Therefore, after this step, you should have an mne.Epochs object consisting of n epochs (let's call this object epochs) and an array of n epoch durations (in seconds) (let's call this array durations).
2. Now we compute a standard time-frequency representation (TFR) from the epoched data. MNE currently includes three TFR functions in mne.time_frequency, namely tfr_morlet(), tfr_multitaper(), and tfr_stockwell(). All of them produce a suitable TFR that can be used for time-warping. These functions can compute either averaged (average=True, the default) or single-epoch (average=False) TFRs. Since time-warping requires single-epoch TFRs (an mne.time_frequency.EpochsTFR object), we need to pass average=False (we will average them later).
3. Finally, we time-warp the mne.time_frequency.EpochsTFR object by passing it to the tfr_timewarp() function, together with the durations array defined in the first step. This will stretch/shrink all single-epoch TFRs to the same length, a process which we refer to as time-warping. Note that tfr_timewarp() returns another mne.time_frequency.EpochsTFR object with the same dimensions as the input object. However, the data from time 0 to the last time point is now time-warped, which means that it cannot be interpreted as time in seconds, but as a duration percentage (ranging from 0% to 100%).
4. Plotting (or post-processing) the time-warped mne.time_frequency.EpochsTFR object usually involves averaging over all epochs first. This can be achieved by calling the average() method.

Example

TODO