Time-frequency on long recordings

Problem. Time-frequency decomposition on a long, high-density recording exhausts memory or takes hours.

The pipeline

1. Filter and epoch first, so the decomposition sees only the trials.
2. Decimate to twice the highest frequency you care about.
3. Morse Wavelet — reduce n_freqs before reducing the range.

What to watch

The output size is trials times channels times frequencies times samples. Halving the frequency bins halves the memory; halving the sampling rate halves it again. Both are usually available before you need to reduce the frequency range.