Here we have the results of two datasets containing recordings from sleep analysis.
The first dataset was recorded at the "National Institute of Mental Health" in Czech Republic and gracefully borrowed for initial investigations by Elizaveta S. The sampling frequency is 250 Hz and multiple 2s windows were labeled for different sleep stages.
The figure below shows an embedding using diffusion maps. Each point corresponds to a 24x24 covariance matrix (containing EEG, EMG and EOG recordings) and the colors indicate different sleep stages: REM, S1, S2, and S3.
The second dataset is a public one available at the Physionet website (link) and contains recordings from a sleep study between 1987 and 1991. The sampling frequency is 100 Hz and we have only 2 EEG channels (Fpz-Cz and Pz-Oz). Several windows of varying length were labeled for different sleep stages.
The figure below shows an embedding using diffusion maps. Each point corresponds to a 2x2 covariance matrix and the colors indicate different sleep stages: S1, S2, and S3.