4rooms/tsinalis-2016-embeddings

Tsinalis-2016 Embeddings

Pre-extracted per-epoch embeddings from TsinalisCNN (Tsinalis et al. 2016), trained on Sleep-EDF via PhysioEx.

Each subject directory contains:

• embeddings.npy — (n_epochs, 500) per-epoch embeddings (bfloat16)
• labels.npy — (n_epochs,) AASM sleep stage labels

Note: Tsinalis concatenates L=5 epochs before convolution. Embeddings are extracted using centered windows (each epoch gets the feature vector from the window centered on it).

Usage

from physioex.models import load_embeddings
path = load_embeddings("tsinalis-2016", "hmc", verbose=True)

Linear Probe Results (5-fold subject-wise CV)

Dataset	Subjects	ACC	MF1	κ	F1-W	F1-N1	F1-N2	F1-N3	F1-REM
sleepedf	153	0.8297	0.7563	0.7631	0.93	0.41	0.86	0.80	0.78
mass_ss05	26	0.8157	0.7004	0.7297	0.72	0.26	0.87	0.82	0.83
mass_ss02	19	0.8062	0.6925	0.7134	0.70	0.28	0.87	0.83	0.78
mass_ss03	62	0.7950	0.7039	0.6923	0.77	0.33	0.87	0.76	0.80
mass_ss04	40	0.7842	0.6791	0.6865	0.73	0.24	0.86	0.79	0.77
mass_ss01	53	0.7473	0.6691	0.6388	0.83	0.36	0.83	0.61	0.72
stages_GSDV	232	0.7368	0.5371	0.5643	0.75	0.08	0.83	0.47	0.56
dcsm	255	0.7329	0.5575	0.5947	0.82	0.02	0.73	0.77	0.45
shhs_visit2	2651	0.7301	0.5995	0.6097	0.80	0.11	0.77	0.71	0.60
shhs_visit1	5793	0.7267	0.5957	0.6001	0.80	0.10	0.77	0.69	0.62
stages_GSLH	45	0.7178	0.5317	0.5367	0.79	0.13	0.80	0.57	0.37
stages_MSMI	63	0.7121	0.5953	0.5608	0.75	0.15	0.79	0.64	0.64
stages_GSSW	105	0.7096	0.5009	0.5234	0.72	0.07	0.81	0.36	0.54
stages_GSSA	26	0.6989	0.4432	0.4483	0.64	0.01	0.81	0.30	0.46
stages_GSBB	30	0.6912	0.5546	0.5269	0.81	0.16	0.77	0.59	0.45
stages_MSQW	153	0.6813	0.5477	0.5151	0.77	0.23	0.79	0.45	0.51
stages_STLK	158	0.6639	0.4954	0.4560	0.67	0.05	0.76	0.43	0.56
hmc	151	0.6634	0.6088	0.5473	0.72	0.25	0.70	0.74	0.63
hpap	247	0.6595	0.5774	0.5192	0.71	0.17	0.72	0.69	0.59
mesa	2056	0.5748	0.4225	0.3371	0.55	0.02	0.67	0.42	0.46
stages_STNF	460	0.5484	0.4682	0.3925	0.63	0.00	0.57	0.67	0.47
stages_MSNF	38	0.5108	0.2886	0.1714	0.45	0.02	0.65	0.26	0.06
stages_MSTH	31	0.4942	0.1529	-0.0035	0.02	0.02	0.67	0.03	0.03
stages_MSTR	285	0.4938	0.2544	0.1110	0.30	0.01	0.64	0.25	0.08
stages_BOGN	85	0.4578	0.1892	0.0419	0.04	0.00	0.62	0.27	0.01

Model Details

• Architecture: TsinalisCNN (Tsinalis et al. 2016) — 2-layer CNN on concatenated 5-epoch signal
• Training data: Sleep-EDF (153 subjects, single EEG channel)
• Pipeline: identity (Sleep-EDF) / raw (other datasets, resample to 100 Hz)
• Sequence length: L=5 epochs
• Embedding dim: 500

Datasets

Dataset	Source	URL
Sleep-EDF	PhysioNet	https://physionet.org/content/sleep-edfx/1.0.0/
HMC	PhysioNet	https://physionet.org/content/hmc-sleep-staging/1.1/
DCSM	ERDA/KU	https://erda.ku.dk/public/archives/db553715ecbe1f3ac66c1dc569826eef/published-archive.html
SHHS	NSRR	https://sleepdata.org/datasets/shhs
MESA	NSRR	https://sleepdata.org/datasets/mesa
HomePAP	NSRR	https://sleepdata.org/datasets/homepap
STAGES	NSRR	https://sleepdata.org/datasets/stages
MASS	CEAMS	http://ceams-carsm.ca/mass/

Citations

@article{gagliardi2025physioex,
    author={Gagliardi, Guido and Alfeo, Luca and Cimino, Mario G C A and Valenza, Gaetano and De Vos, Maarten},
    title={PhysioEx, a new Python library for explainable sleep staging through deep learning},
    journal={Physiological Measurement},
    url={http://iopscience.iop.org/article/10.1088/1361-6579/adaf73},
    year={2025},
}

@article{tsinalis2016automatic,
    title={Automatic Sleep Stage Scoring with Single-Channel EEG Using Convolutional Neural Networks},
    author={Tsinalis, Orestis and Matthews, Paul M and Guo, Yike and Zafeiriou, Stefanos},
    journal={arXiv:1610.01683},
    year={2016},
}