bruAristimunha commited on
Commit
ad04a21
·
verified ·
1 Parent(s): 95903b5

Card: list the training data (datasets:, base_model:) and add a Training data section

Browse files

Follow-up to braindecode PR #1493: `from_pretrained` reads `datasets:` into `model.card_data`. Ids: MOABB names, NEMAR nm/on ids, OpenNeuro ds ids; evidence per checkpoint.

Files changed (1) hide show
  1. README.md +30 -0
README.md CHANGED
@@ -7,6 +7,12 @@ tags:
7
  - sleep-staging
8
  - foundation-model
9
  - braindecode
 
 
 
 
 
 
10
  ---
11
 
12
  # SleepFMStager — pretrained sleep stager
@@ -38,6 +44,30 @@ bit-identical outputs.
38
  The upstream artifacts themselves are kept, byte-for-byte, in
39
  [`braindecode/SleepFM`](https://huggingface.co/braindecode/SleepFM).
40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
41
  ## Usage
42
 
43
  ```python
 
7
  - sleep-staging
8
  - foundation-model
9
  - braindecode
10
+ datasets:
11
+ - SSC
12
+ - MESA
13
+ - MrOS
14
+ - SHHS
15
+ base_model: braindecode/SleepFM
16
  ---
17
 
18
  # SleepFMStager — pretrained sleep stager
 
44
  The upstream artifacts themselves are kept, byte-for-byte, in
45
  [`braindecode/SleepFM`](https://huggingface.co/braindecode/SleepFM).
46
 
47
+ <!-- braindecode:training-data -->
48
+ ## Training data
49
+
50
+ The data that shaped these weights. The ids are the `datasets:` field of this card,
51
+ which braindecode reads into `model.card_data.datasets` when you call
52
+ `from_pretrained` (MOABB class names for MOABB datasets, OpenNeuro / NEMAR
53
+ accessions, Hub ids otherwise). `nm`/`on` ids are datasets on [NEMAR](https://nemar.org)
54
+ (`on` = NEMAR's copy of the OpenNeuro dataset with the same number). Rows marked
55
+ `private:` are data the authors did not release; they are not in `datasets:`.
56
+
57
+ | `datasets:` id | Dataset | Stage | Modality |
58
+ |---|---|---|---|
59
+ | `SSC` | [Stanford Sleep Clinic PSG cohort (Stanford Sleep Dataset)](https://sleepdata.org/datasets/ssc) | pretraining, fine-tuning | PSG |
60
+ | `private:BioSerenity` | [BioSerenity / SleepMed proprietary PSG collection (18,869 overnight recordings)](https://doi.org/10.1038/s41591-025-04133-4) | pretraining | PSG |
61
+ | `MESA` | [Multi-Ethnic Study of Atherosclerosis (sleep ancillary, NSRR)](https://sleepdata.org/datasets/mesa) | pretraining, fine-tuning | PSG |
62
+ | `MrOS` | [Outcomes of Sleep Disorders in Older Men (MrOS Sleep, NSRR)](https://sleepdata.org/datasets/mros) | pretraining, fine-tuning | PSG |
63
+ | `SHHS` | [Sleep Heart Health Study (NSRR)](https://sleepdata.org/datasets/shhs) | fine-tuning | PSG |
64
+
65
+ - Encoder weights = braindecode/SleepFM (leave-one-out contrastive pretraining, SHHS held out).
66
+ - Sleep-staging head (modality pooling, Transformer, BiLSTM, 5-class FC) trained on frozen encoder embeddings; upstream model_sleep_staging/config.json lists "dataset": "ssc_stanford,mesa,mros,shhs" (train split; SHHS test = 2,000 nights). BioSerenity has no staging labels and is not listed.
67
+ - Starts from the weights of [`braindecode/SleepFM`](https://huggingface.co/braindecode/SleepFM), whose card lists the data those saw.
68
+ - Sources: Thapa et al., Nature Medicine 2026, doi:10.1038/s41591-025-04133-4, Results 'Dataset and SleepFM architecture'; https://github.com/zou-group/sleepfm-clinical/blob/main/sleepfm/checkpoints/model_sleep_staging/config.json; same config.json; braindecode SleepFMStager docstring.
69
+
70
+
71
  ## Usage
72
 
73
  ```python