Note
Go to the end to download the full example code or to run this example in your browser via Binder.
Regress observed p-factor with a neural EEG decoder#
Estimated reading time:9 minutes
Train a compact Braindecode EEGNet on recorded windows and aggregate predictions for held-out participants. Two epochs exercise CPU training; this is not a converged or clinically validated model.
Use six explicitly selected participants from the real
HBN ds005505 release. Their
RestingState signal files total approximately 595.8 MB, cached under
EEGDASH_CACHE_DIR. This applied example is larger than the introductory
21 MB SSVEP subset. Cropping after loading reduces computation, not download.
Targets come from the observed participant metadata. Six participants are
sufficient to exercise the workflow, not to support clinical conclusions.
Before you start#
Install EEGDash with EEGPrep, Braindecode and PyTorch. Tutorial 02 explains minibatches; tutorial 11 explains grouped evaluation. The feature-based p-factor project provides a simpler comparison, but no saved output is required here. This script trains a fresh compact network in each fold and reports participant errors after aggregating its real window predictions.
import os
from pathlib import Path
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from braindecode.preprocessing import (
create_fixed_length_windows,
preprocess,
Resampling,
RemoveDrifts,
)
from eegdash import EEGDashDataset
import torch
from braindecode.models import EEGNet
from sklearn.metrics import mean_absolute_error
from sklearn.model_selection import GroupKFold
from torch.utils.data import DataLoader, TensorDataset
1. Load the named cohort and inspect observed participant metadata#
Every window from a participant will share that participant’s observed
p_factor. Repeating the score over windows is a supervised-learning
representation of one measurement, not many independent clinical targets.
Consequently all those windows must remain together during evaluation.
The two assertions require one selected recording per named participant.
description_fields makes the source attributes available alongside BIDS
identifiers; the EEG still comes from EEGDashDataset. These are the original
OpenNeuro recordings, not the separately filtered/downsampled challenge
release supplied by EEGChallengeDataset.
subjects = [
"NDARBH024NH2",
"NDARAM704GKZ",
"NDARAC904DMU",
"NDARAN385MDH",
"NDARAG143ARJ",
"NDARAP359UM6",
]
cache_dir = Path(os.environ.get("EEGDASH_CACHE_DIR", ".eegdash_cache"))
dataset = EEGDashDataset(
dataset="ds005505",
task="RestingState",
subject=subjects,
cache_dir=cache_dir,
description_fields=["subject", "task", "age", "sex", "p_factor"],
n_jobs=1,
)
assert len(dataset.datasets) == len(subjects)
assert dataset.description["subject"].nunique() == len(subjects)
print(dataset.description[["subject", "age", "sex", "p_factor"]])
╭────────────────────── EEG 2025 Competition Data Notice ──────────────────────╮
│ This notice is only for users who are participating in the EEG 2025 │
│ Competition. │
│ │
│ EEG 2025 Competition Data Notice! │
│ You are loading one of the datasets that is used in competition, but via │
│ `EEGDashDataset`. │
│ │
│ IMPORTANT: │
│ If you download data from `EEGDashDataset`, it is NOT identical to the │
│ official │
│ competition data, which is accessed via `EEGChallengeDataset`. The │
│ competition data has been downsampled and filtered. │
│ │
│ If you are participating in the competition, │
│ you must use the `EEGChallengeDataset` object to ensure consistency. │
│ │
│ If you are not participating in the competition, you can ignore this │
│ message. │
╰─────────────────────────── Source: EEGDashDataset ───────────────────────────╯
subject age sex p_factor
0 NDARAC904DMU 11.3386 F -0.603
1 NDARAG143ARJ 7.6648 M -0.258
2 NDARAM704GKZ 10.9449 M 0.062
3 NDARAN385MDH 10.7089 M 0.034
4 NDARAP359UM6 12.8422 F 0.467
5 NDARBH024NH2 8.6883 F -0.818
2. Prepare the first minute of recorded EEG#
The fixed first-minute interval and four named electrodes bound computation and establish a common channel order. They are not selected by test accuracy. The source reference channel can be flat; the chosen channels exclude that reference rather than pretending a zero-variance channel can be standardized.
Braindecode’s EEGPrep Resampling adapter reduces the rate to 100 Hz,
with anti-alias filtering, before RemoveDrifts applies its high-pass
transition from 0.5 to 1 Hz. Resampling first reduces the later filter’s
computation. These published components run on each recording separately;
no custom filtering function or cross-participant fit is needed.
A two-second window then contains 200 voltage samples. The source reference is retained: we do not average-reference an arbitrary four-channel subset. EEGPrep’s forward-backward drift filter is offline, not causal streaming preprocessing. This small recipe neither detects every artifact nor invokes channel rejection or ASR. The neural decoder receives the resulting waveform.
Fixed-length windows are used because the target belongs to a person, not an
event. They do not relabel eyes-open/closed intervals or remove every
instruction or artifact from the first minute. Equal size and stride avoid
overlap; a final short remainder is discarded. The printed array has axes
(windows, 4 channels, 200 samples). Check its actual row count rather than
assuming every recording produces an identical number of windows.
channels = ["E11", "E62", "E75", "E22"]
for recording in dataset.datasets:
raw = recording.raw
print(
recording.description["subject"],
raw.info["sfreq"],
raw.ch_names,
"observed annotations:",
sorted(set(raw.annotations.description)),
)
assert set(channels).issubset(raw.ch_names)
raw.crop(tmax=59.99).load_data().pick(channels).reorder_channels(channels)
# Preserve annotation times in seconds across EEGPrep format conversions.
# Retain the measurement date too: it anchors annotations with absolute times.
annotations_before = [
recording.raw.annotations.copy() for recording in dataset.datasets
]
measurement_dates = [recording.raw.info["meas_date"] for recording in dataset.datasets]
durations_before = [
recording.raw.n_times / recording.raw.info["sfreq"]
for recording in dataset.datasets
]
preprocess(
dataset, [Resampling(sfreq=100), RemoveDrifts(transition=(0.5, 1.0))], n_jobs=1
)
for recording, annotations, duration, measurement_date in zip(
dataset.datasets, annotations_before, durations_before, measurement_dates
):
raw = recording.raw
assert raw.ch_names == channels and raw.info["sfreq"] == 100
assert abs(raw.n_times / 100 - duration) <= 1 / 100
raw.set_meas_date(measurement_date)
raw.set_annotations(annotations)
assert raw.annotations.orig_time == annotations.orig_time
np.testing.assert_array_equal(raw.annotations.onset, annotations.onset)
np.testing.assert_array_equal(raw.annotations.description, annotations.description)
windows = create_fixed_length_windows(
dataset,
window_size_samples=200,
window_stride_samples=200,
on_last_window="drop",
preload=True,
)
metadata = windows.get_metadata().reset_index(drop=True)
X_windows = np.stack([item[0] for item in windows])
groups = metadata["subject"].astype(str).to_numpy()
assert X_windows.shape == (len(metadata), len(channels), 200)
assert np.isfinite(X_windows).all()
assert not metadata.duplicated(["subject", "i_start_in_trial"]).any()
print("Real two-second windows:", X_windows.shape)
NDARAC904DMU 500.0 ['E1', 'E2', 'E3', 'E4', 'E5', 'E6', 'E7', 'E8', 'E9', 'E10', 'E11', 'E12', 'E13', 'E14', 'E15', 'E16', 'E17', 'E18', 'E19', 'E20', 'E21', 'E22', 'E23', 'E24', 'E25', 'E26', 'E27', 'E28', 'E29', 'E30', 'E31', 'E32', 'E33', 'E34', 'E35', 'E36', 'E37', 'E38', 'E39', 'E40', 'E41', 'E42', 'E43', 'E44', 'E45', 'E46', 'E47', 'E48', 'E49', 'E50', 'E51', 'E52', 'E53', 'E54', 'E55', 'E56', 'E57', 'E58', 'E59', 'E60', 'E61', 'E62', 'E63', 'E64', 'E65', 'E66', 'E67', 'E68', 'E69', 'E70', 'E71', 'E72', 'E73', 'E74', 'E75', 'E76', 'E77', 'E78', 'E79', 'E80', 'E81', 'E82', 'E83', 'E84', 'E85', 'E86', 'E87', 'E88', 'E89', 'E90', 'E91', 'E92', 'E93', 'E94', 'E95', 'E96', 'E97', 'E98', 'E99', 'E100', 'E101', 'E102', 'E103', 'E104', 'E105', 'E106', 'E107', 'E108', 'E109', 'E110', 'E111', 'E112', 'E113', 'E114', 'E115', 'E116', 'E117', 'E118', 'E119', 'E120', 'E121', 'E122', 'E123', 'E124', 'E125', 'E126', 'E127', 'E128', 'Cz'] observed annotations: ['break cnt', 'dot_no1_OFF', 'dot_no1_ON', 'dot_no2_OFF', 'dot_no2_ON', 'dot_no3_OFF', 'dot_no3_ON', 'dot_no4_OFF', 'dot_no4_ON', 'dot_no5_OFF', 'dot_no5_ON', 'dot_no6_OFF', 'dot_no6_ON', 'dot_no7_OFF', 'dot_no7_ON', 'dot_no8_OFF', 'dot_no8_ON', 'instructed_toCloseEyes', 'instructed_toOpenEyes', 'learningBlock_1', 'resting_start', 'seqLearning_start']
NDARAG143ARJ 500.0 ['E1', 'E2', 'E3', 'E4', 'E5', 'E6', 'E7', 'E8', 'E9', 'E10', 'E11', 'E12', 'E13', 'E14', 'E15', 'E16', 'E17', 'E18', 'E19', 'E20', 'E21', 'E22', 'E23', 'E24', 'E25', 'E26', 'E27', 'E28', 'E29', 'E30', 'E31', 'E32', 'E33', 'E34', 'E35', 'E36', 'E37', 'E38', 'E39', 'E40', 'E41', 'E42', 'E43', 'E44', 'E45', 'E46', 'E47', 'E48', 'E49', 'E50', 'E51', 'E52', 'E53', 'E54', 'E55', 'E56', 'E57', 'E58', 'E59', 'E60', 'E61', 'E62', 'E63', 'E64', 'E65', 'E66', 'E67', 'E68', 'E69', 'E70', 'E71', 'E72', 'E73', 'E74', 'E75', 'E76', 'E77', 'E78', 'E79', 'E80', 'E81', 'E82', 'E83', 'E84', 'E85', 'E86', 'E87', 'E88', 'E89', 'E90', 'E91', 'E92', 'E93', 'E94', 'E95', 'E96', 'E97', 'E98', 'E99', 'E100', 'E101', 'E102', 'E103', 'E104', 'E105', 'E106', 'E107', 'E108', 'E109', 'E110', 'E111', 'E112', 'E113', 'E114', 'E115', 'E116', 'E117', 'E118', 'E119', 'E120', 'E121', 'E122', 'E123', 'E124', 'E125', 'E126', 'E127', 'E128', 'Cz'] observed annotations: ['boundary', 'break cnt', 'instructed_toCloseEyes', 'instructed_toOpenEyes', 'resting_start']
NDARAM704GKZ 500.0 ['E1', 'E2', 'E3', 'E4', 'E5', 'E6', 'E7', 'E8', 'E9', 'E10', 'E11', 'E12', 'E13', 'E14', 'E15', 'E16', 'E17', 'E18', 'E19', 'E20', 'E21', 'E22', 'E23', 'E24', 'E25', 'E26', 'E27', 'E28', 'E29', 'E30', 'E31', 'E32', 'E33', 'E34', 'E35', 'E36', 'E37', 'E38', 'E39', 'E40', 'E41', 'E42', 'E43', 'E44', 'E45', 'E46', 'E47', 'E48', 'E49', 'E50', 'E51', 'E52', 'E53', 'E54', 'E55', 'E56', 'E57', 'E58', 'E59', 'E60', 'E61', 'E62', 'E63', 'E64', 'E65', 'E66', 'E67', 'E68', 'E69', 'E70', 'E71', 'E72', 'E73', 'E74', 'E75', 'E76', 'E77', 'E78', 'E79', 'E80', 'E81', 'E82', 'E83', 'E84', 'E85', 'E86', 'E87', 'E88', 'E89', 'E90', 'E91', 'E92', 'E93', 'E94', 'E95', 'E96', 'E97', 'E98', 'E99', 'E100', 'E101', 'E102', 'E103', 'E104', 'E105', 'E106', 'E107', 'E108', 'E109', 'E110', 'E111', 'E112', 'E113', 'E114', 'E115', 'E116', 'E117', 'E118', 'E119', 'E120', 'E121', 'E122', 'E123', 'E124', 'E125', 'E126', 'E127', 'E128', 'Cz'] observed annotations: ['boundary', 'break cnt', 'instructed_toCloseEyes', 'instructed_toOpenEyes', 'resting_start']
NDARAN385MDH 500.0 ['E1', 'E2', 'E3', 'E4', 'E5', 'E6', 'E7', 'E8', 'E9', 'E10', 'E11', 'E12', 'E13', 'E14', 'E15', 'E16', 'E17', 'E18', 'E19', 'E20', 'E21', 'E22', 'E23', 'E24', 'E25', 'E26', 'E27', 'E28', 'E29', 'E30', 'E31', 'E32', 'E33', 'E34', 'E35', 'E36', 'E37', 'E38', 'E39', 'E40', 'E41', 'E42', 'E43', 'E44', 'E45', 'E46', 'E47', 'E48', 'E49', 'E50', 'E51', 'E52', 'E53', 'E54', 'E55', 'E56', 'E57', 'E58', 'E59', 'E60', 'E61', 'E62', 'E63', 'E64', 'E65', 'E66', 'E67', 'E68', 'E69', 'E70', 'E71', 'E72', 'E73', 'E74', 'E75', 'E76', 'E77', 'E78', 'E79', 'E80', 'E81', 'E82', 'E83', 'E84', 'E85', 'E86', 'E87', 'E88', 'E89', 'E90', 'E91', 'E92', 'E93', 'E94', 'E95', 'E96', 'E97', 'E98', 'E99', 'E100', 'E101', 'E102', 'E103', 'E104', 'E105', 'E106', 'E107', 'E108', 'E109', 'E110', 'E111', 'E112', 'E113', 'E114', 'E115', 'E116', 'E117', 'E118', 'E119', 'E120', 'E121', 'E122', 'E123', 'E124', 'E125', 'E126', 'E127', 'E128', 'Cz'] observed annotations: ['boundary', 'break cnt', 'instructed_toCloseEyes', 'instructed_toOpenEyes', 'resting_start']
NDARAP359UM6 500.0 ['E1', 'E2', 'E3', 'E4', 'E5', 'E6', 'E7', 'E8', 'E9', 'E10', 'E11', 'E12', 'E13', 'E14', 'E15', 'E16', 'E17', 'E18', 'E19', 'E20', 'E21', 'E22', 'E23', 'E24', 'E25', 'E26', 'E27', 'E28', 'E29', 'E30', 'E31', 'E32', 'E33', 'E34', 'E35', 'E36', 'E37', 'E38', 'E39', 'E40', 'E41', 'E42', 'E43', 'E44', 'E45', 'E46', 'E47', 'E48', 'E49', 'E50', 'E51', 'E52', 'E53', 'E54', 'E55', 'E56', 'E57', 'E58', 'E59', 'E60', 'E61', 'E62', 'E63', 'E64', 'E65', 'E66', 'E67', 'E68', 'E69', 'E70', 'E71', 'E72', 'E73', 'E74', 'E75', 'E76', 'E77', 'E78', 'E79', 'E80', 'E81', 'E82', 'E83', 'E84', 'E85', 'E86', 'E87', 'E88', 'E89', 'E90', 'E91', 'E92', 'E93', 'E94', 'E95', 'E96', 'E97', 'E98', 'E99', 'E100', 'E101', 'E102', 'E103', 'E104', 'E105', 'E106', 'E107', 'E108', 'E109', 'E110', 'E111', 'E112', 'E113', 'E114', 'E115', 'E116', 'E117', 'E118', 'E119', 'E120', 'E121', 'E122', 'E123', 'E124', 'E125', 'E126', 'E127', 'E128', 'Cz'] observed annotations: ['boundary', 'break cnt', 'instructed_toCloseEyes', 'instructed_toOpenEyes', 'resting_start']
NDARBH024NH2 500.0 ['E1', 'E2', 'E3', 'E4', 'E5', 'E6', 'E7', 'E8', 'E9', 'E10', 'E11', 'E12', 'E13', 'E14', 'E15', 'E16', 'E17', 'E18', 'E19', 'E20', 'E21', 'E22', 'E23', 'E24', 'E25', 'E26', 'E27', 'E28', 'E29', 'E30', 'E31', 'E32', 'E33', 'E34', 'E35', 'E36', 'E37', 'E38', 'E39', 'E40', 'E41', 'E42', 'E43', 'E44', 'E45', 'E46', 'E47', 'E48', 'E49', 'E50', 'E51', 'E52', 'E53', 'E54', 'E55', 'E56', 'E57', 'E58', 'E59', 'E60', 'E61', 'E62', 'E63', 'E64', 'E65', 'E66', 'E67', 'E68', 'E69', 'E70', 'E71', 'E72', 'E73', 'E74', 'E75', 'E76', 'E77', 'E78', 'E79', 'E80', 'E81', 'E82', 'E83', 'E84', 'E85', 'E86', 'E87', 'E88', 'E89', 'E90', 'E91', 'E92', 'E93', 'E94', 'E95', 'E96', 'E97', 'E98', 'E99', 'E100', 'E101', 'E102', 'E103', 'E104', 'E105', 'E106', 'E107', 'E108', 'E109', 'E110', 'E111', 'E112', 'E113', 'E114', 'E115', 'E116', 'E117', 'E118', 'E119', 'E120', 'E121', 'E122', 'E123', 'E124', 'E125', 'E126', 'E127', 'E128', 'Cz'] observed annotations: ['boundary', 'break cnt', 'instructed_toCloseEyes', 'instructed_toOpenEyes', 'resting_start']
/tmp/tmpibh9l8qa.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmpu8r00q7h.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmp_z922mod.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmpytcz16um.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmpda8c2q0a.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmpjd46splc.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmpidhhcp_h.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmpxmqy06ve.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmpt_t_2zr7.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmpud2kx7rq.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmpi6qi82h6.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
/tmp/tmp36e4m7j5.set
Field 'subject' is missing from the EEG dictionnary, adding it.
Field 'group' is missing from the EEG dictionnary, adding it.
Field 'condition' is missing from the EEG dictionnary, adding it.
Field 'session' is missing from the EEG dictionnary, adding it.
Field 'comments' is missing from the EEG dictionnary, adding it.
Field 'times' is missing from the EEG dictionnary, adding it.
Field 'icaact' is missing from the EEG dictionnary, adding it.
Field 'icachansind' is missing from the EEG dictionnary, adding it.
Field 'urchanlocs' is missing from the EEG dictionnary, adding it.
Field 'urevent' is missing from the EEG dictionnary, adding it.
Field 'eventdescription' is missing from the EEG dictionnary, adding it.
Field 'epoch' is missing from the EEG dictionnary, adding it.
Field 'epochdescription' is missing from the EEG dictionnary, adding it.
Field 'stats' is missing from the EEG dictionnary, adding it.
Field 'specdata' is missing from the EEG dictionnary, adding it.
Field 'specicaact' is missing from the EEG dictionnary, adding it.
Field 'splinefile' is missing from the EEG dictionnary, adding it.
Field 'icasplinefile' is missing from the EEG dictionnary, adding it.
Field 'dipfit' is missing from the EEG dictionnary, adding it.
Field 'history' is missing from the EEG dictionnary, adding it.
Field 'saved' is missing from the EEG dictionnary, adding it.
Field 'etc' is missing from the EEG dictionnary, adding it.
Field 'datfile' is missing from the EEG dictionnary, adding it.
Field 'run' is missing from the EEG dictionnary, adding it.
Field 'roi' is missing from the EEG dictionnary, adding it.
Real two-second windows: (180, 4, 200)
3. Attach observed participant targets and define subject-disjoint folds#
The metadata join maps each window’s subject to its released score. That join is preferable to relying on recording order, which can change after filtering or concatenation. Missing/non-numeric targets stop the workflow. The random seed controls network initialization and minibatch shuffling; it does not create EEG, labels or a favourable target relationship.
participant_targets = dataset.description.set_index("subject")["p_factor"]
y = pd.to_numeric(
metadata["subject"].map(participant_targets), errors="raise"
).to_numpy(dtype=np.float32)
assert np.isfinite(y).all()
torch.manual_seed(42)
torch.set_num_threads(1)
predictions = np.full(len(y), np.nan, dtype=np.float32)
baseline = np.full(len(y), np.nan, dtype=np.float32)
4. Fit signal scaling, target scaling and network inside each fold#
Three grouped folds reserve two participants at a time and train on the other four. Per-channel mean and standard deviation use only training windows and samples. Subtracting/dividing these arrays broadcasts over windows and time, producing dimensionless float32 network inputs. Target normalization uses one score per training participant, not one repeated value per window. The held-out scores never enter either normalization.
EEGNet receives four channels, 200 time samples and one continuous output.
F1=4 and D=2 keep the temporal/spatial filter bank small;
kernel_length=32 corresponds to 0.32 seconds at 100 Hz. These fixed choices
are for a short CPU demonstration. Batch size 16 and two epochs expose the
training operations without claiming that the network has converged.
For each batch, zero old gradients, compute a prediction and squared-error
loss on normalized targets, backpropagate, then update weights with Adam.
The printed batch-mean training MSE measures optimization in normalized
units; it is neither participant MAE nor held-out performance. eval()
switches dropout and normalization layers to inference behavior, while
no_grad() disables gradient recording. Both are needed for a clear
evaluation path; they serve different purposes.
Finally, reverse the training target normalization so test predictions are back on the original score scale. The same fold’s training mean supplies the non-EEG reference prediction.
for fold, (train, test) in enumerate(
GroupKFold(n_splits=3).split(X_windows, y, groups)
):
assert set(groups[train]).isdisjoint(groups[test])
mean = X_windows[train].mean(axis=(0, 2), keepdims=True)
scale = X_windows[train].std(axis=(0, 2), keepdims=True)
assert (scale > 0).all()
X = ((X_windows - mean) / scale).astype(np.float32)
training_targets = participant_targets.loc[sorted(set(groups[train]))].astype(float)
target_mean, target_scale = training_targets.mean(), training_targets.std(ddof=0)
assert target_scale > 0
train_loader = DataLoader(
TensorDataset(
torch.from_numpy(X[train]),
torch.from_numpy(
((y[train] - target_mean) / target_scale).astype(np.float32)
),
),
batch_size=16,
shuffle=True,
)
model = EEGNet(
n_chans=len(channels), n_outputs=1, n_times=200, F1=4, D=2, kernel_length=32
)
optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)
for epoch in range(2):
model.train()
losses = []
for signals, targets in train_loader:
optimizer.zero_grad()
output = model(signals).reshape(-1)
loss = torch.nn.functional.mse_loss(output, targets)
assert torch.isfinite(loss)
loss.backward()
optimizer.step()
losses.append(loss.item())
print(f"Fold {fold}, epoch {epoch + 1}, training MSE {np.mean(losses):.4f}")
model.eval()
with torch.no_grad():
predictions[test] = (
model(torch.from_numpy(X[test])).reshape(-1).numpy() * target_scale
+ target_mean
)
baseline[test] = target_mean
assert np.isfinite(predictions).all()
Fold 0, epoch 1, training MSE 1.0096
Fold 0, epoch 2, training MSE 0.9398
Fold 1, epoch 1, training MSE 1.0977
Fold 1, epoch 2, training MSE 0.9613
Fold 2, epoch 1, training MSE 0.9861
Fold 2, epoch 2, training MSE 0.9736
5. Score one prediction per held-out participant#
Average the window predictions within each person, then compute MAE across those six people. Scoring all windows directly would treat repeated target measurements as independent observations and give extra weight to longer recordings. Here the target is constant within each group, so its grouped mean is still the original observed score.
A prediction near the training mean can indicate that the short training run learned little person-specific information. A small MAE difference from the mean baseline on six people is not evidence of a reliable biomarker. The scatter plot and per-participant table show the actual outputs rather than a prescribed learning curve or minimum score.
results = (
pd.DataFrame(
{
"subject": groups,
"observed": y,
"predicted": predictions,
"training_mean": baseline,
}
)
.groupby("subject")
.mean()
)
assert len(results) == len(subjects)
print(results)
print("Participant MAE:", mean_absolute_error(results.observed, results.predicted))
print(
"Training-mean MAE:", mean_absolute_error(results.observed, results.training_mean)
)
fig, ax = plt.subplots(figsize=(5, 4), layout="constrained")
ax.scatter(results.observed, results.predicted)
limits = [results.observed.min(), results.observed.max()]
ax.plot(limits, limits, "k--")
ax.set(xlabel="Observed p-factor", ylabel="Held-out participant prediction")
plt.show()
observed predicted training_mean
subject
NDARAC904DMU -0.603 -0.132672 -0.13675
NDARAG143ARJ -0.258 -0.320921 -0.33125
NDARAM704GKZ 0.062 -0.087971 -0.09000
NDARAN385MDH 0.034 -0.130850 -0.13675
NDARAP359UM6 0.467 -0.324978 -0.33125
NDARBH024NH2 -0.818 -0.089630 -0.09000
Participant MAE: 0.3947364389896393
Training-mean MAE: 0.398083359003067
Develop a model using validation participants#
To choose a longer epoch budget, reserve validation people inside each training fold and select the epoch there. Keep the outer test participants untouched until that choice is fixed. The feature-based p-factor project provides a useful simpler model, but its leave-one-out folds differ from these three grouped folds; use identical folds before making a paired model comparison.
Total running time of the script: (0 minutes 3.147 seconds)
