Build VneuroTK neural data#

BaseData represents neural recordings while NeuroData is its neural-array container. NeuroData is not an ndarray subclass; use .data for the underlying array or numpy.asarray() for array conversion.

This notebook uses deterministic synthetic arrays and requires only the core package.

Choose an explicit data mode#

Mode

Raw shape

Trial structure

continuous

(n_samples, n_channels)

Configure onsets before using .epochs

epochs

(n_trials, n_timebins, n_channels)

Already trial-structured

patterns

(n_rows, n_channels)

Aggregated rows

Use the matching factory whenever a 2-D array’s meaning would otherwise be ambiguous.

import numpy as np

import vneurotk as vtk

continuous = vtk.BaseData.for_continuous(
    neuro=np.arange(80, dtype=float).reshape(20, 4),
    neuro_info={"ch_names": ["a", "b", "c", "d"], "sfreq": 10.0},
)
epochs = vtk.BaseData.for_epochs(
    neuro=np.arange(96, dtype=float).reshape(3, 8, 4),
    neuro_info={"ch_names": ["a", "b", "c", "d"], "sfreq": 10.0},
)
patterns = vtk.BaseData.for_patterns(
    neuro=np.arange(24, dtype=float).reshape(6, 4),
    neuro_info={"ch_names": ["a", "b", "c", "d"]},
)
continuous.data_mode, epochs.data_mode, patterns.data_mode

('continuous', 'epochs', 'patterns')

Inspect the neural container#

neuro = patterns.neuro
neuro.shape, neuro.dtype, np.asarray(neuro).shape, neuro.data.shape

((6, 4), dtype('float64'), (6, 4), (6, 4))

Configure continuous trial structure#

stim_ids = np.array(["image-1", "image-2", "image-1"])
continuous.configure(
    vision_onsets=np.array([2, 8, 14]),
    stim_ids=stim_ids,
    vision_db={
        "image-1": np.zeros((8, 8, 3), dtype=np.uint8),
        "image-2": np.full((8, 8, 3), 255, dtype=np.uint8),
    },
    trial_window=[-1, 3],
)
continuous.neuro.epochs.shape

(3, 4, 4)

configure() binds stimulus IDs, onset samples, a trial window, and an image database. Pattern rows require trial_meta["stim_index"] only when they need alignment with vision features.

Read recording sources#

from pathlib import Path

from vneurotk.io import EphysPath, MNEPath

# These examples describe inputs; reading requires the matching local files.
source_root = Path("data")
mne_source = MNEPath(
    root=source_root,
    subject="01",
    session="ImageNet01",
    task="ImageNet",
    run="01",
    suffix="meg_clean",
    extension=".fif",
)
ephys_source = EphysPath(
    root=source_root,
    session_id="251024_FanFan_nsd1w_MSB",
    dtype="TrialRaster",
    extension="h5",
)
# meg_data = vtk.read(mne_source)
# ephys_data = vtk.read(ephys_source)

Electrophysiology products are configured by their loaders. MNE recordings remain lazy until neural values are accessed and need configure() before trial-aligned views are available.