biosigIO 1.1.7

biosigIO 1.1.7

Typed read-error hierarchy (biosigio.exceptions)

biosigIO previously raised a generic ValueError for every read failure, so a
caller could not tell why a recording failed to load — a trial-averaged
derivative, a truncated file, and an unsupported format were indistinguishable.

This release adds a typed hierarchy so callers can classify the failure and
surface a specific reason (the NEMAR Zarr viewer uses this to explain why a
recording has no viewer instead of a blank "not available"):

• BiosigIOError — base; every error carries a stable .code.
• UnsupportedFormatError (unsupported_format)
• FileReadError (file_read_error) — generic fallback for a recognized format
  that still couldn't be read.
• NotContinuousRecordingError (not_continuous) — evoked *-ave.fif / epoched
  *-epo.fif derivatives: valid data, but no continuous time series.
• CorruptFileError (corrupt_or_truncated) — truncated/non-compliant files,
  incomplete CTF .meg4, or an incomplete split chain.
• EmptyRecordingError (empty_recording)
• MixedSamplingRateError (mixed_sampling_rate)
• classify_read_error(exc, filepath) maps a low-level MNE/pyedflib failure to
  the right type (shared by the importers).
• REASONS: code → default user-facing copy.

The MEG, EDF, BrainVision, and EEGLAB importers now raise these typed errors;
_infer_importer raises UnsupportedFormatError for unknown extensions.

Backward compatibility

BiosigIOError subclasses ValueError, so existing except ValueError /
pytest.raises(ValueError) callers keep working unchanged. The EDF mixed-rate
policy error is now MixedSamplingRateError (still a ValueError, same message).

Tests

Real-file verification (no mocks) for each NEMAR failure mode: a real MNE evoked
-ave.fif → not_continuous, an incomplete split-FIF chain → corrupt, a truncated
CTF .meg4 → corrupt, a truncated EDF → corrupt, plus classifier units and
back-compat assertions.

biosigIO 1.1.6

biosigIO 1.1.6

First release since v1.1.3 (1.1.4 and 1.1.5 were internal version bumps, folded in here).

MEG import: CTF and KIT, alongside FIF

• CTF/VSM .ds directories now import via read_raw_ctf.
• KIT/Yokogawa/RICOH .con / .sqd / .kdf now import via read_raw_kit.
• Recording.from_file dispatches all three (plus .fif) to the MEG importer.

Mixed-rate EDF/BDF (#737)

• EDFImporter / Recording.from_file now error by default when an EDF/BDF
  mixes sampling rates across channels (e.g. PSG: EEG at 200 Hz + SpO2 at
  12.5 Hz), instead of failing obscurely.
• Opt in with mixed_rate="resample" to upsample slower channels onto the
  fastest channel's grid for a plotting/ML serving copy. Uses
  resample_poly(padtype="line") to avoid edge ringing; the original per-channel
  rate is recorded in metadata.

Streaming Zarr export

• New stream_to_zarr converts arbitrarily large recordings in bounded
  memory: lazy MNE read → channel-major float32 memmap → per-channel
  resample/quantize/write. Output is byte-equivalent to the in-memory exporter,
  so multi-GB recordings convert without changing the result.

Tests and fixtures

• Real vendored fixtures: KIT .sqd and CTF .ds (catch-alp-good-f,
  244 ch incl. simultaneous EEG, 1250 Hz).
• Regression tests for FIF + multi-file split-FIF chains, CTF streaming, and the
  mixed-rate EDF paths. No mocks.

Fixes

• Preserve importer-loaded events when an events.tsv is unparsable (no onset
  column / missing forced column) instead of wiping them.
• Log non-benign MEG find_events failures rather than swallowing them.

v1.1.3

biosigIO 1.1.3

EEGLAB importer robustness

• Load the standard single-EEG-struct save form that real EEGLAB writes (loadmat returns {'EEG': struct}), in addition to the legacy flat layout. Previously a real .set could import as an empty recording (#100).

Citation and DOI

• Add CITATION.cff (CFF 1.3.0, structured affiliation with ORCID 0000-0001-5557-259X and the SCCN ROR 01bt2qm76) and .zenodo.json, so this release is archived to Zenodo with a DOI and the repository shows a "Cite this repository" widget. .zenodo.json cross-links the PyPI project.

Packaging

• Broaden the description and keywords to the full biosignal scope (EEG/EMG/MEG/iEEG; EDF/BDF, EEGLAB, BrainVision, FIF, and the Zarr serving store).

Documentation

• Document the EEGLAB two-save-form robustness and the Zarr store's forward-compatible extended attributes.

Full changelog: v1.1.2...v1.1.3

v1.1.2

EEGLAB importer now unwraps the single EEG struct that scipy.io.loadmat returns for real .set files, so they load with signals/channels instead of silently producing an empty Recording.
Robust to both the nested and flat save forms. Unblocks the NEMAR Zarr serving-copy pipeline.

v1.1.1

biosigio 1.1.1

Bound peak memory for large-file Zarr conversion so multi-GB recordings fit a free 16 GB ubuntu-latest runner. No API or behavior change. Closes #95.

Fixed

• Zarr exporter streaming — per (modality, rate) group the exporter no longer builds a float64 list of every channel plus a float64 vstack copy plus the int16 copy at once (~4.5x the output). It now preallocates the output-dtype array and resamples + quantizes one channel at a time straight into it, dropping the float64 double-buffer. Round-trip output is bit-identical.
• EEGLAB one-shot frame build — the signal DataFrame is built in a single allocation (samples x channels) instead of one column at a time, avoiding the O(n_channels) block-manager reallocation that roughly doubled peak RAM for high-channel-count .fdt recordings; the .fdt's native float32 is preserved (no silent upcast).

Measured: a 5 GB EEGLAB .set+.fdt (346ch x 3.6M @ 500 Hz) now converts at ~10 GB peak RSS (load 5.5 GB + export), down from a fragmented multi-GB path; round-trip events and rate-group splitting unchanged. New test_zarr_memory.py guards the export's marginal peak via a subprocess ru_maxrss check.

v1.1.0

biosigio 1.1.0

Two importer features, found while validating real BIDS -> Zarr conversion.

Added

• EEGLAB .fdt support — .set files whose signal matrix lives in a separate float32 .fdt (EEGLAB's default for large recordings, where EEG.data holds the .fdt filename) now load. The sibling .fdt is resolved by the .set path, so BIDS-renamed files load even though EEG.data keeps the original .fdt name. Unblocks multi-GB .set recordings.
• bids.apply_events_tsv — load the authoritative BIDS _events.tsv into rec.events (onset, duration with n/a->0.0, description from trial_type then value per row), the symmetric counterpart of apply_channels_tsv. Carries events through to EDF+/Parquet/Arrow/Zarr export for files whose native markers are empty.

No public API or behavior change to existing paths. Closes #94.

v1.0.3

biosigio 1.0.3

Internal naming cleanup: the legacy emg local variable (a leftover from the EMG-only era, retired as the EMG class alias was removed in 1.0.0) is renamed to rec throughout the package, examples, and docs so variable names match the modality-agnostic Recording API. Domain-descriptive names that genuinely denote EMG data (emg_only, emg_channels, emg_subset, emg_data, subset_emg, bicep_emg) are preserved.

Changed

• Renamed the emg/new_emg locals to rec/new_rec across importers, exporters, visualization, and the Recording core (945 replacements over 74 files).
• Renamed the MNE helper raw_to_emg → raw_to_recording (definition + MEG/BrainVision callers).
• Renamed visualization parameters emg_object/emg_original/emg_reloaded → rec_object/rec_original/rec_reloaded.
• Updated README, all docs examples, and the API reference to the rec convention.

No public API or behavior change: the Recording class, channel_type="EMG" modality, channel labels, and all method signatures are unchanged. Closes #89.

1.0.2

EEGLAB events fix. No breaking changes.

Fixed

• EEGLAB .set events now load into the standard Recording.events DataFrame (onset/duration in seconds, description = the EEGLAB event type), converting the 1-based sample latency via (latency - 1) / srate, instead of being stashed as a list of dicts in metadata['events']. This makes EEGLAB consistent with the EDF+/BDF+ and WFDB importers, so EEGLAB events are now carried by the EDF/Parquet/Arrow/Zarr exporters. A present-but-zero srate now falls back to 1000 Hz.

Internal

• The EEGLAB importer's local emg variable was renamed to rec (the broader emg→rec cleanup is tracked in #89).

Docs: https://neuromechanist.github.io/biosigio/

1.0.1

Documentation accuracy pass + small Recording convenience accessors. No breaking changes.

Added

• Recording.get_n_channels(), get_n_samples(), get_sampling_frequency() (raises on mixed per-channel rates), get_duration() (total window length in seconds), has_metadata(key).

Docs

• Complete docs-vs-code revision (driven by an in-house multi-agent audit + a Codex pass): removed examples using nonexistent methods/kwargs, corrected stale claims (EDF+/BDF+ annotations load into events and embed on export; EDF/BDF export rejects mixed sampling rates; OTB/EEGLAB/XDF specifics; auto-detect extensions; Trigno structure; _channels.tsv naming; optional extras; version), and added API pages for the CLI, modality vocabulary, BIDS helpers, serialization schema, and signal-analysis module.

CI

• Workflows path-filtered: docs-only changes skip the Python test matrix, and code-only changes skip the docs build.

Docs: https://neuromechanist.github.io/biosigio/

1.0.0

biosigIO 1.0.0 — the package is renamed emgio -> biosigio and is now the multimodal biosignal I/O library (EEG, EMG, iEEG, MEG, and behavioral/marker streams).

Breaking changes

• Package renamed: import emgio -> import biosigio (pip install biosigio). The emgio project on PyPI is frozen at 0.6.0; a transitional emgio shim release will follow.
• EMG alias removed: use Recording (it was a deprecated alias since 0.5.0).

Highlights since 0.x

• Importers: EEGLAB, Trigno, OTB, EDF/BDF(+), WFDB, XDF, MEG (.fif/CTF) and BrainVision via MNE, and proprietary electrophysiology (Intan/Blackrock/Spike2/Plexon/Micromed/Neuralynx) via python-neo.
• Exporters: EDF/BDF(+) with auto format selection, Parquet + Arrow/Feather (lossless), and a Zarr serving store (viewing + inference + training from one derived store).
• Modality-agnostic Recording class, anti-aliased resample, blocking ty type-check gate, and a full docs site.

Install

pip install biosigio              # core
pip install 'biosigio[all]'       # + meg, arrow, neo, zarr, dev, docs

Docs: https://neuromechanist.github.io/biosigio/