Glossary

Key concepts in the neuralset pipeline, from data source to batched tensors. See the tutorials for how they fit together.

---

Chain

Sequences Steps — each step’s output feeds the next. Auto-created when a list is assigned to a Step-typed field. Resumes from the latest cached intermediate result. → Chain · Tutorial

Event

Something that happens in time: a stimulus, a neural recording, a word. Core fields: start, duration, timeline. Concrete subclasses (Meg, Image, Word, Fmri, …) add modality-specific fields. Events are collected into a pandas DataFrame for bulk processing. → Event · Tutorial

EventsTransform

A processing step that takes an events DataFrame and returns a modified one (filtering, chunking, linking word context, …). Subclasses override _run(events) -> events. → EventsTransform · Tutorial

Extractor

Converts events into numeric arrays for a given time window — e.g. MegExtractor returns MEG time-series at a given frequency. → BaseExtractor · Tutorial

Frequency

A float subclass with helpers for converting between seconds and sample indices (to_ind, to_sec). → Frequency

infra

Parameter controlling disk caching and cluster execution. Its type varies by component: Backend on Steps and Chains (whole-step caching), MapInfra on extractors and Study.infra_timelines (per-item caching and batch dispatch), TaskInfra on neuraltrain experiments (full computation and job arrays). · Caching & Cluster Execution

prepare

Method on extractors that precomputes results for all events in one pass, triggering disk caching and optional cluster dispatch. Called automatically by Segmenter.apply(). · Caching & Cluster Execution

Segment

A time window (start, duration) within a Timeline, anchored to a Trigger. Defines which slice of data each Extractor reads. → Segment · Tutorial

SegmentDataset

A torch Dataset that pairs segments with extractors. Each __getitem__ reads data for one segment across all extractors. → SegmentDataset · Tutorial

Segmenter

Creates segments from an events DataFrame and wraps them into a SegmentDataset. Configured with a trigger query, a time window, and a dict of extractors. → Segmenter · Tutorial

Step

A composable pipeline unit — a pydantic model with a _run() method and an optional infra parameter for caching and cluster execution. Base class for Study, EventsTransform, and Chain. → Step · Caching & Cluster Execution

Study

An interface to an external dataset that knows how to iterate timelines and load events from raw data files. Subclasses implement iter_timelines() and _load_timeline_events(). No data is bundled — each study points to an externally hosted repository. All studies can share a single path — each resolves its own subfolder automatically. → Study · Tutorial

TimedArray

A numpy array with time metadata (frequency, start, duration). Used by dynamic extractors for time-aligned slicing. → TimedArray

Timeline

A recording session or run within a Study, typically identified by BIDS entities (subject, session, task, run). Study.iter_timelines() yields one dict per timeline; events loaded from it share that timeline identifier.

Trigger

The Event that anchors a Segment — for example a word onset or an image presentation. Selected by the trigger_query parameter of the Segmenter.

← Encoding & Decoding Contributing →