Documentation

Ecosystem and Scope

This page exists to keep the package honest. EchoTime should be easy to place next to sktime, tsfresh, DTAIDistance, STUMPY, and other neighboring tools.

Where EchoTime fits in the ecosystem

Package	Family	Strongest fit	How EchoTime fits
tsfresh	feature extraction	large handcrafted feature extraction, feature filtering, series-as-features pipelines	EchoTime is not a feature-zoo replacement; it is explainable structural similarity. Use EchoTime when the first question is 'what kind of temporal dataset is this?' and tsfresh when you need a broad handcrafted feature matrix for downstream supervised learning.
aeon	time-series machine learning toolkit	estimators and pipelines, benchmarking, forecasting / classification / regression / clustering	EchoTime sits earlier in the workflow: dataset characterization, ontology axes, summary cards, and compact context for collaborators or agents. aeon is the modelling-and-evaluation toolbox you use once you already know what sort of problem you have.
sktime	unified time-series ML framework	forecasting pipelines, composite estimators, time-series transformations	EchoTime is for structural triage, communication, and dataset cards; sktime is for building, tuning, and evaluating models.
tslearn	time-series ML toolkit	time-series clustering, classification and regression, distance-based learning	EchoTime is not primarily a learner library. It focuses on understanding and comparing datasets before or beside modelling, while tslearn focuses on machine-learning algorithms over time-series objects.
DTAIDistance	distance / alignment	DTW distance, alignment paths, pairwise distance matrices	EchoTime does include high-level similarity summaries, but it is not a dedicated DTW engine. Use DTAIDistance when alignment paths, distance matrices, or lower-level DTW controls are the main deliverable.
STUMPY	matrix profile / subsequence mining	motif discovery, subsequence anomaly search, matrix-profile workflows	EchoTime is not a matrix-profile or subsequence-mining library. It helps you profile whole datasets and summarize temporal structure at the dataset or pairwise-comparison level.
Darts	forecasting / anomaly detection	forecasting models, backtesting, anomaly detection	EchoTime does not train forecasting models. It helps decide whether the data look trend-dominant, rhythmic, bursty, nonstationary, heterogeneous, or irregular before forecasting choices are made.
Kats	general time-series analysis toolkit	forecasting, detection, TSFeatures	Kats spans multiple analysis tasks; EchoTime is narrower and more explicit about dataset ontology, structural profiling, agent context, and plain-language dataset reports.

Capability coverage

Capability	Role	Companion packages	Notes
Dataset structure and similarity context	Primary	-	EchoTime's main job is to turn a dataset into ontology axes, archetypes, reliability summaries, and task hints.
Dataset card JSON / Markdown	Primary	-	Useful for benchmark curation, cross-team handoff, and plain-language documentation.
Plain-language summary card and narrative report	Primary	-	Built for non-method collaborators, dataset owners, and cross-disciplinary teams.
Explicit agent-driving and compact context	Primary	-	Helps an LLM or agent choose a lean path and emit a compact reusable context bundle.
Raw feature extraction matrix	Complementary	tsfresh, Kats TSFeatures	EchoTime intentionally avoids becoming a giant feature zoo.
Forecasting models and backtesting	Out of scope	Darts, sktime, aeon, Kats	Use EchoTime before or beside forecasting, not instead of it.
Classification / regression / clustering estimators	Out of scope	aeon, tslearn, sktime	EchoTime profiles datasets and compares trajectories; it does not train supervised estimators.
DTW engine and alignment paths	Complementary	DTAIDistance	EchoTime surfaces similarity summaries; DTAIDistance is for distance computation and warping control.
Subsequence motif discovery / matrix profile	Out of scope	STUMPY	EchoTime works at dataset and whole-series levels, not matrix-profile subsequence mining.
Irregular, event-stream, and longitudinal typed wrappers	Primary	pandas, xarray, MNE, nilearn, clinical ETL pipelines	The goal is to preserve observation semantics before analysis rather than silently flattening them away.