EchoTime Explainable time-series similarity for humans and agents.
Documentation

Scenarios

Use this page the way scikit-learn users use problem-oriented documentation: find your domain, then jump into the right entrypoint.

Resting-state or task fMRI cohort profiling
Domains fmriEnvironments notebook, python_script, neuro_stack, ml_benchmark

Data shape: subjects × time × ROI, ROI-wise tables, or xarray-like neuroimaging containers

Where EchoTime helps: Acts as a dataset-card and structure-audit layer before graph modelling, connectome analyses, or benchmark comparison.

Typical inputs
  • FMRIInput
  • xarray-like object
  • 3D NumPy array
What you can do
  • Compare datasets or cohorts structurally before training models.
  • Summarize how networked, heterogeneous, or low-frequency-dominated a dataset is.
  • Export a dataset card for benchmark documentation or supplement material.
Caveats
  • Provide TR whenever possible for Hz-aware metrics.
  • EchoTime does not replace neuroimaging preprocessing or connectome estimation packages.
EEG or electrophysiology recording triage
Domains eegEnvironments notebook, python_script, neuro_stack, ml_benchmark

Data shape: time × channel arrays, subject × time × channel cohorts, or MNE-like objects

Where EchoTime helps: Provides a fast structural summary layer before decoding, spectral analysis, or representation learning.

Typical inputs
  • EEGInput
  • MNE Raw/Epochs/Evoked-like object
  • 2D/3D arrays
What you can do
  • Spot low-rhythmicity or high-noise recordings before downstream analysis.
  • Compare cohorts by bandpower-heavy vs noise-heavy structure.
  • Generate reproducible cards for datasets used in decoding benchmarks.
Caveats
  • Sampling rate is important for EEG-specific proxies.
  • The package profiles structure; it does not perform artifact rejection or source reconstruction.
Irregular clinical monitoring or sparse hospital telemetry
Domains clinicalEnvironments notebook, python_script, pandas_pipeline, cli_batch

Data shape: subject-wise irregular observations, asynchronous channels, long hospital tables, or CSV/parquet extracts

Where EchoTime helps: Audits observation irregularity, eventness, drift, and cohort heterogeneity before modelling or cohort QC.

Typical inputs
  • IrregularTimeSeriesInput
  • pandas DataFrame
  • CSV/parquet path
  • list of record dicts
What you can do
  • Decide whether timestamp irregularity is a primary modelling concern.
  • Document missingness and follow-up instability for cohort reports.
  • Feed dataset cards into benchmarking or data-governance workflows.
Caveats
  • For very sparse data, interpret frequency-aware metrics conservatively.
  • EchoTime does not impute, resample, or fit clinical prediction models for you.
Wearable or digital biomarker longitudinal cohort
Domains wearable, clinicalEnvironments notebook, python_script, pandas_pipeline, ml_benchmark, cli_batch

Data shape: long tables with subject / visit / time / channel / value columns or wide wearable frames

Where EchoTime helps: Profiles adherence, repeated-visit instability, subject fingerprintability, and cohort heterogeneity for longitudinal studies.

Typical inputs
  • pandas DataFrame
  • parquet/CSV path
  • list of records
What you can do
  • Understand whether your cohort is dominated by dropout, visit imbalance, or subject-specific structure.
  • Decide whether leave-subject-out or leave-visit-out validation is more defensible.
  • Generate methods-ready summaries for study documentation.
Caveats
  • Parquet reading requires a parquet engine such as pyarrow at runtime.
  • The package reports longitudinal structure; it does not replace biostatistical mixed-effects analysis.
Environmental or industrial multichannel sensor datasets
Domains genericEnvironments notebook, python_script, pandas_pipeline, ml_benchmark, cli_batch

Data shape: dense or mildly gappy multichannel series, station networks, or machine telemetry tables

Where EchoTime helps: Provides trend, rhythmicity, drift, coupling, and noise summaries before forecasting or anomaly pipelines.

Typical inputs
  • 2D/3D arrays
  • pandas DataFrame
  • xarray-like object
What you can do
  • Compare stations, machines, or datasets structurally.
  • Screen for seasonality-heavy vs drift-heavy problems.
  • Produce dataset cards for shared forecasting benchmarks.
Caveats
  • EchoTime does not replace forecasting model selection or domain simulators.
  • If timestamps are absent, irregularity diagnostics are naturally weaker.
Sparse event streams, alerts, clicks, or treatment logs
Domains generic, clinicalEnvironments notebook, python_script, pandas_pipeline, cli_batch

Data shape: timestamped event records with optional event type, channel, subject, and mark/value columns

Where EchoTime helps: Quantifies burstiness, event diversity, and dataset-level event archetypes before point-process or event modelling.

Typical inputs
  • EventStreamInput
  • long tables
  • JSON/JSONL/CSV records
What you can do
  • Differentiate bursty alarm streams from more regular event streams.
  • Summarize whether event labels are diverse or dominated by one code.
  • Create machine-readable cards for event benchmark datasets.
Caveats
  • Sparse event streams are represented generically; detailed point-process inference remains out of scope.
  • Interpret time-series axes together with event plugin metrics rather than in isolation.
Benchmark curation and dataset-card generation
Domains generic, fmri, eeg, clinical, wearableEnvironments python_script, cli_batch, ml_benchmark

Data shape: any dataset that EchoTime can adapt

Where EchoTime helps: Acts as a repeatable structural profiler and card generator for dataset governance and benchmark transparency.

Typical inputs
  • anything accepted by profile_dataset
What you can do
  • Track structural coverage across benchmark suites.
  • Attach cards to releases, papers, or internal registries.
  • Compare new datasets against existing benchmark structure profiles.
Caveats
  • A dataset card is descriptive, not a substitute for task-specific evaluation.
  • Reliability scores should be reported alongside the axes.