Glossary#

The glossary below defines common terms and API elements used throughout TanaT.

Core Concepts#

Entity: A description of something that happens for an individual during a certain timestamp. An entity represents a single temporal event or state with associated features and temporal information.
Individual: A unique subject or unit of observation in the dataset. Each individual can have multiple sequences and trajectories associated with them over time.
Sequence or Timed Sequence: A longitudinal representation of a collection of events that occurs to an individual along its lifespan. Sequences are ordered collections of entities of the same type for one individual.
Trajectory: A collection of multiple sequences for one individual, possibly with additional static features. Trajectories provide a comprehensive view of an individual’s temporal evolution across different dimensions.
Pool: A collection of sequences or trajectories across multiple individuals. Pools enable batch processing and analysis of temporal data from multiple subjects.
Static Feature: Time-invariant characteristics of an individual that remain constant throughout the observation period. Examples include demographic information, genetic markers, or baseline measurements.

Event Sequence: A sequence of point-in-time events where each entity represents something that happened at a specific moment. Events have no duration and are characterized by their timestamp and associated features.
State Sequence: A sequence representing continuous states over time, where each entity has a duration and states do not overlap. State sequences provide complete temporal coverage with non-overlapping intervals.
Interval Sequence: A sequence of events with duration that can potentially overlap in time. Each entity represents an interval with start and end times, allowing for concurrent events.

Interval: A period of time defined by start and end timestamps. Intervals can represent durations of states, events, or any temporal phenomena with extent.
Timestamp: A specific point in time when an event occurs or a measurement is taken. Timestamps provide the temporal ordering for sequences.
Temporal Extent: The time period covered by a sequence or trajectory, from the earliest to the latest timestamp.
Granularity: The level of temporal precision used in the analysis, such as days, hours, or minutes.

Clustering: A learning task focused on discovering groups consisting of instances with similar timed sequences or trajectories. TanaT provides specialized clustering algorithms for temporal data.
Distance Metric: A function that quantifies the similarity or dissimilarity between two sequences or trajectories. TanaT implements various metrics adapted for temporal sequence analysis.
Criterion: Filtering or selection rules applied to sequences, trajectories, or entities based on temporal, pattern-based, or feature-based conditions.
Aggregation: The process of combining multiple sequences or metrics into summary statistics or reduced representations.

Pool Structure: The organizational framework for managing collections of sequences or trajectories, providing efficient access and manipulation of temporal data across multiple individuals.
Metadata: Additional information about sequences and entity feature(s) that describes their properties, type, etc.

Loader: Components responsible for reading and importing temporal data from various sources and formats into TanaT’s internal data structures.
Simulation: Tools for generating synthetic temporal sequences and trajectories for testing, validation, or augmentation purposes.
Visualization: Methods and tools for creating graphical representations of temporal sequences, trajectories, and analysis results.
Zeroing: The process of aligning temporal sequences to a common reference point (T0 or index date) or normalizing temporal coordinates for comparative analysis.

Survival Analysis: Statistical methods for analyzing time-to-event data, including censoring and hazard modeling, integrated with TanaT’s temporal sequence framework.
Pattern Mining: The discovery of recurring temporal patterns, motifs, or subsequences within individual sequences or across multiple sequences in a pool.
Temporal Alignment: The process of synchronizing sequences or trajectories to enable meaningful comparison and analysis across different individuals or time periods.
Feature Engineering: The creation of derived features from temporal sequences, such as statistical summaries, pattern indicators, or temporal transformations.