Choose a useful resolution
Compare candidate tensors by retained information, structure, sparsity, and computational cost before committing to a grid.
The software
Turn data representation into an explicit decision.
Scientific data are routinely binned, gridded, compressed, or reorganized before analysis. Those choices can remove real detail—or create an impression of structure that was not present in the observations.
TensIT measures raw observations and tensor representations on comparable terms. It keeps information retention, spatial and temporal structure, occupancy, and computational cost separate so the tradeoffs stay visible.
Information measures reveal statistical dependence and predictive relationships. They do not, by themselves, prove physical causation.
Find hidden relationships
Expose nonlinear dependence, redundant attributes, unique contributions, and combinations that only become informative together.
Follow change through time
Scan lags, compare directions, and track local or windowed information as relationships strengthen, weaken, or shift.
Keep uncertainty visible
Use bootstrap intervals, permutation tests, stability analysis, and multiple-comparison controls to distinguish durable patterns from noise.
A transparent workflow
Compare first. Choose second.
TensIT preserves the individual diagnostics behind a recommendation, so scientists can see why one representation is preferred and when a different constraint should change that choice.
- 01
Raw observations
- 02
Candidate representations
- 03
Information diagnostics
- 04
Defensible choice
Built for reproducible scientific work
- Discrete and continuous estimators
- Sparse, threaded, and streaming analyses
- Checkpointable information accumulators
- Optional plotting, separate from core calculations
Built in Julia
TensIT returns structured, reproducible results that can move from exploratory analysis into larger scientific workflows. Core calculations remain separate from optional plotting, and its accumulators support streaming or distributed work without retaining every original observation.
Discuss a TensIT workflowExplore EnviCloud
See the ideas before writing code.
Use approachable public demos to compare tensor resolutions or build intuition about information in one- and two-dimensional data.