Examples

Standalone scripts covering every major DRIADA capability. Each example generates synthetic data internally — no external files needed. Run any script directly: python examples/<folder>/<script>.py

Source code: examples/ on GitHub

INTENSE — Selectivity Detection 

intense_basic_usage	Minimal INTENSE workflow: generate a synthetic population, run two-stage significance testing, extract per-neuron results, and visualize tuning.
full_intense_pipeline	Complete pipeline across all feature types — circular (head direction), spatial (position), linear (speed), and discrete (events) — with ground-truth validation and detection metrics.
mixed_selectivity	Disentangle mixed selectivity patterns using multivariate features, synergy/redundancy decomposition, and interaction information.

Dimensionality Reduction 

compare_dr_methods	Systematic comparison of dimensionality reduction methods (PCA, Isomap, UMAP, t-SNE, etc.) on synthetic datasets with quality metrics (k-NN preservation, trustworthiness, continuity, stress) and timing benchmarks.
dr_simplified_api	Quick-start guide for DRIADA’s dimensionality reduction API: `MVData`, automatic parameter handling, custom metrics, and method-specific configurations.
dr_sequence	Compare direct UMAP vs PCA-then-UMAP sequential dimensionality reduction on synthetic neural manifolds with preservation metrics.
autoencoder_dr	Neural-network-based dimensionality reduction on circular manifold data: standard autoencoder with `continue_learning`, Beta-VAE, and PCA comparison.

Extract 1D circular structure from head direction cells using multiple dimensionality reduction methods and intrinsic dimensionality estimators (correlation dimension, geodesic dimension, participation ratio).

Integration — INTENSE + Dimensionality Reduction 

intense_dr_pipeline	Use INTENSE to identify spatially selective neurons, then compare dimensionality reduction quality (decoding R², distance correlation) using all neurons vs the INTENSE-selected subset.
loo_dr_analysis	Leave-one-out neuron importance analysis: remove each neuron, measure manifold degradation, and compare structural importance with INTENSE selectivity scores.
functional_organization	Reverse the usual direction: treat embedding components as features, run INTENSE on them, and discover which neurons drive which manifold dimensions.

Network Analysis 

network_analysis	Compute cell-cell functional connectivity via pairwise MI significance, build a `Network` object, and analyze graph structure (degree distribution, communities, spectral properties).
network_spectrum	Spectral analysis toolkit: eigendecomposition of adjacency and Laplacian matrices, inverse participation ratio, spectral entropy, communicability, and Gromov hyperbolicity.

Recurrence Analysis 

recurrence_basic	Recurrence fundamentals on classic signals (sine, Lorenz, noise): embedding parameter selection (TDMI, FNN), recurrence plots, RQA measures, three graph representations (RG, HVG, OPN), and windowed regime detection.
recurrence_population	Recover functional modules from population dynamics alone: per-neuron recurrence graphs, pairwise Jaccard similarity, permutation testing, community detection (Louvain), and ground-truth validation (ARI).

RSA — Representational Similarity 

rsa	Compute RDMs from stimulus-selective populations, compare representations across regions and sessions with multiple distance metrics, and run bootstrap significance testing.

Synthetic Data 

rnn_activations

Full DRIADA pipeline on simulated RNN activations: generate behavioral inputs, simulate a driven RNN, load into an Experiment, and run INTENSE + dimensionality reduction + network analysis. Demonstrates that DRIADA works with any (n_units, n_frames) data, not just calcium imaging.

Neuron — Spike Reconstruction & Quality 

neuron_basic_usage	Core `Neuron` class: generate synthetic calcium signals, reconstruct spikes (wavelet method), optimize rise/decay kinetics, and compute signal quality metrics.
spike_reconstruction	Compare wavelet vs threshold spike reconstruction on calcium traces with overlapping events; analyze detection accuracy differences.
threshold_vs_wavelet_optimization	Benchmark reconstruction modes: default kinetics, optimized kinetics, and iterative detection (n_iter=2, 3) with performance tradeoffs.

Utilities & Data Loading 

data_loading	Load real recording data into DRIADA: numpy arrays and feature annotations → `Experiment` object with all downstream analysis enabled.
signal_association	Information-theory primitives: pairwise MI (GCMI vs KSG estimators), time-delayed MI, conditional MI, and interaction information (synergy/redundancy).
behavior_relations	Feature-feature significance testing with FFT-based circular shuffling for behavioral variable correlations, independent of neural data.
spatial_analysis	Spatial data visualization: trajectory maps, occupancy maps, rate maps, and calcium traces for place cell analysis.
visual_utils	Publication-ready figure generation using DRIADA’s visual utilities: embedding comparison plots, selectivity summaries, and consistent styling.