metricspace 0.4.0

A python translation of code originally theorized in: Metric-space analysis of spike trains: theory, algorithms, and application Jonathan D. Victor and Keith Purpura Network 8, 127-164 (1997)

Metric Space Analysis

This repository hosts algorithms, functions and estimators for clustering adapted to the unique characteristics of neuronal entropy and spike timing.

Estimators are wrapped in sklearn.BaseEstimator and sklearn.ClusterMixin classes for integration into sklearn pipelines.

Classification algorith based on:

Nature and precision of temporal coding in visual cortex: a metric-space analysis

Most neural activity is represented relative to the spike rate of the individual neurons.

Neurons that fire significantly faster around the onset of a stimulus is considered a stimulus-active neuron. The activity of these neurons is then used to classify the stimulus. This is a very common method of classification in neuroscience.

However, the brain is complicated, and this may not tell the entire story.

The metric space analysis is a quantification using spike timings rather than spike rates. Using this method, we can quantify the similarity between two spike trains.

This allows us to classify stimuli based on the similarity of the spike trains and how much information is conveyed.

• q-value : precision of temporal coding. q=0 resolves no information in temporal coding, where all information lies in the spike count.
• h-value : amount of information conveyed by the spike train. h=0 conveys no information, where the spike train is completely random.

The crux of this model is attempting to use pairwise distances between spike-trains from different neurons, different animals, or different sessions to classify stimuli.