Position Decoding

Bayesian decoding of spatial position from population neural activity.

place_bayes

Decode animal position using Bayesian inference on population firing rates.

This function implements a naive Bayesian decoder that estimates spatial position from the firing rates of a population of neurons, given their spatial tuning curves (rate maps).

Algorithm:

Given firing rates fr and rate maps rate_map, computes the posterior probability of the animal being at each spatial bin using Bayes’ theorem with a uniform prior.

Dimension Parameters:

N: number of neurons
T: number of temporal bins
X: number of spatial bins

Usage:

import numpy as np
from neuralib.decoding.position import place_bayes

# Example: 50 neurons, 100 time bins, 40 spatial bins
n_neurons = 50
n_time_bins = 100
n_spatial_bins = 40
spatial_bin_size = 2.5  # cm

# Firing rates: (T, N) array
fr = np.random.rand(n_time_bins, n_neurons)

# Rate maps (spatial tuning curves): (X, N) array
rate_map = np.random.rand(n_spatial_bins, n_neurons)

# Decode position
posterior = place_bayes(fr, rate_map, spatial_bin_size)
# posterior shape: (T, X) - probability of each position at each time

# Get decoded position (maximum a posteriori)
decoded_position = np.argmax(posterior, axis=1)  # (T,)

Parameters:

fr: Firing rate array of shape (T, N)
rate_map: Spatial tuning curves of shape (X, N)
spatial_bin_size: Spatial bin size in cm

Returns:

Posterior probability matrix of shape (T, X)