SweetSpot 1.4.0

Efficient psychophysical adaptive procedures

SweetSpot is a small Python package for adaptive psychophysical procedures and psychometric-function estimation. It is intended to keep the core experimental logic lightweight and explicit, with numpy as the only required numerical dependency and matplotlib optionally used for visualization and diagnostics.

The package provides:

• PsychophysicalData, for storing and collating stimulus/response data.

• PsychometricFunction, a four-parameter psychometric-function model with configurable sigmoid shapes, priors, parameter grids, likelihoods, posterior summaries, and simulated responses.

• PsychometricLinkFunction subclasses such as Logistic and Gumbel. Weibull behavior is represented as a Gumbel function in log-stimulus space, via LogTransform, which avoids the numerical awkwardness of native Weibull parameterizations near zero.

• Parameterizable prior distributions such as Gaussian, with mutable named properties (mu, sigma) and change notifications that support live visualization of the effect of the prior.

• WUD, an implementation of Kaernbach's weighted up-down staircase: a simple, useful, old-fashioned adaptive procedure.

• QUEST, a classic Watson-and-Pelli-style Bayesian threshold tracker, built on the same posterior-update machinery as the more general Bayesian procedures.

• PsiMarginal, a discrete-grid implementation of Prins's psi-marginal method. It maintains a posterior over psychometric-function parameters, marginalizes over nuisance parameters, and chooses stimuli to reduce uncertainty in the parameter or parameters of interest. With parameterOfInterest='alpha,beta' and fixed asymptote parameters, it can also be configured to behave like the classic Psi method of Kontsevich and Tyler.

• Simulate(), a convenience function for testing adaptive procedures against simulated observers and visualizing their behavior.

SweetSpot is public-domain software released under CC0.