prefopt.model package¶
Submodules¶
prefopt.model.binary module¶
Gaussian-process model for binary preferences.
-
class
prefopt.model.binary.BinaryPreferenceModel(n_iter=500, n_samples=1, sigma_signal=1.0, sigma_noise=1.5, link='logit', lengthscale=1.0, ard=False)¶ Bases:
prefopt.acquisition.base.PreferenceModelGaussian-process model for binary preferences.
Depending on the choice of the link function, the model is either a Thurstone-Mosteller (probit) or a Bradley-Terry (logit) model.
Attributes
n_iter (int, optional (default: 500)) The number of optimization iterations. n_samples (int, optional (default: 1)) The number of samples for calculating stochastic gradient. sigma_signal (float, optional (default: 1.0)) The variance of the signal. sigma_noise (float, optional (default: 1.5)) The variance of the observation noise. link (str, optional (default: probit)) The link function. Can either be probit in which case the model is a Thurstone-Mosteller model or logit in which case the model is a Bradley-Terry model. lengthscale (float or np.array, optional (default: 1.0)) The lengthscale parameter. Can either be a scalar or vector of size D where D is the number of dimensions of the input space. ard (bool, optional (default: False)) Use automatic relevance determination (ARD) instead of fixed lengthscales. If True, the value of lengthscale will be ignored. -
fit(X, y)¶ Perfom inference in the preference model.
This will update the mean and variance function of the model.
Parameters: X : np.array, shape (N, D)
The preference items.
y : dict
Mapping from indices corresponding to items (rows) in X to preferences. For instance, {(0, 1): 1} means that X[0] is preferred over X[1].
-
mean(X)¶ The posterior mean function evaluated at points X.
-
variance(X)¶ The posterior variance function evaluated at points X.
-