graphdot.model.gaussian_field.weight module¶

class graphdot.model.gaussian_field.weight.RBFOverDistance(metric, sigma, sigma_bounds=(0.001, 1000.0), mopts={})[source]¶

Set weights by applying an RBF onto a distance matrix.

Parameters:

metric (callable) – An object that implements a distance metric.
sigma (float) – The log scale hyperparameter for the RBF Kernel.
sigma_bounds (float) – The bounds for sigma.
sticky_cache (bool) – Whether or not to save the distance matrix upon first evaluation of the weights. This could speedup hyperparameter optimization if the underlying distance matrix remains unchanged during the process.

__call__(X, Y=None, eval_gradient=False)[source]¶

Parameters:	eval_gradient (bool) – If true, also return the gradient of the weights with respect to the log-scale hyperparameters.

class graphdot.model.gaussian_field.weight.RBFOverFixedDistance(D, sigma, sigma_bounds=(0.001, 1000.0), sticky_cache=False)[source]¶

Set weights by applying an (optimizable) RBF onto a fixed distance matrix.

Parameters:	metric (callable) – An object that implements a distance metric. sigma (float) – The log scale hyperparameter for the RBF Kernel. sigma_bounds (float) – The bounds for sigma.

__call__(X, Y=None, eval_gradient=False)[source]¶

Parameters:	eval_gradient (bool) – If true, also return the gradient of the weights with respect to the log-scale hyperparameters.

class graphdot.model.gaussian_field.weight.Weight[source]¶

Bases: abc.ABC

__call__(X, Y=None, eval_gradient=False)[source]¶

Computes the weight matrix and optionally its gradient with respect to hyperparameters.

Parameters:

X (list) – The first dataset to be compared.
Y (list or None) – The second dataset to be compared. If None, X will be compared with itself.
eval_gradient (bool) – If True, returns the gradient of the weight matrix alongside the matrix itself.

Returns:

weight_matrix (2D ndarray) – A weight matrix between the datasets.
weight_matrix_gradients (3D ndarray) – A tensor where the i-th frontal slide [:, :, i] contain the partial derivative of the weight matrix with respect to the i-th hyperparameter.