class pymc.ImplicitGradient(approx, estimator=<class 'pymc.variational.operators.KSD'>, kernel=<pymc.variational.test_functions.RBF object>, **kwargs)[source]#

Implicit Gradient for Variational Inference

not suggested to use

An approach to fit arbitrary approximation by computing kernel based gradient By default RBF kernel is used for gradient estimation. Default estimator is Kernelized Stein Discrepancy with temperature equal to 1. This temperature works only for large number of samples. Larger temperature is needed for small number of samples but there is no theoretical approach to choose the best one in such case.


ImplicitGradient.__init__(approx[, ...])[n, score, callbacks, ...])

Perform Operator Variational Inference

ImplicitGradient.refine(n[, progressbar, ...])

Refine the solution using the last compiled step function

ImplicitGradient.run_profiling([n, score])