API¶
Convenience Methods¶
|
Fit a model using black-box variational inference. |
|
Check variational inference diagnostics. |
Approximation Families¶
|
An abstract class for an variational approximation family. |
|
A mean-field Gaussian approximation family. |
|
A mean-field Student's t approximation family. |
|
A full-rank multivariate t approximation family. |
|
|
|
|
Models¶
|
Base class for representing a model. |
|
Class that encapsulates a PyStan model. |
Variational Objectives¶
|
A class representing a variational objective to minimize |
|
Exclusive Kullback-Leibler divergence. |
|
Inclusive Kullback-Leibler divergence using Distilled Importance Sampling. |
|
Log of the alpha-divergence. |
Diagnostics¶
|
Compute all VI diagnostics. |
|
Compute a bound on the alpha-divergence. |
|
Compute error bounds. |
|
Compute all bounds. |
Optimization¶
An abstract class for optimization |
|
|
Stochastic gradient descent. |
|
Adam optimization method (Kingma and Ba, 2015) |
|
Averaged Adam optimization method (Mukkamala and Hein, 2017, §4) |
|
RMSProp optimization method (Hinton and Tieleman, 2012) |
|
Averaged RMSProp optimization method (Mukkamala and Hein, 2017, §4) |
|
Adagrad optimization method (Duchi et al., 2011) |
|
Windowed Adagrad optimization method (Default optimizer in Pymc3) |
|
Fixed-learning rate stochastic optimization meta-algorithm (FASO) |
|
A robust, automated, and accurate BBVI optimizer (RAABBVI) |