beanmachine
beanmachine copied to clipboard
facebookresearch
A library that allows for inference on probabilistic models
Summary: In order to not blow up the size of the compiled state of some important models, it would be nice to have matrix versions of some common operators; here's...
This commit adds the feature to run diagnostic tools using ArviZ and Bokeh in a Jupyter environment. - Added a `tools` sub-package in the `diagnostics` package. This new sub-package adds...
Summary: To support the use of `bm.param`s in the generative model, beanmachine's VI now: * Traces through `World` in addition to `VariationalWorld` so that `param`s used in `World` are added...
Summary: Due to failure of nightly conjugate test of Metropolis Adapted Langevin Algorithm (MALA), we have increased the number of stepd and decreased step size. Differential Revision: D38319379
Summary: This diffs includes the codes for the statistics gathering and generation module for BMG. It contains: Extensions to graph.h to include an object called Statistics as part of the...
### Motivation Eliminate the need to write inference in two places. Python is the single frontend and native is the backend. ### Changes proposed The change proposed is the addition...
Summary: All functions -- including type constructors -- in the torch modules are treated as "special" by the compiler. For special functions we have a "can have stochastic arguments" allow...
Summary: This file includes the codes for the statistics gathering and generation module for BMG. It contains: 1. Extensions to graph.h to include an object call Statistics as part of...
Summary: The compiler crashed when given a numpy array as a "constant" operand to a stochastic operator; we now detect this situation and automatically convert the value to a tensor...
Summary: Support's torch.optim.LBFGS by passing a closure Differential Revision: D38171088
Metadata
Owner
Metadata
A library that allows for inference on probabilistic models