Storchastic logo.

Storchastic is a PyTorch library for stochastic gradient estimation in Deep Learning. Stochastic deep learning models are becoming increasingly relevant. For example, they are commonly used in the fields of Variational Inference and Reinforcement Learning. We can formalize stochastic models using so-called stochastic computation graphs. While PyTorch computes gradients of deterministic computation graphs automatically, PyTorch will not automatically estimate gradients on such stochastic graphs. This is because they require marginalization over the stochastic nodes in the graph, which is usually intractable and needs to be estimated.

With Storchastic, you can easily define any stochastic deep learning model and let it estimate the gradients for you. Storchastic provides a large range of gradient estimation methods that you can plug and play, to figure out which one works best for your problem. Storchastic provides automatic broadcasting of sampled batch dimensions, which increases code readability and allows implementing complex models with ease.

When dealing with continuous random variables and differentiable functions, the popular reparameterization method is usually very effective. However, this method is not applicable when dealing with discrete random variables or non-differentiable functions. This is why Storchastic has a focus on gradient estimators for discrete random variables, non-differentiable functions and sequence models.

Mail e.van.krieken@vu.nl for any help or questions.

API documentation:

Indices and tables