Fix in benchmark and update in torchdiffeq setup

Bug fix in printed message in benchmark method and usage of torchdiffeq version on PyPI.

Native sampling from truncated normal, HMC with dual averaging, and seaborn-related updates in examples

v0.0.9

Updated requirements and version

Updated examples to use newly added MCMC diagnostics

Merge pull request #26 from papamarkou/example_update

Updated examples to use recent MCMC diagnostics

Improved MCMC diagnostics

Added:

• ChainLists class, making it easier to call multi_rhat and other methods across multiple chains,
• method for approximating a non-positive definite matrix by a (semi-)positive definite matrix,
• finished summary method for a set of chains

Added multivariate ess and psrf (rhat) in torch

v0.0.6

Updated setup script for new version

Fixed bug in setup.py

A bug has been fixed in setup.py, which previously prevented installation via PyPI.

Multiple upgrades

• Added new samplers
• Made small edits in API for models and chains
• Added MMD
• Added benchmark method for samplers

Added robust adaptive Metropolis and fixed minor bugs

v0.0.3

Fix in multivariate normal kernel

Bug fixes and new functionality

Various bugs have been fixed, new samplers, predictive posterior approximation and functionality related to chain management have been added.

Starting codebase

This is the first release, in pre-alpha stage. The main functionality of the code has been thoroughly checked to ensure numerical correctness of the current MCMC samplers. There is more work to be done in terms of adding more samplers and conducting research, while the current framework can be trusted in terms of correct numerical output.