Releases: kazewong/flowMC
flowMC-0.3.4
What's Changed
- update example and add print summary to Sampler by @ThibeauWouters in #169
- The previous load model function does not correctly load the model by @kazewong in #172
Full Changelog: flowMC-0.3.3...flowMC-0.3.4
flowMC-0.3.3
What's Changed
- 165 add probability floor to normalizing flow model -> updating Adam optimizer to optimize with constraint by @kazewong in #168
Full Changelog: flowMC-0.3.2...flowMC-0.3.3
flowMC-0.3.2
Optimization strategy is included in this version.
Also fixed a bug in the flow model that prevent training
What's Changed
Full Changelog: flowMC-0.3.1...flowMC-0.3.2
flowMC-0.3.1
Fixing training bugs and sampling bugs. Previous version incorrectly also start at initial position instead of actually stepping during sampling. This is fixed and tested with community example now.
Full Changelog: flowMC-0.3.0...flowMC-0.3.1
flowMC-0.3.0
The version release focuses on cleaning up the internal API, which should makes the study of auto-tuning way easier now. The top level API remains relatively unchanged. On the performance side, most noticeably the compilation time of the normalizing flow has been reduced significantly.
For the next version of flowMC (0.4.0), the main focus will be some sort of auto-tune and adding more potential strategies in the mix, including 1. using importance sampling with normalizing flow to further improve the sampling results 2. Start the normalizing flow with some variational inference scheme instead of from scratch.
What's Changed
- Bayeux example by @kazewong in #146
- Added plotting utilities and refactored hyperparams by @ThibeauWouters in #143
- small bug fix in keys for stepping optimizer by @tedwards2412 in #148
- Sync branch 134 by @kazewong in #150
- 134 add precommit script and fix formatting by @kazewong in #151
- 144 get rid of random key set by @kazewong in #152
- 153 use scan to reduce nf compilation time by @kazewong in #154
- Sync branch 142 by @kazewong in #155
- 142 sampling from arrays by @kazewong in #156
- Update README.md by @kazewong in #159
- 158 put training loop into nf class by @kazewong in #160
- 157 making sampler composable by @kazewong in #161
New Contributors
- @ThibeauWouters made their first contribution in #143
- @tedwards2412 made their first contribution in #148
Full Changelog: flowMC-0.2.4...flowMC-0.3.0
flowMC-0.2.4
What's Changed
Full Changelog: flowMC-0.2.3...flowMC-0.2.4
flowMC-0.2.3
flowMC-0.2.2
Minor bug fixes and consistency updates in #125
What's Changed
Full Changelog: flowMC-0.2.1...flowMC-0.2.2
flowMC-0.2.1
Two significant changes in this release, following PR #118 and PR #123.
Now there is a black box optimization interface using evosax
, an evolutionary algorithm to find the optimum of the likelihood function. In our experience, this works better and is more paralleled compared to packages such as jaxopt
.
In PR #123, we re-haul a large chunk of the normalizing flow interface. We now have deprecated flax
and distrax
in favor of equinox
. More information can be find in the specific PR note
What's Changed
- small update to dual moon tutorial by @marylou-gabrie in #120
- 118 black box optimization interface for finding maximum of the likelihood by @kazewong in #122
- 123 update dependencies version by @kazewong in #124
Full Changelog: flowMC-0.2.0...flowMC-0.2.1