- Renamed smc to
mc
and mcmc-gan tomcmc
, in both the API and CLI. We're not really doing SMC, and the MCMC isn't really a GAN. - Used KDE reflection in the violin plots, to ameliorate the reduction
in density near the edges of the support. This was already implemented
for the
hist
plots. - Fixed iteration numbering in plots to start at zero.
- Fixed broken pipe when piping
dinf-tabulate
output to, e.g., head. - Updated the tutorial in the docs.
- Removed abc-gan CLI subcommand, and added
smc
subcommand. - Removed the
--resample
option fordinf-plot hist
. - Implement KDE variance shrinkage.
- Added entropy line plot.
- Bring dependencies up to date, e.g. fix deprecations.
- Lazily import scipy, to improve the CLI lag.
- Added
dinf tabulate
subcommand.
- Changed how the CLI is used. Now the model and discriminator are specified with the -m/--model and -d/--discriminator options, even when they are required.
- Removed DinfModel.feature_shape.
- Simplified Discriminator API.
- Add progress bars and do logging with
rich
. - Fix KDE bandwidth for abc-gan.
- Separate the train/test datasets from abc-gan proposals.
- Reduce memory use for jobs with millions of simulations.
- Remove alfi-mcmc-gan.
Initial release, featuring:
dinf
anddinf-plot
CLIs,- train, predict, abc-gan, mcmc-gan, alfi-mcmc-gan, pg-gan,
- two flavours of feature matrices, dinf-style
BinnedHaplotypeMatrix
and pg-gan-styleHaplotypeMatrix
, - two flavours of discriminator networks, dinf-style
ExchangeableCNN
and pg-gan-styleExchangeablePGGAN
, plus multi-population versions of each.
Dummy release to reserve name on Pypi.