v0.2.5

What's Changed

• Deprecated support for Python 3.8
• Bnb/exo agg methods by @bnb32 in #279
• Bnb/net cdf4 to xarray caching by @bnb32 in #280
• cacher time_index to h5s as encoded string (S25 dtype). by @bnb32 in #281
• Add pixi install instructions by @bnb32 in #282
• Bnb/mixed timestep fix by @bnb32 in #286
• Bnb/sliced wasserstein by @bnb32 in #283
• n_projections arg moved to init so it can be specified during mod… by @bnb32 in #287

Full Changelog: v0.2.4...v0.2.5

Miscellaneous QOL additions

What's Changed

• updated readme with new content and graphics by @grantbuster in #267
• Bnb/nc collection by @bnb32 in #268
• added dib article and updated release of sup3rcc v0.2.2 by @grantbuster in #269
• enable multiple exo source files by @bnb32 in #270
• Bnb/improved attrs by @bnb32 in #271
• Bnb/obs scaling by @bnb32 in #272
• feat: implement _match_model_input method in MultiStepGan to ensure f… by @bnb32 in #275
• Bnb/exo agg low mem by @bnb32 in #276
• Bnb/feature aliases by @bnb32 in #274
• Bnb/cache load edit by @bnb32 in #277
• Update pyproject.toml by @bnb32 in #278

Full Changelog: v0.2.3...v0.2.4

Lots of updates, new features for observations in Sup3rWind

What’s Changed

• Gb/bc optm (#265) @grantbuster
• Gb/sup3rcc example (#266) @grantbuster
• Gb/cpu kw (#264) @grantbuster
• Bnb/cond obs fwp (#262) @bnb32
• Added relativistic discriminator loss used in ESRGAN paper. (#261) @bnb32
• multi-term loss function capability: model loss arg can be dictionary… (#260) @bnb32
• Bnb/disc training fix (#259) @bnb32
• Bnb/min padding general (#258) @bnb32
• need to parse FeatureRegistry so no overlap in kwargs (#255) @grantbuster
• interpolation adjustment: option to include single levels in interpol… (#254) @bnb32
• bias refact: run_in_parallel function to remove duplicate calls t… (#253) @bnb32

Bug fixes and improvements for SolarCC

What’s Changed

• Gb/solar experiment (#250) @grantbuster
• preload of coords added to compute and h5 preload test added. (#251) @bnb32
• Bug fix on GCM clearsky ratio (#249) @grantbuster

Version 0.2.1

What's Changed

• Bnb/masked fwp by @bnb32 in #231
• Gb/optm state by @grantbuster in #242
• uv invert for netcdf writes by @bnb32 in #243
• sum multi gpu gradients and apply once to weight equally in adam mome… by @grantbuster in #244
• miscellaneous bias module clean up by @bnb32 in #245
• Gb/disc valid by @grantbuster in #246
• Bnb/1d nc loading by @bnb32 in #247
• Gb/qa fix by @grantbuster in #248

Full Changelog: v0.2.0...v0.2.1

Major refactor of forward pass modules and data / batch handler classes.

What's Changed

• Use rex Regridder port by @ppinchuk in #228
• Gb/presrat updates 2 by @grantbuster in #230
• Bnb/compute on none chunks by @bnb32 in #233
• Gb/bc debug by @grantbuster in #232
• fixed double bias correction of features in bc qdm utility by @grantbuster in #234
• Fix: Typos on optional dependencies related to doc by @castelao in #236
• Updated Sup3rX access patterns - keys can be either features or numpy style indexing but not both by @bnb32 in #235
• if n_writes request is too big, set to n_chunks instead of raising error by @grantbuster in #237
• added simple test on cc batching for daily boundaries by @grantbuster in #238
• Bnb/caching fixes by @bnb32 in #239
• Gb/bc kwargs by @grantbuster in #240
• Bnb/dh refactor by @bnb32 in #220

Full Changelog: v0.1.4...v0.2.0

Version 0.1.4

What’s Changed

• Exclude test data in release tar

Version 0.1.3

Prerelease for v0.2.0 refactor

What’s Changed

• PresRat correction (#215) @castelao
• Time dependent QDM (#226) @castelao
• Explicit keyword argument (#225) @castelao
• save default device to model parameters so model weights can be loade… (#221) @grantbuster
• fixed bias calc class import path (#222) @grantbuster
• Improving ruff setup (#224) @castelao
• Allow some tolerance in test comparison (#219) @castelao
• fix performance issue with slow listcomp on regrid indexing (#217) @grantbuster
• Reorganizing QDM related methods (#216) @castelao
• Pipeline requires explicit arguments for QuantileDeltaMappingCorrection (#211) @castelao
• Basic setup for using pixi (#212) @castelao
• Applying fill_and_smooth on QDM (#214) @castelao
• Bnb/nc4 dep removal (#210) @bnb32
• tensorboard logging (#208) @bnb32
• Fixing minor issues left from PR #200 (#209) @castelao
• Quantile Delta Mapping (#200) @castelao
• combined conditional moments and wind conditional moments. (#207) @bnb32
• Gb/no regrid (#206) @grantbuster
• allow surface spatial met model to interpolate non temp/pres/rh features (#204) @grantbuster
• Gb/precip skill (#195) @grantbuster
• prettifying how examples show up in docs page (#203) @bnb32
• Fix documentation, requires tags to figure out version (#202) @castelao
• Sup3rWind examples (#199) @bnb32
• Fix wrong path to version in documentation setup (#201) @castelao
• Migrating to pyproject.toml (#194) @castelao
• Instructions to use GPU on OSX (#197) @castelao
• Updating pylint setup for pre-commit (#198) @castelao
• Syntax update using numpy and scipy (#196) @castelao
• Bnb/dev (#193) @bnb32
• Defining some minimum versions (#190) @castelao
• Adding missing people (#192) @castelao
• Caching PIP (#191) @castelao
• Adding CITATION.cff (#188) @castelao
• Updating classifiers (metadata) (#189) @castelao
• cs ghi bug fix (#187) @grantbuster
• Bnb/dev (#186) @bnb32
• Gb/lr loss (#185) @grantbuster
• added kwarg to fwp to allow for constant tensor output e.g., in the c… (#184) @grantbuster
• Gb/bug (#182) @grantbuster

Updates

• Fix badges on main README (#213) @ppinchuk

Data Handler Refactor, bias and exo algorithm update, other misc changes

What’s Changed

• Gb/bias nc (#181) @grantbuster
• bug fix with negative test for dual DH operating on data handlers tha… (#179) @grantbuster
• Bnb/ddh norm (#180) @bnb32
• Significant refactor on handling of feature names in DataHandler and model (#178) @grantbuster
• stdev and mean hot fix (#177) @bnb32
• fix get_clearsky_ghi function to use correct time slice (#176) @spodgorny9
• Bnb/dev (#173) @bnb32
• Bnb/nc collector (#175) @bnb32
• Add ability to specify duplicate command names in Pipeline (#174) @ppinchuk
• Bnb/train only features hot fix (#171) @bnb32

Exogenous data refactor with breaking changes. Dual data handling. GAPS integration.

Refactored exogenous data handling in the forward pass and training. This enabled deletion of some specialized models, which creates breaking changes.

Dual data handler added for using different datasets for low res and high res data for training. For example, low_res = ERA5 data and high_res = WTK.

GAPS integration for command line interface and job handling. Allowed removal of reV dependency.

What's Changed

• Bnb/dev by @bnb32 in #148
• WindGanDC model by @bnb32 in #150
• Bnb/bc smoothing by @bnb32 in #151
• Bnb/dev by @bnb32 in #152
• Bug fix for conditional moment with topography by @malihass in #149
• Use multi gpu for conditional moment estimation by @malihass in #157
• Bnb/log interp by @bnb32 in #155
• Gb/skill by @grantbuster in #163
• Bnb/h5 with era training by @bnb32 in #162
• Bnb/dual dh patch by @bnb32 in #164
• Bnb/ws power law by @bnb32 in #165
• bug fix on bias calc object initialization named args and cli parsing by @grantbuster in #166
• Bnb/exo refactor by @bnb32 in #167
• GAPs integration by @ppinchuk in #168
• Bnb/exo refactor plus by @bnb32 in #169
• Update version.py by @bnb32 in #170

New Contributors

• @ppinchuk made their first contribution in #168

Full Changelog: v0.1.0...v0.1.1