Skip to content

Releases

v0.3.2

Choose a tag to compare

@franckalbinet franckalbinet released this 03 Nov 23:03

New Features

  • Accept a list of batch transforms in train_or_load (#25)

Bugs Squashed

  • Make the datasets of get_spectral_dls picklable, so DataLoaders with spawned workers work (#26)

v0.3.1

Choose a tag to compare

@franckalbinet franckalbinet released this 27 Oct 21:37

New Features

  • Pass extra keyword arguments of train_or_load to spectral_learner (#24)
  • Add load_weights, to load a saved model without its training data (#23)

v0.3.0

Choose a tag to compare

@franckalbinet franckalbinet released this 27 Oct 19:46

New Features

  • Add the transfer module, to train, apply and adapt models across datasets and instruments (#22)
  • Add group_folds, for cross-validation that keeps each group in one fold (#21)
  • Add groups to get_ks_splitter, to keep each group on one side of the split (#20)
  • Add get_mask_splitter, for a fixed test set given as a boolean mask (#19)
  • Add CutCNN: the first blocks of a SpectralCNN with a new linear head (#18)

v0.2.0

Choose a tag to compare

@franckalbinet franckalbinet released this 22 Oct 21:09

Breaking Changes

  • Remove read_spa and load_spa, now in soilspecdata (#17)
  • Rename StandardizeSpectrum to BatchSNV, dividing by n (#16)

v0.1.4

Choose a tag to compare

@franckalbinet franckalbinet released this 21 Oct 13:40

New Features

  • Add load_spa to build a Spectra from .spa files (#15)
  • Add read_spa to read Thermo OMNIC .spa files (#14)

v0.1.3

Choose a tag to compare

@franckalbinet franckalbinet released this 18 Oct 10:59

New Features

  • Add plot_cv and a stage column to cross_validate (#13)

v0.1.2

Choose a tag to compare

@franckalbinet franckalbinet released this 18 Oct 09:55

New Features

  • Add cross_validate, returning per-fold, per-epoch losses and metrics (#12)
  • Add get_kfold_splitters for cross-validation (#10)

Bugs Squashed

  • get_splitter and get_pre_splitter reset the global NumPy random state (#11)

v0.1.1

Choose a tag to compare

@franckalbinet franckalbinet released this 17 Oct 19:27

New Features

  • Add spectral_learner, with backbone and head parameter groups (#9)

v0.1.0

Choose a tag to compare

@franckalbinet franckalbinet released this 02 Oct 19:45

Breaking Changes

  • Require soilspecdata 0.2.1, soilspectfm 0.1.1, fastai 2.8.12 and fastcore 2.2.22 (#4)
  • Remove unused and dataset-specific API (#3)
  • Store wavenumbers once in dls.wns instead of a TensorSpectrum class attribute (#2)
  • Generalise the CNN: add SpectralCNN, make MirzaiCNN a configuration of it (#1)

New Features

  • Rewrite the docs front page and align the docs site with soilspectfm (#7)
  • Add soilspecdl.all (#6)
  • Add a bundled toy dataset, load_toy_kssl (#5)

Bugs Squashed

  • get_splitter never rejects train_pct + valid_pct above 1 (#8)