Skip to content

Changelog#

All project's changes will be documented here.

[2.3.0]#

Changed#

  • No more need to create ìndex_xp_deep Metadata Feature for Schemas.
  • categories attribute of CategoricalFeature class becomes optional and should not be set in client side, so categories will be automatically inferred in server side.
  • torch '2.4' -> '2.5'
  • Instantiate projects using Project.create class method instead of Project(). IDs of projects are no more set by the client, but generated in server side.
  • Update projects using project.update instance method instead of Project()
  • Load projects using Project.load instance method instead of Project()
  • Project names are unique.
  • DictMetrics constructor only takes as arguments TorchGlobalMetric and TorchLeafMetric. Metric objects of torchmetrics library are only indicated through TorchGlobalMetric and TorchLeafMetric.
  • UnivariateTimeSeriesand MultivariateTimeSeries now have a asynchronous boolean attribute to indicate they are synchronous or not (dynamic time warping required).

Added#

  • additional_attributes keyword argument added to TorchPreprocessor __init__ method.
  • Project.load_all class method replaces get_all_projects function. Returns a Projects list.
  • Parameter target_as_indexes on Metric. It computes target indexes from onehot vectors in classification. Especially useful for torchmetrics multiclass classification metrics which expects targets to be indexes.
  • Metadata is replaced by IndexMetadata. It is automatically added to the Schema and represents the "index_xp_deep" column.
  • Explainable Features have an optional feature_augmentation attribute.
  • ParquetDataset, AnalyzedParquetDataset and FittedParquetDatasetoptional global_augmentation attribute.
  • ImageFeature is now available to work on images, as an example the MNIST dataset is available within a tutorial.

Removed#

  • get_all_projects function has been deleted and replaced by class method Project.load_all.