Skip to content

Changelog#

All project's changes will be documented here.

[3.2]#

Added#

  • Load a FittedParquetDataset, ExplanationArtifact, Filter or TrainedModelArtifact instances by id or name using get_by_id and get_by_name class methods.
  • Load all FittedParquetDataset, ExplanationArtifact, Filter or TrainedModelArtifact instances of a project using load_all class method.
  • Delete a FittedParquetDataset, ExplanationArtifact, Filter or TrainedModelArtifact instances using delete method.
  • Updating TrainedModelArtifact or ExplanationArtifact instances disabled, except for name and description attributes.
  • Updating Filter instances attributes dataset, min_index and max_index disabled.
  • Updating FittedParquetDataset instances attributes disabled.
  • Saving FittedParquetDataset and Filter using save method.
  • Documentation on Xpdeep Platform Accesss

Changed#

  • Additional feature types available.
  • ImageFeatureAugmentation is renamed FeatureAugmentation.
  • ExplainableFeature feature_augmentation attribute is renamed augmentation.
  • Project.list_fitted_datasets method replaced by Filter.load_all.
  • Project.list_trained_models method replaced by TrainedModelArtifact.load_all
  • Project.list_explanations method replaced by ExplanationArtifact.load_all`
  • Project.list_computed_statistics method replaced by FittedParquetDataset.load_computed_statistics
  • FittedParquetDataset.artifact_id property replaced by FittedParquetDataset.id
  • Building an XpdeepModel using from_torch class method requires an example dataset instead of a FittedSchema
  • The trainer loss must be an XpdeepLoss
  • Remove target_homogeneity_pruning_threshold of ModelDecisionGraphParameters parameters
  • Rework tutorials to use more default values

Removed#

  • split_name is deleted in ParquetDatasetArtifactInsert.
  • target_homogeneity_pruning_threshold and reset_optimizer_on_pruning are deprecated and will be removed in the next release.

[3.1]#

Added#

  • ImageFeature type.
  • Data augmentation is available for ImageFeature with ImageFeatureAugmentation, backed by torchvision.
  • Object Detection task, with a tutorial, and provided models.
  • Handle error on creating or getting a project with same name as another one you do not have access to.

Fixed#

  • Serialization of TorchPreprocessor transform and inverse_transform methods was incorrect.
  • ApiError when the AnalyzedSchema returned a BaseFeature if no match was found.
  • Bugs related to duplicated insertions of Filters and Datasets.
  • Bug related to updating project name and description.

Changed#

  • TorchModel is replaced by ApiModule for model serialization.
  • is_post_hoc is replaced by frozen_model in ModelDecisionGraphParameters.
  • PostHocTrainer is replaced by FrozenModelTrainer.
  • IndexMetadata should be added in your parquet file and specified in the Schema.