oliveiraph17/holoclean-incremental

holoclean-incremental's Issues

Identify the difference in the precision metric

The precision metric in the report does not correspond to the number of correct repairs over the total number of repairs.

Adjust other featurizers to be attribute-based

Only the OccurAttrFeaturizer was done attribute-based in the attribute-based_incremental branch. The other featurizers should be adjusted accordingly.

Routine to export datafile and ground truth to CSV

The idea is to be able to export to CSV a Data Frame both to a data file and to a ground truth file following the format accepted by HoloClean. This will be useful to create new datasets and/or change the record order of existing ones.
It should also contain example code to shuffle the Data Frame or sort it according to a criterion before exporting.

Merge batch featurizing from dev-scale into dev

The dev-scale branch has several contributions on featurizers to featurize a dataset using batches. This limits the amount of memory needed for featurizing to the provided parameter.
Essentially, it makes the FeaturizedDataset class to use the torch.utils.data.Dataset interface for interactive handling the training and inferring datasets.

Merge incremental functionalities into dev

Clean up branches

• Rename (?) merge_devmixed_into_dev to dev_inc
  ==> Remember that attribute-based_incremental is a branch from it!

• Delete branch dev_mixed (it is still in the fork branch)

• Delete branch merge_hcqemb3_into_dev (it was already merged into dev)

Prepare a script to run tests

It should accept several datasets as well as parameter variations.