how were the splits obtained/mapped?
#1
by
VictorSanh
HF staff
- opened
In the original repo, i see the files metadata.csv, and dataset_split/train_index_mapping.csv, dataset_split/val_index_mapping.csv, dataset_split/test_index_mapping.csv but the sizes don't seem to match.
metadata contains (for pew) 2096 lines (including 1 for header), but then are 9288 total lines in the split csv files. how do I link the old_id that seem to be in the split files to their new id?
Sorry for late reply, You can check dataset_split directory. (Multi-column dataset)
It has 7799 items totally
@VictorSanh
https://github.com/vis-nlp/Chart-to-text/tree/main/pew_dataset/dataset/dataset_split