lhallee's picture
Update README.md
4745a7e verified
metadata
dataset_info:
  features:
    - name: IDs
      dtype: string
    - name: score
      dtype: float64
    - name: OrgA
      dtype: string
    - name: OrgB
      dtype: string
    - name: SeqA
      dtype: string
    - name: SeqB
      dtype: string
  splits:
    - name: train
      num_bytes: 681153007.5956708
      num_examples: 645692
    - name: valid
      num_bytes: 7149732
      num_examples: 5854
    - name: test
      num_bytes: 1401546.5899105743
      num_examples: 1322
  download_size: 632661073
  dataset_size: 689704286.1855814
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: valid
        path: data/valid-*
      - split: test
        path: data/test-*

lhallee/Stringv12ModelOrgPairs90 but negatome sequences are removed, and splits are formed. The test set has no sequence overlap with the training set. The valid set is essentially a random split (leftover from test creation).