Datasets are not splited
#2
by
weinwen
- opened
Thanks for your great work!
When I loaded the dataset accoring to Readme, with:
from datasets import load_dataset
ds = load_dataset("diarizers-community/callhome", "jpn")
print(ds)
I got a DatasetDict without spliting:
DatasetDict({
data: Dataset({
features: ['audio', 'timestamps_start', 'timestamps_end', 'speakers'],
num_rows: 120
})
})
Did I miss something?
Hi @weinwen ,
The dataset doesn't come with offcial splits in the TalkBank version: https://ca.talkbank.org/access/CallHome/jpn.html.
In our training scripts, we use the train_test_split method from the Datasets library to automatically split the dataset. You could do it this way:
from datasets import load_dataset, DatasetDict
ds = load_dataset("diarizers-community/callhome", "jpn")
train_testvalid = ds['data'].train_test_split(test_size=0.2, seed=0)
test_valid = train_testvalid['test'].train_test_split(test_size=0.5, seed=0)
ds = DatasetDict({
'train': train_testvalid['train'],
'validation': test_valid['test'],
'test': test_valid['train']
})
Hope it will help you!