A dataset split partitions examples by workflow purpose: training fits a model, testing evaluates performance, and unsupervised data supports learning without labels. section 5.1.3
Load one split with Hugging Face Datasets
Use load_dataset("stanfordnlp/imdb", split="train") to download only the training split instead of the complete DatasetDict. section 5.1.3
Inspect split contents before using them:
- IMDb exposes train and test splits with 25,000 rows each, plus an unsupervised split with 50,000 rows. section 5.1.3
- Yelp Polarity provides 560,000 training rows and 38,000 test rows, with text and binary label features. section 6.1.1
- Fine-tuning assigns the train split to training and the test split to evaluation through Trainer. section 6.1.5
from datasets import load_dataset dataset_id = 'stanfordnlp/imdb' dataset = load_dataset(dataset_id) #1 print(dataset) #2
Run the minimal loading pattern, then inspect the resulting DatasetDict and its named splits. section 5.1.3
