Time Saving Stated Aim Classifier

This is a RoBERTa-large model that is trained to classify whether an explicit stated aim described in a British historical patent is designed to save time.

Hyperparameters:

lr = 3e-5
batch size = 50

Validation set results:

{'eval_loss': 0.6435797214508057,
 'eval_accuracy': 0.85,
 'eval_precision': 0.8466448445171849,
 'eval_recall': 0.85,
 'eval_f1': 0.8480286738351254,
 'eval_runtime': 1.075,
 'eval_samples_per_second': 55.816,
 'eval_steps_per_second': 1.861}

Test set results:

{'eval_loss': 0.9021337032318115,
 'eval_accuracy': 0.8333333333333334,
 'eval_precision': 0.8278777959629023,
 'eval_recall': 0.8333333333333334,
 'eval_f1': 0.825925925925926,
 'eval_runtime': 0.8154,
 'eval_samples_per_second': 73.583,
 'eval_steps_per_second': 2.453}