domnasrabadi/juree_v1

Using combined dataset from multiple sources for safe vs unsafe text classification.

This is an initial experiment with results below:

{'test_loss': 0.0818108394742012,
 'test_accuracy': 0.9797578509269769,
 'test_f1': 0.9796642969399756,
 'test_recall': 0.9901676829268292,
 'test_precision': 0.9854357885155124,
 'test_runtime': 75.8752,
 'test_samples_per_second': 209.001,
 'test_steps_per_second': 13.074}

142k samples for train, 15k for test.

---

metrics: - accuracy - recall - precision - f1 pipeline_tag: text-classification