Skip to content

Hyperparameter optimization with early stopping and gradient clipping #2 #11

Description

@hbandukw

DATA: NON AUGMENTED / IMBALANCED
METRIC: weighted F1

Best Hyperparameters: {'learning_rate': 0.00011934626028114851, 'dropout_rate': 0.14349283542403013, 'num_trainable_layers': 2, 'train_batch_size': 32, 'grad_clip': 0.5835870650159247, 'step_size': 6, 'gamma': 0.8223727743347635, 'max_length': 319}

I ran it again in my code:
Epoch 1 - Train Loss: 1.2385, Val Loss: 1.0269, Val Accuracy: 0.6440
Epoch 2 - Train Loss: 0.8503, Val Loss: 0.9978, Val Accuracy: 0.6538
Epoch 3 - Train Loss: 0.6682, Val Loss: 1.0414, Val Accuracy: 0.6694
Epoch 4 - Train Loss: 0.4806, Val Loss: 1.1070, Val Accuracy: 0.6645
Epoch 5 - Train Loss: 0.3256, Val Loss: 1.2361, Val Accuracy: 0.6743
Early stopping at epoch 5
Final Model Accuracy: 0.6743

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions