Skip to content

Latest commit

 

History

History
5 lines (5 loc) · 292 Bytes

File metadata and controls

5 lines (5 loc) · 292 Bytes

Key takeaways

  • I have experimented with different model architectures and found this network best
  • Don't blindly put dropouts in every layers
  • Learning rate scheduler is very good if your training loss is not converging
  • Don't train for too many epochs in small dataset like this