Chapter 15 from the book, Siamese training issue

[Hawxyde (Bori)]

Hi everyone. I’m a bit stuck on chapter 15 from the book. When running the book jupyter notebook without any modifications, the Siamese model has terrible performance (of around 50% in accuracy, for a boolean classification task, is basically random). I can’t find the part that is problematic. Did something change in the way fastai works since it has been written down?

Thanks

[Ethan96 (Ethanmartinez)]

It sounds like you’re experiencing issues with the Siamese model in Chapter 15. Since the book was written in 2019, changes in the Fastai library may affect performance. Try these steps:

1. Review the data setup: Ensure that the data loading and preprocessing steps match the current API.
2. Preformatted textRevisit the loss function and model architecture: Verify that they’re set up correctly for your task.

[mary598chess (Mary Chess)]

@Hawxyde

Hi everyone. I’m a bit stuck on chapter 15 from the book. When running the book jupyter notebook without any modifications, the Siamese model has terrible performance (of around 50% in accuracy, for a boolean classification task, is basically random). I can’t find the part that is problematic. Did something change in the way fastai works since it has been written down?

Thanks

Hello,
It sounds like you’re McDonald’s Customer Survey experiencing a common issue with Siamese models in fastai. Since the book was written, there have been updates to fastai that might affect model performance.

Could you share the specific part of the code where you define and train the Siamese model? This will help identify any potential issues. Additionally, checking for updates in the fastai documentation or forums might provide some insights.

Best Regards,
Mary Chess