Skip to content

Instantly share code, notes, and snippets.

Embed
What would you like to do?
DHCD (Devanagari) dataset ResNet32 model fitting
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@sauravrt

This comment has been minimized.

Copy link

@sauravrt sauravrt commented Jun 3, 2020

Hey Suvash
I'm running your notebook hoping to reproduce your results. My runs resulted in an error rate of ~2.45%.
In the notebook, you mention using DHCD dataset available from UCI. I obtained the same dataset, but it does not have an explicit validation set. However, the output to path.ls() shows that you have a 'valid' folder. Did you manually create the validation set?

@suvash

This comment has been minimized.

Copy link
Owner Author

@suvash suvash commented Jun 3, 2020

Hi there. From what I remember, I did create a separate validation set. I'll have to hunt down the script I used to do that. Unfortunately, that's not included in this notebook.
I could actually update this notebook to use the newly released v2, and post the results. It shouldn't deviate a lot. will update here once I get around doing that. Give or take, in some weeks I'd assume.

@sauravrt

This comment has been minimized.

Copy link

@sauravrt sauravrt commented Jun 3, 2020

Thanks for confirming.
I'm using fastai v2 and with that there is a new function cnn_learner() to init a pre-trained model based learner. This function has a parameter to specify the fraction of dataset to use as the validation set.
Anyways, will look forward to your updates to this notebook.

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