Fine-tune a Vision Transformer Model with a custom biomedical dataset #85
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
|
Yes, looking forward to reviewing your notebook! 🤗 |
Added and edited the notebook
|
Hello @stevhliu, made the first commit I edited and organized all sections, waiting for your opinions and reviews |
| @@ -0,0 +1,767 @@ | |||
| { | |||
There was a problem hiding this comment.
In "Dataset Info", it'd be nice to briefly explain what the images are of so users have more context about what they're training the model to do.
Reply via ReviewNB
| @@ -0,0 +1,767 @@ | |||
| { | |||
There was a problem hiding this comment.
Is this dataset available on the Hub? I think it'd be easier for users to follow along if they could also download the dataset or if you provided some more information/details about how a user can create their own dataset with their images.
Reply via ReviewNB
| @@ -0,0 +1,767 @@ | |||
| { | |||
There was a problem hiding this comment.
Maybe say something like the following to avoid confusion with the next sentence where you say we can see the features again.
"We can the image is a PIL.Image with a label associated with it."
Reply via ReviewNB
|
Make sure to add your notebook to the |
|
Hello @stevhliu, I made some changes I put some images from dataset to "Dataset Info" section, but I have some questions. The user can find similar datasets from Kaggle, and I can also upload the dataset to Hub. What should I do? |
I think it'd be easiest to upload the dataset to the Hub so users can follow along with your notebook without putting in the extra work of finding a similar dataset from Kaggle if they don't want to.
I think you can create a new "Computer vision" section.
Sorry for the typo. You can put that text before you call We can see the image is a PIL.Image with a label associated with it.
train_ds[0] |
- Added notebook to toctree - Put an image about dataset info - Pushed the dataset to Hub
|
Hello @stevhliu, I made the changes you asked for.
|
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
View / edit / reply to this conversation on ReviewNB merveenoyan commented on 2024-04-30T13:32:17Z maybe you could give link to the base model :) emre570 commented on 2024-04-30T17:52:20Z I already did, it should direct you to model's HF page |
|
View / edit / reply to this conversation on ReviewNB merveenoyan commented on 2024-04-30T13:32:18Z nit: let's snake case the variable names for consistency with the rest of the recipe |
|
View / edit / reply to this conversation on ReviewNB merveenoyan commented on 2024-04-30T13:32:19Z maybe call push_to_hub explicitly as well |
|
View / edit / reply to this conversation on ReviewNB merveenoyan commented on 2024-04-30T13:32:20Z nit: scikit-learn |
|
View / edit / reply to this conversation on ReviewNB merveenoyan commented on 2024-04-30T13:32:21Z this is nice, but also we could put classification score because in this case we care about recall a lot (we don't want to miss malignant ones that look like benign or malign) |
|
I already did, it should direct you to model's HF page View entire conversation on ReviewNB |
|
Hello @merveenoyan, I made changes you asked. The notebook had issues. Your request for recall score saved nearly everything 😁. Some cells had errors and could've ruined all work. I made the notebook from scratch and fixed all code. It is fully working now and ready to use. |
|
View / edit / reply to this conversation on ReviewNB stevhliu commented on 2024-05-02T17:22:27Z The link to the image doesn't work. Can you upload it to https://huggingface.co/datasets/huggingface/cookbook-images and then link from there? |
|
One more comment, then we can merge! 🤗 |
- Opened PR and added the image.
|
Hey @stevhliu, I think I did it, opened a PR and uploaded the image, it should work now. |
|
@emre570 I just merged your PR to dataset repository |
|
Thanks folks, it was a pleasure ❤️ |
|
HUGE thanks folks, again, it was a pleasure ❤️ |
Hello folks @merveenoyan @stevhliu ,
I'm planning to make a notebook about fine-tuning a vit model with a custom biomedical dataset.
I have a code ready to use, made for my graduation project.
I used HF Datasets for dataset works, HF Transformers and Trainer for fine-tuning.
(Optional) Metrics won't show at training process momentarily, I can add a custom callback function.
If it's okay, i will push my code, then begin the editing for beginners. I am waiting for your opinions and suggestions.