Create a tutorial on Colab

[davidbuniat]

Create a tutorial on Colab

Users should be able to load a dataset, train a model, and upload the dataset. Feel free to start from a small example and then make the example comprehensive.

[Paola351]

I can do this i you need help

[keshav340]

can I do this? It is easy for me.

[davidbuniat]

can I do this? It is easy for me.

Assigning you as well, to avoid replication of the task, please think of the extended introduction of Hub to an user.

[Paola351]

Sorry, i have to give up as I'm very busy in these days, sorry!

[davidbuniat]

Sorry, i have to give up as I'm very busy in these days, sorry!

No worries at all, thanks for thinking of us!

[Tanujcbe]

@davidbuniat Can I work on it?

[davidbuniat]

Thanks for your willingness to contribute. @Tanujcbe assigning you the task.

[Tanujcbe]

@davidbuniat Please accept pull request #83

[davidbuniat]

@Tanujcbe thanks for the PR. Probably the issue was not clear enough from the beginning.

We are looking for a Hub tutorial on Google Colab to demonstrate the user Hub in action, e.g. how users can create datasets and load them into pytorch/tensorflow.

Finally we look for adding the created notebook into the readme.

[thisiseshan]

Hi. Is this issue resolved? I find this repo interesting and would wish to contribute to this issue.

[davidbuniat]

@thisiseshan thanks for your interest, no this issue is still pending, would love your help!

[thisiseshan]

Thank you for assigning this to me, would MNIST classification be an appropriate example?

[muskanlalit18]

Hi, this looks interesting. I would like to work on it too

[davidbuniat]

Thank you for assigning this to me, would MNIST classification be an appropriate example?

@thisiseshan yes, as long as you the tutorial demonstrates in very simple terms the main values of the package Mnist is fine. @muskanlalit18 assigning you the task as well, please feel free to choose another dataset.

[drashtipatel2503]

Hii, can I go for this?

[muskanlalit18]

@muskanlalit18 assigning you the task as well, please feel free to choose another dataset.

Thanks for assigning it to me, should I try CIFAR classification example?

[AbhinavTuli]

Sure @muskanlalit18 CIFAR sounds good! Feel free to post here or in the discussions https://github.com/activeloopai/Hub/discussions if you need any help.

[thisiseshan]

Hi, I am training my model on MNIST dataset but it's taking way too long on Colab (3 hrs per epoch). The model seems to be running fine.

I wanted to know

• Do datasets from hub take a bit longer to train?
• Should I change the model?
• Should I proceed to the upload dataset part and send a PR so you can review it?

[davidbuniat]

Hi, I am training my model on MNIST dataset but it's taking way too long on Colab (3 hrs per epoch). The model seems to be running fine.

I wanted to know

• Do datasets from hub take a bit longer to train?
• Should I change the model?
• Should I proceed to the upload dataset part and send a PR so you can review it?

It is not supposed to take that long, can you please proceed with the PR with collab link. I will give a try and see what's wrong.

[thisiseshan]

Hi! Please review my PR and let me know your thoughts. I can make necessary changes.

[davidbuniat]

@thisiseshan added comments there, please also take a look at other examples of colabs to demostrate core values of hub.

• TFDS
• Transformers

[thisiseshan]

Hi, Current functionality implemented to Colab is: [PyTorch]

• Load MNIST dataset using Hub
• Create Model
• Train model on dataset
• Print Accuracy of Model from Test subset

Should the 'upload dataset' be integrated to this notebook. If Yes, What dataset is to be uploaded as an example (should I use the same example as in docs)?

from hub import dataset, tensor

tensor1 = tensor.from_zeros((20,512,512), dtype="uint8", dtag="image")
tensor2 = tensor.from_zeros((20), dtype="bool", dtag="label")

dataset.from_tensors({"name1": tensor1, "name2": tensor2})

dataset.store("username/namespace")

[davidbuniat]

@thisiseshan for uploading to work, the user would need to authenticated inside collab. We don't have yet token so my guess would be tricky to do with colab.

I think this would be out of scope for the first tutorial. We can do a separate PR for uploading the dataset with colab and add any required functionality :)

[thisiseshan]

Okay perfect!

[thisiseshan]

Hi. I have made the fixes. Please let me know your thoughts and any changes you need :)

[mikayelh]

Hi, @muskanlalit18 ! Hope this finds you well.

Dropping a note to check in on you and ask if you need a hand with uploading the dataset. Feel free to ask us in the GitHub Discussions (we have beta access!) or our dedicated Slack channel. Thanks a mil!

[mikayelh]

Hi @thisiseshan! Thanks so much for following up. We will be reviewing the code shortly. Meanwhile, if there is any other task in the project board you'd like to work on, feel free to let us know!

[thisiseshan]

Sure, Thank you!

[mikayelh]

thank YOU, @thisiseshan :)

[muskanlalit18]

Hi, @muskanlalit18 ! Hope this finds you well.

Dropping a note to check in on you and ask if you need a hand with uploading the dataset. Feel free to ask us in the GitHub Discussions (we have beta access!) or our dedicated Slack channel. Thanks a mil!

Hi, I'm working on it, I shall ask in the Discussions if I need any help.

[mikayelh]

Hi, @muskanlalit18, thanks!

[davidbuniat]

@muskanlalit18 please take a look at @thisiseshan contribution colab and feel free to think of another exciting tutorial to demonstrate hub capabilities. e.g. uploading a dataset or more advanced use case such as training coco.

[mynameisvinn]

Closing this due to inactivity, will reopen if theres widespread interest.