Training KAN on a diabetes classification dataset

[Chukwuebuka-2003]

Now, I tried to train the KAN on a dataset that deals on classification.

My question are these:

Now when I tried to run this:

This was the error that I was seeing: ValueError: array must not contain infs or NaNs

I'm currently stuck here:

[KindXiaoming]

hi, does model.train() work well (doesn't show any NaN)? Can you nail down to whether the problem comes from train or auto_symbolic?

[ChrisD-7]

@KindXiaoming i think his issue is with the auto_symbolic function cuz it looks like his model is trained and pruned

[Chukwuebuka-2003]

@KindXiaoming I do think that this issue might be from the auto_symbolic.

Like @ChrisD-7 said, the model has been trained and pruned.

[ChrisD-7]

@Chukwuebuka-2003 are u facing an issue with the dataset? like what the size is when building this classifier I have a larger dataset and run an issue when passing all features through it so I tried loading it through my GPU and faced an issue how big is ur dataset?

[ChrisD-7]

yup the low visualization is not that good for classification probs

[Chukwuebuka-2003]

@ChrisD-7
The dataset that I used isn't big but a little dataset used to train and test classification models. I will share my notebook

[ChrisD-7]

I ran a model and will share it as well will link my repo

[ChrisD-7]

@Chukwuebuka-2003 take a look : https://github.com/ChrisD-7/Fraud-Detection-in-Supply-Chains-with-Kolmogorov-Arnold-Networks

[ChrisD-7]

@Chukwuebuka-2003 were u able to run it ??

[Chukwuebuka-2003]

@ChrisD-7 I was able to run it, but I came across related errors

Here's the link to my repo on this: https://github.com/Chukwuebuka-2003/kan_diabetes

[ChrisD-7]

@Chukwuebuka-2003 let me take a look

[ChrisD-7]

@Chukwuebuka-2003 I have submitted a pull request. Please let me know if there are any additional changes needed:

@KindXiaoming I believe this issue can now be closed. The model is functioning well and achieved around 80% accuracy. I encountered some issues with converting the formula to NumPy, but I managed to resolve them.