Share an example notebook

[aadimator]

First of all, thanks for such an awesome implementation. One suggestion is to add a simple example training notebook so one can easily understand how to use this on an actual dataset like IMDB and such. That'd would be huge help. Thanks

[batrlatom]

What about this?

https://colab.research.google.com/drive/1rCZWPpFlgPZC_sqiUtKRSf16rScJi0JW?usp=sharing

I have prepared notebook for very bare 'object detection' . Would like to improve on it later on

@lucidrains Is this sufficient to show how perceiver works?

[lucidrains]

@batrlatom I think it would be better to run this on cifar-10 or 100!

[batrlatom]

I will do an example on cifar100

[lzmax888]

@batrlatom How do you determine your hyperparameters? Such as num_freq_bands and max_freq

BTW,
"
attn_dropout = 0.5,
ff_dropout = 0.5,
"
half dropped out?

[batrlatom]

It was just trial and error. It worked for me on mnist and my custom problem with object detection. Honestly, not so much on cifar. I am trying the original params now.