Mahmoud Asem
Mahmoud Asem
Maybe adding tree util cookbook would be useful? @jakevdp
Hello, Any updates or feedback? Additionally, I implemented a Gaussian filter based on FFT depthwise convolution, which should be faster for large kernels. https://github.com/ASEM000/serket/blob/main/serket/nn/blur.py Let me know if you are...
Check [here](https://github.com/ASEM000/PyTreeClass) for generalized concept based on lenses, and [here](https://pytreeclass.readthedocs.io/en/latest/notebooks/common_recipes.html#%5B7%5D-Use-PyTreeClass-with-Flax/Equinox) on how to integrate with flax/equinox (or any registered pytree)
The example below should answer the questions, notably: 1) it can work on dict/class. the distinction between dict/class attribute is dealt with on the path entry level defined using jax...
> As a user, I think I'd prefer the lambda approach to the chain of at because it is more concise and direct. You can try this if you do...
`jax` from 0.4.6, supports named leaves for [pytrees](https://jax.readthedocs.io/en/latest/_autosummary/jax.tree_util.tree_map_with_path.html). So it's up to @patrick-kidger to support it. If adopted, then this will simplify a lot of tree manipulation in equinox (...
Nice :D Once this is merged, you can use -shameless plug- [pytreeclass](https://github.com/ASEM000/PyTreeClass) to add functional and composable setters/getters (~lenses like) to your equinox trees to modify values based on their...
I think `tree_at` needs a makeover, possibly to match something like the above example. `tree_at` did a good job when as you said, named leaves did not have a meaning...
I think maybe you don't need to check for `if field.init` , AFAIK the field should be initialized at that point of your code, even if `init=False`. So that line...
> it seems to me that it performs 0-padding. Is this the case? Yes, > If yes, it would be useful to support different schemes like "mirror", "edge", or ideally...