pykan
pykan copied to clipboard
KindXiaoming
Kolmogorov Arnold Networks
Hi, I have a question about extracting symbolic activation functions after fixed symbolic for each activation functions in KAN models. The tutorial shows how to get the activation function values...
In the `KAN.train` function, the dataset is sent to the device within the function. However, for `KAN.initialize_from_another_model`, the dataset needs to be sent before calling the function. This PR allows...
--------------------------------------------------------------------------- AssertionError Traceback (most recent call last) Cell In[26], [line 10](vscode-notebook-cell:?execution_count=26&line=10) [7](vscode-notebook-cell:?execution_count=26&line=7) device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') [8](vscode-notebook-cell:?execution_count=26&line=8) print(device) ---> [10](vscode-notebook-cell:?execution_count=26&line=10) model = KAN(width=[15,6,3,1], grid=5, k=3,device=device) #, grid_range=(0,1))...
## Description: This issue proposes an enhancement to the Kolmogorov-Arnold Networks (KAN) architecture that involves the development of an automated method for converting trained models into symbolic activation functions. This...
I am trying to use a massive CSV (1M+ rows), with the X input having 400 dimensions. It's technically working but I had two main questions: 1. It seems like...
In certain applications (e.g., causal inference) we also want to measure the output in its natural units, pre-normalization. In this case we need something like this in the symbolic_formula method:...
model(dataset['train_input']) model.plot(beta=10) --------------------------------------------------------------------------- RuntimeError Traceback (most recent call last) Cell In[51], [line 1](vscode-notebook-cell:?execution_count=51&line=1) ----> [1](vscode-notebook-cell:?execution_count=51&line=1) model(dataset['train_input'].double()) [2](vscode-notebook-cell:?execution_count=51&line=2) model.plot(beta=10) File c:\Users\thnog\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\nn\modules\module.py:1532, in Module._wrapped_call_impl(self, *args, **kwargs) [1530](file:///C:/Users/thnog/AppData/Local/Programs/Python/Python311/Lib/site-packages/torch/nn/modules/module.py:1530) return self._compiled_call_impl(*args, **kwargs) # type:...
adds the possibility of normalizing the outputs of a symbolic fn as well. ```python import torch import numpy import pandas n = 10000 w = numpy.random.normal(0, 1, n) x =...
Many thanks to the author for proposing this amazing KAN. I modified the KANLayer.py file for better compatibility in non-CUDA environments. BTW, it is my first pull request. I would...
Is it acceptable to use KAN as a way to approximate f(x) when computing expressions such as y = g(f(x)) over other layers or functions rather than over a function...
