search

lhnguyen102 / cuTAGI

CUDA implementation of Tractable Approximate Gaussian Inference
MIT License
24 stars 9 forks source link

Issue with removing omega_tol #57

Closed jamesgoulet closed 3 months ago

jamesgoulet commented 3 months ago

I have completed the implementation and tested the new formulation for all the mixture-based activation functions in all four files

  1. activation.cpp
  2. activation_fun_cpu.cpp
  3. activation_fun.cu
  4. activation_cuda.cu that can be found in the branch: New-formulation-mixture-activations

I have found a bug in the existing MixtureSigmoid() where the ma = ma/2 should have been done as a separate step. The result is that the number of epochs in the LSTM example test_lstm.py needs to be reduced to avoid NaNs.

  1. I have re-updated the unit tests that were all minimally changed by the new formulation.
  2. I have tested with test.py all activation functions with CPU and GPU.

The issue that I am running into is that I am unable to remove omega_tol from all the classes' inputs. I tried searching for all occurence in the cuTAGI repo and removing them, but I am then unable to compile and I do not have the experience to understand what I am doing wrong. If you can provide me with guidance with how to do so, I can finish this implementation. You will find below a screenshot of the errors I am running into when trying to compile after having remove the omega_tol everywhere:

Screenshot 2024-04-02 at 21 16 03
lhnguyen102 commented 3 months ago

@jamesgoulet Did you push the latest code to your branch New-formulation-mixture-activations? It compiled just fine on my side

jamesgoulet commented 3 months ago

@lhnguyen102 No, have deleted because I was afraid I would break something... I can redo it if you want?

lhnguyen102 commented 3 months ago

@jamesgoulet could you push the latest code that caused this error to your New-formulation-mixture-activations? Otherwise I wont be able to troubleshoot the issues. As long as we dont merge to main, it wont break main branch

jamesgoulet commented 3 months ago

I re-did it from scratch and now it compiles. I will prepare the PR.