I'm back

Hello. It's been a a while. Not yet back to normal, but as normal as it can be.

I'm here to tell about an error I got while trying to run the TempoGAN on a brand new server. The error happens when "session.run" is called, and it gives:

Internal: Blas SGEMM launch failed : m=216000, n=32, k=4

If you search on the internet, everyone will tell this is due to lack of memory in the GPU. Ok, this net is quite large but... I have 48GB available to me! This net can't be that greedy.

So, what did I found? Yes, it needs more memory in the graphics card... BECAUSE IT IS NOT ALLOCATING ANY!  The code is from 2018, the GPU architecture is modern, and cuBLAS from CUDA 10 does not know how to do its job.

I'm sorry, but you and I will have to upgrade the code to Tensorflow 2.X.

Good luck for us. 

Change of plans

In the last post I said I would update the tutorial of the deep learning API. But an urgent matter came first: learn to use SHAP. The subject will be a Toy Model for predicting the area of a 2D triangle based on its points.

The tutorial is not ready yet, but you can check it every week for new content, and the code is not open source yet.

When there is more updates, I'll let you know.

See ya.

Problems with PyTorch in Conda

So... I made a tutorial for using the Deep Lagrangian Fluids, but the training gets stuck in what I suppose it is a deadlock. However, the training works fine with Tensorflow. I know I said it was impossible to install the version I needed via conda (still is), but via pip, it is possible.

I'll post the tutorial update ASAP. There are some differences in the steps. I'm thinking about how to organize them.

New page released

I just made a tutorial to build and install all the dependencies to run the Deep Learning network I'm studying for fluid simulation, right here. For now, I can just execute it and visualize the results. The presentation video can be remade on your computer if you follow all the steps, no skipping or changing the order (please).

Soon I'm gonna find out the next step.

If you are interested on this topic and you want to work with Disney, check out the lib this project uses: Partio. And also, the SPlisH SPlasH that uses it.

I'm Back

Hello there. Almost a year, huh? 

Last time I stopped because I needed to generate data for a paper, which is now at the last step to be accepted. It is about Rigid Registration with Lie Groups. When it is released, I'll post here.

Since then, I started to research about Deep Learning for this topic. But I found out it is very complicated when you don't know much of the theory. 

For instance, I made a Toy Model with a Multilayer Perceptron to compute the triangle area. However, I couldn't make it have a good accuracy. 

So the task was assigned to another student who knows better about neural networks, and he concluded that triangles with very small area are being interpreted as outliers by the network, and making the training harder. Also, there is an error accumulation due to numerical instability, so the training needed 64 bit float and the error measurement, a 32 bit float.

Right now, I changed my research to fluid simulation with Deep Learning, far more interesting to me. I just made this code work, and very soon I'll post the step by step to make it work for you too.

See ya.