tnn | Biologically-realistic recurrent | Machine Learning library

Input:

I recently implemented a two-layer GRU network in Jax and was disappointed by its performance (it was unusable).

This is my minimal working example and the output was created on Google Colab with GPU-runtime. notebook in colab

I am a newbie in PyTorch.

I have implemented a custom model (based on a research paper), I get this error when trying to train it.

element 0 of tensors does not require grad and does not have a grad_fn

Here is my code for model:

I came across this simple problem, but I haven't found my way around it. I have two datasets (DS_clim and DS_yield), which I would like to compare across the three dimensions (time, lat, lon). However, their dimensions are not exactly the same, therefore I thought of using xr.dataarray.where to mask/crop both of them and therefore have the exact same dimensions. Funny enough, the output is still not compatible, with DS_yield having more datapoints than DS_clim. If anyone could help me make them identical in terms of dimension, I would really appreciate. I uploaded both .nc files and below you can find a self-standing piece of code that should replicated it.

Cheers!

Link for downloading the two files: https://drive.google.com/file/d/1gDSoKOY6eFLHqZ4AM0TTr4tXEBu3Y6yM/view?usp=sharing https://drive.google.com/file/d/1ysLqxNz-FBykJS2KojAx0UgTy6Hd9Wc2/view?usp=sharing

Good afternoon ,

Assume we have the following matrix :

I tried to make a copy of a neural network in pytorch and subsequently train the copied network, but training does not seem to change the weights in the network after copying. This post suggests that deepcopy is a convenient way to make a copy of a neural network, so I tried using that in my code.

The code below works just fine and shows that the weights and accuracy of the network are different after training from before training. However, when I toggle so that network_cp=deepcopy(network) and optimizer_cp=deepcopy(optimizer), the accuracy and weights before and after training are exactly the same.

I'm relatively inexperienced in C++, and am trying to compile a library to DLL for use in Unity. As far as generating the DLL and interfacing with it, I've had success in getting my below function to call and return a dummy value in Unity, no worries.

Where I'm struggling is working out how to interface with this Approximate Nearest Neighbour library, and could use some pointers. (This may be a hard problem to define without downloading the above-linked source and having a look yourself, but I will attempt to make it simple)

In Unity, I have a three-dimensional array (points[] = float[x,y,z]), and a query point in space (float[x,y,z]). I wish to feed this data into something like the below function (which I have modified from the ann_sample.cpp - original linked at the bottom), and return the nearest neighbour to this point.

My function