kirushyk / le Goto Github PK

cuBLAS uses column major order only

We need combobox element added to left part of UI to specify training and test sets size.

Now SVM is very slow on datasets with many examples.
For MNIST, we need few minutes just to pass 100 of 60000 examples.

We want Neural Networks to report gradients for Gradient Descent optimizers to use.

I fetched the project in case it can help me downsize my ML and still keep it portable. I was pleasantly surprise to see it build on MacOS after fetching bits and pieces from "brew", but the latest Ubuntu (tried both gcc9 and gcc10) gave me

/home/ubuntu/le/bin/../le/tensors/letensor.c:38: undefined reference to LE_ERROR'
`

On the Mac, I had no luck with the MNIST examples, it might be something trivial but I don't know at this stage

./mnist-snn Segmentation fault: 11

I also don't know if this is expected behavior

./polynomial-logistic-regression Train set: x = [1.000 2.000 3.000 4.000; 4.000 3.000 2.000 1.000] y = [0.000 0.000 1.000 1.000; 4.000 3.000 2.000 1.000] Assertion failed: (le_shape_equal(h->shape, y->shape)), function le_logistic_loss, file ../le/leloss.c, line 17. Iteration 0. Abort trap: 6

Sequential model needs gradient checking test.

We need it written in C, Python and Rust.

Sequential is not enough this days...

We need to provide demo code for those ones who will try our framework:

• Polynomial Regression;
• Regularization;
• SVM;
• MNIST-related stuff;
• Dense Neural Network.

Is there a way to compile the frame work to ONLY use SVM algorithm on STM32F103C8T6 board?
Any instructions to install/compile this library will be encouraged.
Thanks

Everything should be easy to read and understand. Ideas are welcome here.

https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/tr-98-14.pdf

Also, maybe use secure_getenv .

We need examples on all the languages supported.

We need some acceleration on Linux.

Some ML Researches use Microsoft Windows.

At this moment, all gradients and weights going to NAN soon. First fixed error was stride not being taken into account when copying signal in Sequential model.

We need to provide demo code for those ones who will try our framework:

• Polynomial Regression;
• Regularization;
• SVM;
• MNIST-related stuff;
• Dense Neural Network.