Ergonomic ndarrays and deep learning in Nim – Arraymancer v0.5 (github.com)

Hey library author here, you can ask me anything.

I just released the 0.5 version of my multidimensional arrays (Numpy-like) + Deep Learning library (PyTorch-like) that I've written from scratch in Nim.

Key highlights of this version:

  - Sequence/time series prediction end-to-end example[1]
  - Text generation with Char-RNN on Shakespeare and Jane Austen work end-to-end example[2]
  - IMDB dataset
  - read and write: Numpy .npy files, images (jpg, png, bmp, tga) and H5
  - KMeans clustering
  - GRU, Embedding layers
  - Adam optimizer
  - Yann Lecun, Xavier Glorot and Kaiming He initialisations
  - fancy indexing
  - tensor splitting, chunking stacking with autograd support

And in the ecosystem:

  - a neural network training demo with live input and loss monitoring[3]
  - Nim wrapper for the Arcade Learning Environment to agent on Atari games[4]

Nim[5] is a high performance compiled language with a syntax similar to Python. Nim compiles to C, C++ or Javascript.

  [1] https://github.com/mratsim/Arraymancer/blob/v0.5.0/examples/ex05_sequence_classification_GRU.nim
  [2] https://github.com/mratsim/Arraymancer/blob/v0.5.0/examples/ex06_shakespeare_generator.nim
  [3] https://github.com/Vindaar/NeuralNetworkLiveDemo
  [4] https://github.com/numforge/agent-smith
  [5] https://nim-lang.org/

Motivation? I mean why should I click on the link, let alone consider yet another DL library?