Mirror of forum.nim-lang.org

3214 :: Hacktoberfest project contributions?

• perturbation2 (orginal) [2017-10-02T01:23:16+02:00] view original

  Are there any packages / issues that would be good candidates for Hacktoberfest? That is, issues that are pretty self-contained or good introductions to the package?

  I have my eye on NimData, Neo, and ArrayMancer for myself, but it might be worth adding the "Hacktoberfest" label to issues on your project so that others participating can find them.

  I know we already have the "Easy" label for the Nim project itself, so I think it shouldn't be too hard to find good Hacktoberfest issues from there.

---

• dom96 (orginal) [2017-10-02T12:31:14+02:00] view original

  Be sure to also look at the nimble and choosenim issues. There aren't many and they're pretty well labelled so you can easily go through them all.

---

• andrea (orginal) [2017-10-03T13:16:44+02:00] view original

  Hi perturbation2! I have not had much time to work on Neo lately, but there is a list of issues which I use to track features that I had planned to add. Most should be rather self contained. If you want to contribute, I will be glad to help :-)

---

• mratsim (orginal) [2017-10-04T09:30:41+02:00] view original

  @perturbation, Thanks for your interest.

  Many of the low hanging fruits in Arraymancer are done but I could use your help in implementing reduction functions like "max", "min", and in-place functions like msqrt, msin, as mentioned here: https://github.com/mratsim/Arraymancer/issues/58

---

• perturbation2 (orginal) [2017-10-04T21:51:16+02:00] view original

  Thanks guys! I'll take a look at those issues this weekend and let you know if I have questions. :)

---

• bpr (orginal) [2017-10-06T03:42:32+02:00] view original

  From their docs, it looks like neo and Arraymancer are both intended to be substitutes for NumPy. Can someone explain why I would prefer one over the other?

  Nim could be very useful in data science and scientific computing; it's good to see some libraries sprouting up.

---

• andrea (orginal) [2017-10-06T10:11:36+02:00] view original

  I will reply for Neo and then let mratsim speak for ArrayMancer.

  Neo was born first (at least, its precedessor linalg was). I tried to give it an interface more familiar for people coming from linear algebra (everything there is expressed in vectors and matrices), as opposed from the popular Numpy approach where everything is a n-dimensional tensor. Not that tensors are not from mathematics, but - well, there are many kinds of tensors in mathematics, they are a bit more complicated than simply an n-dimensional table, so I left that part for later. I tried to focus on providing bindings to the BLAS, LAPACK and CUDA libraries, in order for the actual computations to run fast, and tried to postpone abstractions such as tensors for later (I have not decided on the interface yet).

  For similar reasons, only lately I introduced operations such as the element-wise product of vectors, which do not have a natural mathematical interpretation (but it is there at last). Also, I tried to design Neo in such a way that it could accomodate matrices and vectors allocated on the stack.

  My recent interest is in trying to adapt Neo to work with sparse matrices, both on the CPU and GPU side - which is complicated, since - unlike BLAS for dense matrices - there is not a common approach on the two sides. I would also like to focus on matrices decompositions and other features from LAPACK (but unfortunately, these are not always implemented on CUDA).

  I also moved the actual libraries bindings into separate packages: CUDA, BLAS, LAPACK so that they could be used by other libraries in the Nim ecosystem.

  All that said, work on Neo is now quite slowed down. In general, I had not much time to work with Nim lately - I still have plans to go on with Neo, but I have made no actual progress in the latest few months. ArrayMancer is currently more active, but I will let mratsim speak about its features.

---

• mratsim (orginal) [2017-10-07T10:27:13+02:00] view original

  Arraymancer was born in-between linalg and neo and is targeted at machine learning, with a focus on deep learning and computer-vision (and later speech and language hopefully).

  I started Arraymancer because you need 4-dimensional tensors for deep learning on images (Number of images, RGB colors, height, width) and 5d for videos or 3D images.

  It takes a Numpy/Julia/Matlab like multidimensional array approach meaning operations like reshaping, concatenating arrays over an arbitrary dimensions, permuting dimensions, working with stacked 2d matrices in a 3d tensors is possible and efficient.

  ---

  There are a few twists:

  • Like Andrea, I like linear algebra operations like + and * to mean math (linear algebra) things, so broadcasting and element-wise matrix product is not implicit like in Numpy but opt-in: you must use custom operators .+ and .*. The benefit is that if you have strange results you can grep those.

  • Like Nim, I choose to have value semantics for my tensors, meaning it's copy on assignation. You can opt-in to use "unsafe" operations like "unsafeReshape" or "unsafeTranspose" that will provide a view over the same data.

    It avoids a lot of beginners gotchas and like broadcasting makes it easy to grep for potential issues.

    Note: For CudaTensors I currently can't due to = overloading issues.

  ---

  My short-term plans (consolidating the base):

  • Convolutional neural networks are just around the corner
  • Improve CudaTensors support
  • Doing demos
  • Cover common deep-learning use cases outside of Recurrent Neural Networks.

  Mid-term plans (focus on data):

  • CSV loading and/or conversion from a Nim dataframe package like NimData
  • Basic image manipulation (rotating, flipping, blurring ...), in an extra package like Arraymancer-vision.
  • Saving and loading models from disk.
  • Provide popular datasets (IRIS, MNIST ...)
  • Recurrent neural networks.

  Long-term plans (hardware and format):

  • More hardware support (AMD GPUs at least)

  • Provide utilities for general machine learning (and not only deep learning):

    • Sparse tensors but like Andrea said, apart on Cuda there is no standard :/
    • PCA and SVD matrices decomposition
    • Clustering tools like KMeans, DBScans

  Very long-term plans (wishlist):

  • Bindings to gradient boosted trees libraries (XGboost / LightGBM)
  • Videos loading and analysis
  • New backends: OpenCL, Javascript, Apple Metal

  Very (very !) long-term plans (wishlist reloaded):

  • More natural language utilities or bindings like spaCy
  • REST API for example, send Arraymancer an image and it returns what was detected in it.
  • Bayesian Neural Networks which works with uncertainty built-in

  What is not a focus:

  • Linear algebra/LAPACK not used in Machine Learning (like solving equations)
  • 2D/3D geometry and quaternions (nim-glm is probably more suited than either neo or Arraymancer)

---