ragged-buffer
rust-numpy
ragged-buffer | rust-numpy | |
---|---|---|
2 | 10 | |
19 | 1,019 | |
- | 2.1% | |
3.8 | 8.0 | |
about 1 year ago | 17 days ago | |
Rust | Rust | |
Apache License 2.0 | BSD 2-clause "Simplified" License |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
ragged-buffer
-
Entity Gym: A new entity based API for reinforcement learning environments
We are also releasing enn-trainer, a PPO implementation that takes full advantage of the Entity Gym interface. Variable-length observations are efficiently processed using ragged sample buffers and a general ragged batch transformer implementation that can be applied to any Entity Gym environment. With many performance optimizations still missing, enn-trainer can already reach a throughput of 10s of thousands of samples per second per GPU when it is not bottlenecked by stepping the environment. More typically, environments implemented in Python reach thousands of samples per second, but can share a single GPU between multiple concurrent training runs.
-
Writing Rust libraries for the Python scientific computing ecosystem
One of Rust's many strengths is that it can be seamlessly integrated with Python and speed up critical code sections. I recently wrote a small library with an efficient ragged array datatype, and I figured it would make for a good example of how to set up a Rust Python package with PyO3 and maturin that interoperates with numpy. There are a lot of little details that took me quite a while to figure out:
rust-numpy
-
Numba: A High Performance Python Compiler
On the contrary, it can use and interface with numpy quite easily: https://github.com/PyO3/rust-numpy
- Carefully exploring Rust as a Python developer
-
Hmm
Once I figured out the right tools, it was easy. Its just "maturin new". It automatically converts python floats and strings. Numpy arrays come through as a special Pyarray type, that you need to unwrap, but that's just one builtin function. Using pyo3, maturin and numpy, https://github.com/PyO3/rust-numpy it's fairly easy.
-
Man, I love this language.
If I'm understanding this documentation correctly then you may be able to pass the numpy array directly with func(df['col'].to_numpy) which may save some conversion.
-
[D] Is Rust stable/mature enough to be used for production ML? Is making Rust-based python wrappers a good choice for performance heavy uses and internal ML dependencies in 2021?
Otherwise, though, Rust is an excellent choice. The many advantages of Rust (great package manager, memory safety, modern language features, ...) are already well documented so I won't repeat them here. Specifically for writing Python libraries, check out PyO3, maturin, and rust-numpy, which allow for seamless integration with the Python scientific computing ecosystem. Dockerizing/packaging is a non-issue, with the aforementioned libraries you can easily publish Rust libraries as pip packages or compile them from source as part of your docker build. We have several successful production deployments of Rust code at OpenAI, and I have personally found it to be a joy to work with.
-
Writing Rust libraries for the Python scientific computing ecosystem
Integration with numpy uses the rust-numpy crate: Example of method that accepts numpy arrays as arguments Example of a method that returns a numpy array to Python (this performs a copy, there ought to be a way to avoid it but the current implementation has been plenty fast for my use case so far)
-
Feasibility of Using a Python Image Super Resolution Library in My Rust App
This example maybe helpful.
-
Julia is the better language for extending Python
Given that it's via pyO3, you could even pass the numpy arrays using https://github.com/PyO3/rust-numpy and get ndarrays at the other side.
Same no copy, slightly more user friendly approach.
Further criticism of the actual approach - even if we didn't do zero copy, there's no preallocation for the vector despite the size being known upfront, and nested vectors are very slow by default.
So you could speed up the entire thing by passing it to ndarray, and then running a single call to sum over the 2D array you'd find at the other end. (https://docs.rs/ndarray/0.15.1/ndarray/struct.ArrayBase.html...)
-
Parsing PDF Documents in Rust
I believe converting between pandas Series (e.g. columns) and numpy ndarrays can be pretty cheap, right? Once they're in that format, you can use rust to work directly on the numpy memory buffer with rust-numpy. Otherwise, feather is a format designed for IPC of columnar data; pyarrow is in pandas (might be an optional dependency) and may be pretty quick for that, and rust has an arrow implementation too.
-
PyO3: Rust Bindings for the Python Interpreter
https://github.com/PyO3/rust-numpy
What are some alternatives?
maturin-action - GitHub Action to install and run a custom maturin command with built-in support for cross compilation
RustPython - A Python Interpreter written in Rust
rogue-net - Entity Gym compatible ragged batch transformer implementation.
julia - The Julia Programming Language
enn-trainer - Reinforcement learning training framework for entity-gym environments.
polars - Dataframes powered by a multithreaded, vectorized query engine, written in Rust
entity-gym - Standard interface for entity based reinforcement learning environments.
rayon - Rayon: A data parallelism library for Rust
PyO3 - Rust bindings for the Python interpreter
image-super-resolution - 🔎 Super-scale your images and run experiments with Residual Dense and Adversarial Networks.
enn-zoo - Collection of entity-gym bindings for different reinforcement learning environments.