brax
Numba
Our great sponsors
brax | Numba | |
---|---|---|
11 | 124 | |
2,058 | 9,432 | |
3.3% | 1.8% | |
5.2 | 9.9 | |
10 days ago | 10 days ago | |
Jupyter Notebook | Python | |
Apache License 2.0 | BSD 3-clause "New" or "Revised" 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.
brax
-
4000x Speedup in Reinforcement Learning with Jax
There is Brax with its Ant, Humanoid and other rigid body articulated Gym environments: https://github.com/google/brax
- Physic engine for 3D simulation: which one to use?
-
Brax vs TDS for differentiable rigid body dynamics
I need differentiable rigid body dynamics because I want to do nonlinear MPC. One library that can do this is C++ is Tiny Differentiable Simulator https://github.com/erwincoumans/tiny-differentiable-simulator. As I understand it, this software uses a C++ auto-diff library and code generation to create CUDA kernels to compute fast derivatives in parallel. This seems pretty fast because it's C++. Another option is Brax https://github.com/google/brax. Brax uses JAX which I've never used, but from what I've seen online, JAX is popular for researchers and probably very good.
-
Deep learning for robotics
I am doing a MSc on robotics with a focus on machine learning, especially attention based architectures. There is a lot simulation and reinforcement learning going on. I have a funding of ~2500$ for the hardware system (no flexibility here, cannot use it for cloud either). I used pcpartpicker.com to select compatible components, as shown below. I am not located in the western part of the world; which means I have difficulty accessing some components and prices are higher here than that of pcpartpicker.com. That is why I am aiming towards 2000 - 2200$ range in the pcpartpicker.com. - Overall, what do you think of my planned setup? - Since there is a lot of simulation planned including rigid body dynamics with contact (libraries like https://github.com/raisimTech/raisimLib, https://github.com/deepmind/mujoco), I need some powerful CPU to use these libraries. I know that Intel has MKL over AMD; however, I am not sure how relevant that is for my case. The robotics simulators are generally written with C++, uses Eigen or their own math libraries. I feel like there is a lot of linear algebra involved and Intel combined with MKL should give me less headache. I have chosen i9-12900K, but what about AMD Ryzen9 5950X for example? - There is a new generation of rigid body simulators which use GPU instead of CPU (https://github.com/google/brax, https://developer.nvidia.com/isaac-gym). I do not think they are as mature as the previously mentioned simulators. Perhaps I am mistaken. Shall I focus on them instead? In terms of hardware that means I can downgrade the CPU to Ryzen5, and upgrade to RTX3080, roughly. - Do you think this system is easy to upgrade in future? What can I change to make it easier for long-term use and upgrades? Thanks for any help!
-
[D] Advice on Hardware Setup for Robotics
There is a new generation of rigid body simulators which use GPU instead of CPU (https://github.com/google/brax, https://developer.nvidia.com/isaac-gym). I do not think they are as mature as the previously mentioned simulators. Perhaps I am mistaken. Shall I focus on them instead? In terms of hardware that means I can downgrade the CPU to Ryzen5, and upgrade to RTX3080, roughly.
-
DeepMind open-sourcing MuJoCo simulator
I wonder what this means for the future of Brax [1].
1. https://github.com/google/brax
-
Any tutorial on how to create RL C++ environments?
If you want raw speed, parallel execution on GPU or TPU is best. Checkout out our Brax simulator, which uses the XLA compiler and JAX Python frontend: https://github.com/google/brax
-
Best environment to train RL agents
Check out Brax, hardware accelerated RL training in a Google Jupyter Colab. It trains typical RL tasks in minutes on TPU, also on GPU or CPU. And it is free, you can train with just a browser: https://github.com/google/brax
-
[N] Mujoco is free for everyone until October 31 2021
Anyone made any progress with Brax? That was sold as a massively-parallel Mujoco alternative but not sure if anyone's actually using it yet.
- [R] Brax: A Differentiable Physics Engine for Large Scale Rigid Body Simulation, with a focus on performance and parallelism on accelerators, written in JAX.
Numba
-
Mojo🔥: Head -to-Head with Python and Numba
Around the same time, I discovered Numba and was fascinated by how easily it could bring huge performance improvements to Python code.
-
Is anyone using PyPy for real work?
Simulations are, at least in my experience, numba’s [0] wheelhouse.
[0]: https://numba.pydata.org/
-
Any data folks coding C++ and Java? If so, why did you leave Python?
That's very cool. Numba introduces just-in-time compilation to Python via decorators and its sole reason for being is to turn everything it can into abstract syntax trees.
- Using Matplotlib with Numba to accelerate code
-
Python Algotrading with Machine Learning
A super-fast backtesting engine built in NumPy and accelerated with Numba.
-
PYTHON vs OCTAVE for Matlab alternative
Regarding speed, I don't agree this is a good argument against Python. For example, it seems no one here has yet mentioned numba, a Python JIT compiler. With a simple decorator you can compile a function to machine code with speeds on par with C. Numba also allows you to easily write cuda kernels for GPU computation. I've never had to drop down to writing C or C++ to write fast and performant Python code that does computationally demanding tasks thanks to numba.
-
Codon: Python Compiler
Just for reference,
* Nuitka[0] "is a Python compiler written in Python. It's fully compatible with Python 2.6, 2.7, 3.4, 3.5, 3.6, 3.7, 3.8, 3.9, 3.10, and 3.11."
* Pypy[1] "is a replacement for CPython" with builtin optimizations such as on the fly JIT compiles.
* Cython[2] "is an optimising static compiler for both the Python programming language and the extended Cython programming language... makes writing C extensions for Python as easy as Python itself."
* Numba[3] "is an open source JIT compiler that translates a subset of Python and NumPy code into fast machine code."
* Pyston[4] "is a performance-optimizing JIT for Python, and is drop-in compatible with ... CPython 3.8.12"
[0] https://github.com/Nuitka/Nuitka
[1] https://www.pypy.org/
[2] https://cython.org/
[3] https://numba.pydata.org/
[4] https://github.com/pyston/pyston
-
This new programming language has the potential to make python (the dominant language for AI) run 35,000X faster.
For the benefit of future readers: https://numba.pydata.org/
-
Two-tier programming language
Taichi (similar to numba) is a python library that allows you to write high speed code within python. So your program consists of slow python that gets interpreted regularly, and fast python (fully type annotated and restricted to a subset of the language) that gets parallellized and jitted for CPU or GPU. And you can mix the two within the same source file.
- Numba Supports Python 3.11
What are some alternatives?
mujoco - Multi-Joint dynamics with Contact. A general purpose physics simulator.
NetworkX - Network Analysis in Python
pybullet-gym - Open-source implementations of OpenAI Gym MuJoCo environments for use with the OpenAI Gym Reinforcement Learning Research Platform.
jax - Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more
tiny-differentiable-simulator - Tiny Differentiable Simulator is a header-only C++ and CUDA physics library for reinforcement learning and robotics with zero dependencies.
Dask - Parallel computing with task scheduling
cupy - NumPy & SciPy for GPU
RustyNEAT - Rust implementation of NEAT algorithm (HyperNEAT + ES-HyperNEAT + NoveltySearch + CTRNN + L-systems)
Pyjion - Pyjion - A JIT for Python based upon CoreCLR
open_spiel - OpenSpiel is a collection of environments and algorithms for research in general reinforcement learning and search/planning in games.
SymPy - A computer algebra system written in pure Python