Facebook’s AI team is bringing GPU-powered machine learning to Python. The team has released a Python package that can either supplement or partially replace the existing Python packages like NumPy. The Python package named PyTorch has been well received by Twitter, Salesforce, Facebook, Carnegie Mellon University etc. PyTorch is a deep learning research platform offering maximum flexibility and speed. It provides GPU acceleration for many functions. While Torch is wrapped in Lua scripting, PyTorch wraps the core Torch binaries in Python.
Advantages of PyTorch
Torch works as a tensor library for working on different data matrices that are multidimensional (utilized in machine learning and math intensive applications). PyTorch offers libraries for performing basic tensor manipulations on CPUs, GPUs etc. Moreover, it functions as a multiprocessing library that can function with shared memory. In a developer’s jargon, PyTorch is “useful for data loading and hogwild training.”
Additionally, PyTorch also offers the following benefits:
- Enables developers to tune into the extensive ecosystem of Python libraries as well as software
- Encourages Python programmers to utilize the styles they are used to rather than writing code to create a wrapper for an external library
- Supports most of the existing packages- NumPy, SciPy, Cython
- Ability to modify the current neural networks without having the necessity to rebuild from scratch
- PyTorch’s remarkable memory efficiency owing to a custom-written GPU memory allocator
- With its memory adeptness, PyTorch allows you to train huge deep learning models
- PyTorch is optimized for machine learning and beyond
- Capacity to drop back to CPU if GPU is unavailable
With all the above features, PyTorch becomes a comfortable coding option for Python developers and enthusiasts.
Related Posts:
- Guide to languages Python and R: Machine learning and data analysis
- Microsoft Cognitive Toolkit’s new version launched with Python support
- Big data, Python and the fight against Human Trafficking
