by This is the first in the series of simplified tutorial on PyTorch.
What is PyTorch?
PyTorch is a Machine Learning library that is based on the Torch library developed in FaceBook (Meta AI). It is used for developing models for Natural Language Processing and Computer Vision.
PyTorch is a leading Deep Learning framework similar to TensorFlow. Some of the existing deep learning applications are build on top of PyTorch. These include PyTorch Lightning, Uber’s Pyro, Tesla AutoPilot and Hugging Face’s Transformers.
At a high level, PyTorch provides the following features:
- Deep Neural Networks
- Tensor Computing with GPUs
PyTorch Tensors
PyTorch provides the torch.tensor class which enables storing and performing operations on multidimensional arrays (Tensors). Tensors are similar to Numpy arrays but can also be operated on a GPU (Graphics Processing Unit)
PyTorch Neural Networks
PyTorch defines the torch.nn module for building neural network models. This module provides a comprehensive collection of building blocks for neural networks. This includes layers and activation functions.
To create a neural network, you would need to extend the torch.nn module and override the forward() function which defines a sequence of operations.
Benefits of PyTorch
PyTorch offers the following benefits:
- Integration With Python – PyTorch is written in Python and can seamlessly integrate with well-known Python libraries including Numpy for numeric operations, Cython for compiling Python to C and SciPy.
- Cloud Support – PyTorch is supported in major cloud platforms including AWS, Azure and Google Cloud.
- Support for GPU and parallel processing
- PyTorch Hub – this is a repository of pre-trained models that can easily be invoked.
- Dynamic Computational Graph – PyTorch provides support for dynamic computational graph which allows the behaviour of the network to change at runtime.
