Chapter 1- PyTorch for Beginners: Basics
We are living in a tech-driven world where each year many new technologies come and serve the people. So it's important for everyone to keep up to date with these new technologies. Artificial Intelligence is the most trending field in this era. Every company tries to integrate AI with their machinery.
Training a model requires a large amount of data to perform the model really well. So where it comes Deep learning that is used in a variety of tasks like image translation, image captioning, predicting next sentences, generating new images, and many more.
Pytorch is the best choice when you are starting deep learning.
Taking into account all the advantages of knowing PyTorch, we have decided to write a series of blog posts on Deep Learning with PyTorch. We are going to start with the first-day tutorial which is PyTorch basics.
What is PyTorch
Pytorch is a deep learning framework and a scientific computing package. This is how the PyTorch team defines it. Originally torch was built on Lua programming language and for the ease of use, it is converted in Python by the Facebook AI research teams and many others.
It’s a Python-based scientific computing package targeted at two sets of audiences:
1- A replacement for NumPy to use the power of GPUs.
2- A deep learning research platform that provides maximum flexibility and speed
PyTorch uses Tensor as its core data structure, which is similar to Numpy's ndarrays. If you are wondering about this specific choice of data structure, the answer lies in the fact that with appropriate software and hardware available, tensors provide acceleration of various mathematical operations. These operations when carried out in a large number in Deep Learning make a huge difference in speed.
Why should I learn PyTorch?
In the previous block, we learn what is PyTorch and in this section, we learn about why should I learn PyTorch.
There are many deep learning frameworks that are available other than PyTorch such as Keras, Tensorflow, Mxnet, Caffe, and many more. But what makes PyTorch different?
The goal of the PyTorch is to have maximum flexibility and speed in building our scientific algorithms while making the process extremely simple.
There are many features of PyTorch are
1- Pytorch offers native supports for Python and the use of all of its libraries and in this way, it becomes "Pythonic" in nature.
2- It is actively used by the huge MNC's like the Facebook AI team along with many other subsidiaries.
3- Pytorch APIs are easy to use due to its easy nature it is used many Ph.D. scholars and researchers for their research purpose.
4- The main features oaf PyTorch is Imperative programming which means the program describes steps that change the state of the computer. And in this way, it generates the graph at each step making graph dynamic in nature.
Overview of Pytorch libraries
The most important PyTorch libraries are torch.nn, torch.optim, torch.utils and torch.autograd.
Let us see what is the use of all these libraries
1- Loading Dataset
The first step in any machine learning or deep learning project is we have to load the dataset and handling the dataset.
There are 2 important classes in the torch.utils.data are:
1- Dataset: It is used for loading custom datasets and built-in datasets.
2- DataLoader: It is used for loading large datasets parallel. It gives us the option to shuffle data, determining the batch size, and the number of workers to load data in parallel.
2- Defining the neural network
To define the neural network we torch.nn module. It helps to set up the neural network layers like fully connected layers, convolutional layers, activation and loss functions, etc.
After that, we have to update the weight and biases so that our neural network can learn. For this task we use torch.optim module. Now we perform a backward pass to compute the gradient of input Tensors with respect to that same scalar value. This can easily be done torch.autograd module.
3- Performing inference and converting into other dl frameworks
Finally, we save the model by using torch.save the module and then we load the model for further predictions. This process is referred to as model inference.
You can only convert your Pytorch model into the ONNX model so that other deep learning frameworks can use like MXNet, CNTK, Caffe2.
Other Pytorch libraries
torchaudio: It is an audio library for PyTorch that is used to deal with audio data and perform audio preprocessing and deploy it on production.
Some examples are Cornell BirdCall Identification, UrbanSound8k, and many others.
torchtext: It is used for text data and it is majorly used in natural language processing tasks. It provides many modules for text preprocessing tasks.
Some examples are Sentiment Analysis, Question- Answering, and many others.
torchvision: It deals with image data and its transformation. It is used in computer vision and deep learning.
Some examples are MNIST, COCO, CIFAR, and many more.
torchserve: It is used to deploy our machine learning model to production.
Introduction to Tensors and it's operation
Tensors are similar to NumPy’s ndarrays, with the addition being that Tensors can also be used on a GPU to accelerate computing.
If the tensor is of 1 dimension it is called 1-D tensors, similarly, If the tensor is of 2 dimensions it is called 2-D tensors, If the tensor is of 3 dimensions it is called 3-D tensors, Similarly, If the tensor is of n dimensions it is called n-D tensors.
Let us created our first tensors using PyTorch. So the first step is to import our torch library.
In the above block of code, we have shown the examples of a 1D tensor. In the next block of code, we will code tensors of more than 1 dimension.
Accessing elements in a tensor
To access an element in a tensor we use square bracket [ ] and specify the index position or range of values in that bracket.
Converting Numpy to tensors and vice-versa
In the above section of code, we learn that how to create tensors and access its elements, in this section we will how to convert a tensor to Numpy and vice-versa.
Performing Arithmetic Operations on tensors
There are some common operations that you can perform on tensors like addition, subtraction, multiplication, matrix multiplication, and division.
How to load the tensor into CPU & GPU and vice-versa
There are two versions of the implementation of PyTorch tensor. One is for Cpu and the other is for GPU. Gpu is used when we have to perform massively parallel, fast computations.
To use the power of GPU, first, we have to every tensor to GPU and then it performs at a much faster rate.
If you don't have GPU you can use Google Colab or Kaggle kernels to get one. In our case, we are using Google Colab. And then go to the runtime menu and change the runtime type from None to GPU.
Let's see the code
As we can see that we using P100 GPU and the memory usage is 791MiB out of 16280MiB.
Wrap up the Session
So in this blog, we cover about what is PyTorch, why should we learn PyTorch, Pytorch pipeline, libraries in Pytorch, and introduction to tensor and its operation.
In the next blog post, we will cover how to implement machine learning models using PyTorch. So stay tuned.