LiterallyTheOne
About Me
Posts
Tags
Tutorials
Introduction
Introduction
PyTorch
Deep-Learning
Python

Table of content

Introduction

What is PyTorch

PyTorch is an open-source framework for Machine Learning. It is developed by Facebook AI Research ( Meta ). This framework is mostly used to build and train Deep Learning models. Because of its flexibility and Pythonic inference, it has become so popular, especially among researchers. So, let’s write a hello world to understand PyTorch better. Then, in the future, we will complete this hello world example step by step.

Hello world

Problem definition

Imagine that we have 3 samples of data. Each sample has 8 features. We want to classify this data into 4 classes. So, the shape of our data would be [3, 8] and the shape of our result should be [3, 4]. Now, our plan is to just make a model that we can feed our data to. The simplest way to do that is to have a fully connected layer, with the input size of 8 and the output size of 4, like the image below:

pytorch_hello_world

Implementation

At first, we should import the necessary modules as below:

# -------------------[ Imports ]-------------------
import torch
from torch import nn

In the code above, we import torch and nn (neural network). Now, let’s create random data:

# -------------------[ Data ]-------------------
data = torch.rand((3, 8))  # (number_of_samples, features)

Now, we have random data that has 3 samples, and each sample has 8 features. After that, let’s create a simple linear model like the image that we provided in the implementation section.

# -------------------[ Model ]-------------------
model = nn.Linear(8, 4)  # (features, number_of_classes)

The code above creates a fully connected neural network layer that takes 8 features as its input and produces 4 classes as its output. For the next step, let’s feed that data to our model.

# -------------------[ Feed the data to the model ]-------------------
logits = model(data)
print(logits)

"""
--------
output: 

tensor([[ 5.3127e-01,  6.7324e-01, -1.7548e-01, -2.0279e-02],
       [ 5.3984e-01,  1.0462e+00, -1.0124e-01,  8.4969e-03],
       [ 4.6493e-01,  1.0864e+00, -3.6424e-01,  8.6406e-04]],
      grad_fn=<AddmmBackward0>)
"""

As you can see, we could simply call the model with our data. The output would be something like the probability of each class in each sample. As you might have noticed, there is a grad_fn in the output. When we call the model like this, PyTorch stores the gradient that we are going to use it in the future. So, if we want to get the class that we want, we should just report the index of the maximum probability.

result = logits.argmax(1)
print(result)

"""
--------
output: 

tensor([1, 1, 1])
"""

In the code above, I took the argmax of the dimension 1, which had the probabilities on it. Because we haven’t trained our model yet, the results are biased and all of them are predicting that this sample belongs to class 1.

In further we are going to explain each of them and try to complete our code step by step. You must have so many questions right now, but don’t worry, they will be answered soon.

Materials

© 2025 LiterallyTheOne