Similarly to **NumPy**, PyTorch also offers several built-in functions to **create tensors directly**. These functions help in initializing data placeholders and generating structured or custom tensors.

## Tensor of Zeros and Ones

To create a tensor filled with **zeros**, use `torch.zeros()`. The arguments represent the **size of each dimension**, with the number of arguments corresponding to the **number of dimensions**:

import torch 
tensor = torch.zeros(4, 2)
print(tensor)

This is useful for **initializing biases** or placeholders where the initial values are set to zero.

Similarly, use `torch.ones()` to create a tensor filled with **ones**:

import torch
tensor = torch.ones(3, 3)
print(tensor)

This can be particularly useful for **initializing weights**, **bias terms**, or performing operations where a tensor of ones serves as a neutral element or a specific multiplier in mathematical computations.

## Arange and Linspace

Similarly to `numpy.arange()`, `torch.arange()` generates a sequence of values with a specified **step size**:

import torch
tensor = torch.arange(0, 10, step=2)
print(tensor)

We have successfully created a tensor with values from `0` to `10` **exclusive** with the step size equal to `2`.

To create **evenly spaced** values between a start and end point, use `torch.linspace()`:

import torch
tensor = torch.linspace(0, 1, steps=5)
print(tensor)

This generates a tensor with `5` **equally spaced** values between `0` and `1` **inclusive**.

## Tensor from Shape

You can create tensors with a specific shape by using the **"like" variants** of creation functions. These create tensors with the **same shape** as an existing tensor:

import torch
x = torch.tensor([[1, 2, 3], [4, 5, 6]])
zeros_tensor = torch.zeros_like(x)
ones_tensor = torch.ones_like(x)
print(f"Tensor of zeros: {zeros_tensor}")
print(f"Tensor of ones: {ones_tensor}")

Neural Networks with PyTorch

Tensor Creation Functions

Tensor of Zeros and Ones

Arange and Linspace

Tensor from Shape