🧱 Understanding Tensors: The LEGO Blocks of Deep Learning

The Big Picture

Imagine you have a magical toy box filled with LEGO blocks. These blocks can be arranged in different ways—a single block, a row, a flat mat, or even a 3D cube.

Tensors are exactly like these LEGO blocks for computers!

They hold numbers in organized ways, and PyTorch uses them to build amazing things like image recognizers, chatbots, and self-driving cars.

🎯 What is a Tensor?

A tensor is just a container for numbers. Think of it like different ways to organize your toys:

graph TD
    A[🔵 Scalar: One Number] --> B[🔵🔵🔵 Vector: A Row]
    B --> C[📊 Matrix: A Grid]
    C --> D[📦 3D Tensor: Stacked Grids]

Simple Truth: A tensor is just numbers organized in boxes within boxes!

🛠️ Creating Tensors from Data

Let’s build some tensors! It’s as easy as putting toys in a box.

From a Single Number (Scalar)

import torch

# One lonely number
x = torch.tensor(42)
print(x)  # tensor(42)

Like holding one marble in your hand.

From a List (Vector)

# A row of numbers
ages = torch.tensor([5, 8, 12])
print(ages)  # tensor([5, 8, 12])

Like placing marbles in a line.

From Nested Lists (Matrix)

# A grid of numbers
grid = torch.tensor([
    [1, 2, 3],
    [4, 5, 6]
])
print(grid)

Output:

tensor([[1, 2, 3],
        [4, 5, 6]])

Like arranging marbles in a checkerboard pattern.

Quick Tensor Builders

PyTorch gives you shortcuts to make tensors fast:

# All zeros - like empty boxes
zeros = torch.zeros(3, 4)

# All ones - like boxes full of 1s
ones = torch.ones(2, 3)

# Random numbers - surprise boxes!
random = torch.rand(2, 2)

🎨 Tensor Data Types

Just like crayons come in different colors, numbers come in different types:

Setting the Type

# Decimal numbers (default)
prices = torch.tensor(
    [1.99, 2.50, 3.75]
)
print(prices.dtype)
# torch.float32

# Whole numbers only
counts = torch.tensor(
    [1, 2, 3],
    dtype=torch.int64
)
print(counts.dtype)
# torch.int64

Changing Types

# Convert decimals to whole numbers
x = torch.tensor([1.9, 2.7, 3.1])
y = x.to(torch.int64)
print(y)  # tensor([1, 2, 3])

The decimals got chopped off—like rounding down!

📐 Tensor Shapes and Dimensions

Shape tells you how your LEGO blocks are arranged.

What is Shape?

# A row of 4 numbers
row = torch.tensor([1, 2, 3, 4])
print(row.shape)  # torch.Size([4])

# A 2x3 grid
grid = torch.tensor([
    [1, 2, 3],
    [4, 5, 6]
])
print(grid.shape)  # torch.Size([2, 3])

Reading shapes:

• [4] → 4 items in a row
• [2, 3] → 2 rows, 3 columns

Dimensions (How Deep is the Box?)

# Scalar = 0 dimensions
scalar = torch.tensor(5)
print(scalar.ndim)  # 0

# Vector = 1 dimension
vector = torch.tensor([1, 2, 3])
print(vector.ndim)  # 1

# Matrix = 2 dimensions
matrix = torch.tensor([[1, 2], [3, 4]])
print(matrix.ndim)  # 2

graph TD
    S[0D: Scalar] -->|Add dimension| V[1D: Vector]
    V -->|Add dimension| M[2D: Matrix]
    M -->|Add dimension| T[3D: Tensor]

Real-World Shapes

Checking Size

img = torch.rand(3, 224, 224)

print(img.shape)     # torch.Size([3, 224, 224])
print(img.size())    # torch.Size([3, 224, 224])
print(img.ndim)      # 3
print(img.numel())   # 150528 (total numbers)

🎉 Quick Summary

🚀 You’re Ready!

You just learned the foundation of all deep learning!

Every neural network, every AI model—they all start with tensors. You now understand:

✅ What tensors are (organized number containers) ✅ How to create them (from lists, zeros, ones, random) ✅ Data types (float, int, bool) ✅ Shapes and dimensions (the arrangement of numbers)

Next step: Start playing with tensors in PyTorch. Try creating your own!

import torch

# Your turn! Make a tensor
my_tensor = torch.tensor([
    [1, 2],
    [3, 4]
])
print("Shape:", my_tensor.shape)
print("Type:", my_tensor.dtype)

You’ve got this! 🎮