Tensors

The problem

Hi, I'm your model. You're going to build me! I want to write stories. But I only understand numbers. Lots and lots of them.

A model is just numbers. Millions of them.

We need to do math on all of them at once.

numbers = [10, 20, 30]
print(numbers + 5)

A normal list can't do that. We need something better.

A tensor does math on every number at once. Let's meet one.

Meet the tensor

The row of numbers you typed, kept together as one thing, is a tensor.

One line of math changes every number inside it.

A tensor holds many numbers. How do you take out just one?

Pick one number

The small number above each item is its position. Counting starts at 0.

Try it on new numbers.

✏️ Practice 1.1

New numbers, same ideas.
Remember: numbers[0] is the first number. Positions start at 0.
numbers * 2 - 10 changes every number: first × 2, then − 10.

A model needs more than one line of numbers. Let's make a table.

Tables of numbers

A tensor can also be a table. Rows and columns. Like this one.

Try it on new numbers.

✏️ Practice 1.2

A table needs two positions: first the row, then the column. Both start at 0.
Remember: table[row, column]
A table holds rows × columns numbers.

Here's all of it in Python.

The code

Guess what each print shows. Then press ▶ Run.

import torch

# Make a tensor
numbers = torch.tensor([10, 20, 30])
print(numbers)

# Take one number out
print(numbers[0])

# Math on every number at once
print(numbers + 5)

print(numbers * 2)

# A table
table = torch.tensor([[1, 2, 3],
                      [4, 5, 6],
                      [7, 8, 9]])
print(table[1])

print(table[1, 2])

$ python lesson1.py
tensor([10, 20, 30])
tensor(10)
tensor([15, 25, 35])
tensor([20, 40, 60])
tensor([4, 5, 6])
tensor(6)

▶ Run it in Google Colab

Colab runs Python with PyTorch in your browser for free. You just need a Google account.

Quick check: did it all make sense?

Lesson 1 test

5 questions, one for each idea in this lesson. Answer them all right to pass. A wrong answer tells you why, and which page to look at again.

table = torch.tensor([[1, 2, 3],
                      [4, 5, 6],
                      [7, 8, 9]])

What your model still can't do.

What about text?

Your model can do math on lots of numbers at once.

But stories are made of letters, like "cat", and a tensor only holds numbers.

Now I can hold lots of numbers at once! But… how do I read a word like “cat”?

Mark lesson as done