A letter in math stands for a number you don't know yet, or for any number at all. With letters, a rule is said once and works for every number: area = length × width covers every rectangle. This lesson shows how to read and tidy expressions, and how to work one out by putting numbers in. Python does the same with a function. And a machine learning model is, at heart, an expression whose letters get tuned.
What you'll be able to do
Read an expression: its variables, terms and coefficients, and how 3x means 3 × x
Combine like terms, and put numbers in for letters to work an expression out
Write a formula as a Python function with def and return, and call it with inputs
Read a model's formula, ŷ = w·x + b, and write it as a predict function
Quick check
Have a guess before you start. If x is 4, what is 3x?
34
7
12
3.4
Show the answer
12
A number written right before a letter multiplies it: 3x means 3 × x, so 3 × 4 = 12.
The math: letters that stand for numbers
Purpose. Say a rule once, for every number it could apply to.
A variable is a letter that stands for a number: x, n, or any other. An expression is numbers, letters and operations with no equals sign, like 2x + 7. It isn't a question; it's a recipe for a number, once you know x.
Two habits of writing:
A number right before a letter multiplies it: 3x means 3 × x. The number is the coefficient.
x² means x × x, just as 4² meant 4 × 4.
The parts joined by + or − are terms. 2x² + 3x − 1 has three terms: 2x², 3x and −1. The last one, with no letter, is a constant: it never changes.
Quick check
From an earlier lesson: what is 2 × 4²?
64
32
10
16
Show the answer
32
Powers come first: 4² = 16. Then 2 × 16 = 32. The same order applies when a letter takes a number's place.
Like terms combine. Think of each term as a pile of tiles. 3x is three x-tiles, and 17x is seventeen more, so 3x + 17x is twenty x-tiles: 20x. Only the same kind of tile can be counted together. An x²-tile is a different object from an x-tile, just as a square is different from a strip.
3x + 5x = 8x (same tile: add the coefficients)
4x² + x² = 5x² (x² on its own means 1x²)
3x + 2 stays as it is (an x-tile and two plain units)
2x + 3y stays as it is (two different letters)
Quick check
Simplify 4a + 2b + 3a.
9ab
7a + 2
9a
7a + 2b
Show the answer
7a + 2b
4a and 3a are like terms: 7a. The 2b is a different kind, so it stays: 7a + 2b.
Substitution: putting numbers in. To work an expression out, put the number in place of each letter, then do the arithmetic in the usual order. For 2x² + 3x − 1 at x = 4, every x becomes 4.
The picture draws the expression as a tree, as in lesson 3: each operation sits in a box above the two things it works on, and the box at the top happens last. Watch the two x's at the bottom turn into 4s, then the tree fold up from the bottom, one operation a step, until a single number is left.
Try it now
Then set x to -1 and press Use these numbers: the same recipe gives a different number. That's what a variable is for.
Symbols and formula. Combining like terms and substituting, in symbols:
ax+bx=(a+b)x2x2+3x−1x=4=2(16)+12−1=43
In plain words: same-letter terms add their numbers. To work out an expression, put the number in for the letter and do the arithmetic.
Words into symbols. Four words come up again and again: the sum is what you get by adding, the difference by subtracting, the product by multiplying and the quotient by dividing. So "5 more than twice a number" is 2n + 5, and "the quotient of a number and 4" is n/4.
Quick check
Which expression is "the product of 3 and a number, minus 4"?
3n − 4
3(n − 4)
4 − 3n
3 + n − 4
Show the answer
3n − 4
The product of 3 and n is 3n; minus 4 takes 4 away from it: 3n − 4.
In your own words
In your own words: why can 3x and 5x be added into one term, but 3x and 5x² can't?
Write a sentence or two, then compare it with a good answer in the lesson.
In Python: a formula with a name
Purpose. Python's version of an expression with letters is a function: a formula with a name and inputs. You write it once with def, and return hands its answer back:
It runs right here in your browser, with nothing to install.
length and width are the function's letters, its parameters. Calling area(3, 4) substitutes 3 and 4 for them, just as you did by hand. The indented line is the body: the formula itself.
Now step through the same function, called twice. At each call, watch Python jump into the body: the parameters take that call's numbers, return hands the answer back, and the parameters vanish again when the call ends. Press Next until small has its value, then keep going into the second call.
A math variable and a Python variable. They're close but not the same. In math, x stands for one number throughout a problem, so x = x + 1 is impossible. In Python, = means "store", so it's an instruction: work out the right side, then store it in the name.
x = 5
x = x + 1print(x)
A named mistake: print instead of return. A function that prints its answer shows it, but hands nothing back, so the result can't be used. Python gives back None, meaning nothing:
defdouble_print(n):
print(n * 2)
r = double_print(5)
print(r)
Worked example
Worked example: two formulas from earlier lessons
Write lesson 8's percentage change and this lesson's 2x² + 3x − 1 as functions, then call them.
defpercent_change(old, new):
return (new - old) / old * 100deff(x):
return2 * x ** 2 + 3 * x - 1print(percent_change(40, 50))
print(f(4), f(-1), f(0))
In the lesson you can change this and run it.
Each function is the formula written once, with its letters as parameters. percent_change(40, 50) puts 40 in for old and 50 for new, giving 25.0. f(4) is the 43 you worked out by hand. f(0) is −1: with x = 0, only the constant is left. Writing f(x) for "the formula at x" is how math names functions too, as you'll see in lesson 19.
Quick check
What does this print?
defdouble(n):
return n * 2print(double(3) + double(4))
Which function hands back x squared plus 1, so it can be used later?
def f(x): return x ** 2 + 1
f(x) = x ** 2 + 1
def f(x): print(x ** 2 + 1)
def f(x) = x ** 2 + 1
Show the answer
def f(x): return x ** 2 + 1
def names the function, and return hands the value back to whoever called it.
Why give a formula a name? Once a formula has a name, it can be used anywhere, with any inputs, and changed in one place. If a rule changes, say a tax goes from 8% to 9%, you edit one line inside the function, and every call uses the new rule. Names also make code read like math: area(3, 4) says what it's for, while 3 * 4 on its own could be anything. And a function can call other functions inside it, so big rules are built from small named ones. That's how large programs, and models, are put together.
Drill
Two formulas as functions
Fix the two functions: area(length, width) should return the area of a rectangle, and percent_change(old, new) the percentage change from old to new. Each has one slip.
defarea(length, width):
print(length * width)
defpercent_change(old, new):
return (new - old) / new * 100
a = area(7, 3)
p = percent_change(80, 100)
print(a, p)
In the lesson, your code is checked as you write it.
In your own words
Why does x = x + 1 make sense in Python but not as a math equation?
Write a sentence or two, then compare it with a good answer in the lesson.
In AI and ML: a model is an expression
Purpose. The simplest machine learning model is an expression with adjustable letters, and making a prediction with it is substitution.
The map below is the same in every lesson of this course: a model learning, with the part this lesson's math lives in lit.
A shop that rents tents might predict rentals from the day's temperature:
y^=w⋅x+b
In plain words: x is the input (the temperature), w is the weight (how much each degree adds), b is the bias (the starting amount), and ŷ, read "y hat", is the prediction. The hat marks a guess, not a fact.
Training a model means choosing the numbers for w and b. Once they're chosen, predicting is substitution. Here is the tent model from lesson 3, w = 2 and b = 3, on three days from the shop's log. Each row prints the day's temperature, the prediction, and the tents the shop actually rented:
defpredict(x, w, b):
return w * x + b
print(10, predict(10, 2, 3), 22)
print(20, predict(20, 2, 3), 41)
print(30, predict(30, 2, 3), 59)
Every prediction is the same expression with a new x put in. Draw the predictions for every temperature and they make a straight line, which lesson 21 studies. The picture shows the three days as dots and the model as that line; press Next to close in on the 20-degree day.
Quick check
What does this print?
defpredict(x, w, b):
return w * x + b
print(predict(10, 2, 5))
70
17
30
25
Show the answer
25
x = 10, w = 2 and b = 5: 2 × 10 + 5 = 25.
Greek letters are just names. Papers use Greek letters for some quantities, and they're ordinary names in disguise: η (eta) is usually the learning rate, θ (theta) all of a model's weights together, and σ (sigma) a spread or a squashing function. In code they're written out:
eta = 0.01
theta = 2.5print(theta - eta * 4)
That one line is a tiny version of a training step, which you'll meet properly in Module 5: take the weight, and move it by the learning rate times some amount.
Quick check
In ŷ = w·x + b, which letter is the input, the thing the model is given?
ŷ
x
w
b
Show the answer
x
x is the input, like the temperature. w and b are set by training, and ŷ is what comes out.
In the real tools. In PyTorch, the expression w·x + b with one input is a layer: nn.Linear(1, 1), the name you found on the map. It keeps its w in .weight and its b in .bias. Here is the tent shop's model, its weight set to 2 and its bias to 3, asked about a 20-degree day:
Drill
Predict tent rentals
A tent shop's model is rentals = 3 × temperature − 20. Write predict(x, w, b), then set cool, mild and hot to its predictions for 15, 22 and 30 degrees.
defpredict(x, w, b):
return w + x * b
cool = predict(15, 3, -20)
mild = predict(22, 3, -20)
hot = 0print(cool, mild, hot)
In the lesson, your code is checked as you write it.
In your own words
What's the difference between the x and the w in ŷ = w·x + b?
Write a sentence or two, then compare it with a good answer in the lesson.
Goes deeper in
Functions: functions in more Python, with defaults and return values.
31Formulas with fractions: clear them firstFreeClearing denominators, cross-multiplying, a letter in the bottom, and reciprocal formulas like 1/R = 1/R₁ + 1/R₂.35 min
37Inverse functions, lines and scalingFreeUndoing a function by making x the subject, domains that allow it, line forms, and unscaling predictions.40 min
38Sequences and sums: find n, a, d and rFreeEvery letter of the arithmetic and geometric formulas as the subject, and an average's window from its decay.40 min
40Review: rearrange anything (algebra)FreeBroken rearrangements to fix, mixed practice from the whole algebra block, a strategy chart, and its formula sheet.35 min
41Coordinates and distanceFreePoints as (x, y), Pythagoras, distance in any number of coordinates, and nearest neighbors.42 min
42Angles, degrees and radiansFreeMeasuring a turn, converting units, arcs, and the angle between two arrows.36 min
43Sine and cosineFreeAcross and up: right triangles, the unit circle, and directions and rotations.42 min
44Waves: height, speed and shiftFreeAmplitude, period and phase, sampling a wave, and position codes in transformers.37 min
68Vector equations: rearranging with arrowsFreeSolving for a vector or a scale, recovering a gradient from two weight snapshots, and an angle from a dot product.40 min
87Counting: choices, orders and groupsFreeThe counting principle, factorials, ordered picks and groups, and why attention's cost grows with the square of the length.43 min
88Probability: from counts to chancesFreeEqually likely outcomes, the complement, distributions, seeded simulation, and a next-word model.43 min
89Combining events: and, or, independentFreeThe or rule, independence, probability trees, drawing with and without replacement, and chains of steps.42 min
100Maximum likelihoodFreeLikelihoods as products, log-likelihoods as sums, grid search, underflow, and the negative log-likelihood loss.45 min
101Linear regression: the best lineFreeResiduals, least squares by formula, three roads to one line, the matrix form, and train versus test error.45 min
109One neuron, start to finishFreeWeighted sum, activation, loss, slopes and update, by hand and in code, and what one straight line can and can't split.45 min
111Training a small networkFreeThe backward pass in matrix form, gradient checks, a full training loop, and reading loss curves.45 min
112The same network in PyTorchFreenn.Sequential with the NumPy weights loaded, the same loss and slopes, then SGD and mini-batches.45 min
113Attention as matrix mathFreeScaled dot-product scores, the causal mask, a row-by-row softmax, and blending value vectors.45 min
114Residual connections and layer normFreeAdding the input back and why its slope keeps learning alive, layer norm by hand, and add and norm.40 min
115A transformer block, start to finishFreeA sentence through embeddings, attention, add and norm, a small network and next-word scores, every shape named.45 min
117Final test: transpose any formulaFreeSlips to fix from every topic, mixed practice of every kind of rearranging, the strategy chart and the full sheet.45 min
Learn the math behind AI from arithmetic up: numbers, algebra, vectors and matrices, calculus, probability and statistics, each shown as numbers moving and checked in Python, ending with a network you build and train from the formulas.
See the course →
Never trained an AI? Teach a small one to write, right in your browser. It takes about 20 seconds and needs no account. Teach it to write