Some amounts go below zero: a temperature on a cold night, a bank balance in debt, a prediction that's too low. Negative numbers let one number say both how far and which way. Every model's error is like that: too high or too low, by some amount. This lesson extends the number line through zero and shows the few rules for working with what's on the other side.
What you'll be able to do
Place negative numbers on the number line and compare them
Add, subtract, multiply and divide with negative numbers using the sign rules
Find the distance between two numbers with absolute value, and use abs() and comparisons in Python
Read a model's error as a direction and a size
Quick check
Have a guess before you start. It's −3 degrees at dawn and 5 degrees at noon. How many degrees did it warm up?
2
5
−8
8
Show the answer
8
From −3 up to 0 is 3 degrees, and from 0 up to 5 is 5 more: 8 in all. That's 5 − (−3), and subtracting a negative is adding.
The math: the number line through zero
Purpose. Counting starts at zero, but amounts don't stop there. A negative number is below zero: −3 means 3 steps the other way.
Picture the number line extended to the left of zero:
−5 −4 −3 −2 −1 0 1 2 3 4 5
Further right means bigger. So −1 is bigger than −3, even though 3 is the bigger digit: −1 is closer to warm.
Adding moves you right; subtracting moves you left. Adding a negative is moving left: 6 + (−4) is the same as 6 − 4 = 2. And subtracting a negative moves you right: 5 − (−3) = 5 + 3 = 8. Taking away a debt leaves you better off.
The absolute value, written |−3|, is the distance from zero, with no direction: |−3| = 3 and |3| = 3.
The flow below works 6 + (−4) − (−3), and its picture shows each step as a jump along the line. A minus sign in front of a number flips it to the other side of 0. Adding −4 jumps 4 to the left; subtracting −3 jumps 3 to the right.
Try it now
The minus signs that belong to a number apply first; then it's left to right, one jump at a time. Set the expression to 10 - -4 and press Use these numbers: which way does the last jump go?
Symbols and formula. For multiplying and dividing, there are two sign rules:
same signs→+different signs→−
In plain words: two negatives multiplied make a positive; a negative times a positive makes a negative. Division follows the same rule.
So −3 × −4 = 12, −3 × 4 = −12 and −24 ÷ 6 = −4.
Distance between two numbers is the absolute value of their difference:
distance(a,b)=∣a−b∣
The distance from −3 to 5 is |−3 − 5| = |−8| = 8. Order doesn't matter: |5 − (−3)| = |8| = 8 too.
Smaller minus bigger. Taking a bigger number from a smaller one crosses zero. For 3 − 8, move 3 left to reach 0, then 5 more: −5. A quick way: subtract the smaller from the bigger (8 − 3 = 5), then give the answer the sign of the bigger one's direction. Since you took away more than you had, it's negative.
You meet this all the time. A path climbs from 120 meters below a pass to 45 meters above it: that's 45 − (−120) = 165 meters of climbing. A store is 30 bars short and gets 50: −30 + 50 = 20 bars to spare.
Quick check
Which is the biggest: −7, −2, −10 or −5?
−10
−5
−2
−7
Show the answer
−2
On the number line −2 sits furthest right, closest to zero, so it's the biggest.
Quick check
What is −6 × −3?
−18
9
−9
18
Show the answer
18
Same signs make a positive: 6 × 3 = 18.
Where the sign rules come from. Look at the pattern as you count down: 3 × 2 = 6, 2 × 2 = 4, 1 × 2 = 2, 0 × 2 = 0. Each step takes away 2. Keep going: −1 × 2 = −2, −2 × 2 = −4. A negative times a positive is negative.
Now count down with −2 instead: 2 × −2 = −4, 1 × −2 = −2, 0 × −2 = 0. Each step adds 2. Keep going: −1 × −2 = 2, −2 × −2 = 4. Two negatives make a positive because the pattern has to continue.
In your own words
In your own words: why is subtracting a negative the same as adding?
Write a sentence or two, then compare it with a good answer in the lesson.
In Python: signs, distances and true-or-false
Purpose. Python handles negatives with the same rules, and has a function for distance from zero.
It runs right here in your browser, with nothing to install.
abs( ) gives the absolute value: the number without its sign.
You can also ask Python a yes-or-no question. A comparison gives True or False:
print(-1 > -3) # is −1 bigger than −3?print(-7 < -10) # is −7 smaller than −10?print(abs(-3 - 5) == 8)
The comparison signs are > and < (bigger, smaller), >= and <= (or equal), == for "is equal to" (two equals signs, since one = stores a value), and != for "is not equal to".
Two jobs for one sign. The minus sign does two different jobs. Between two numbers, as in 9 - 4, it subtracts. In front of a single number, as in -4, it makes that number negative. So 4 - -6 reads as "4 minus negative 6", which is 10. Python can tell the two jobs apart from where the sign sits, but people can't always: write brackets, 4 - (-6), when you want the line to be easy to read. Python gives the same answer either way.
Worked example
Worked example: distances on the number line
A trail sign marks 0 at the parking lot. Two hikers stand at −350 meters (behind the sign) and 1,200 meters (ahead). How far apart are they, and who is farther from the sign?
ann = -350
ben = 1200
gap = abs(ann - ben)
print(f"gap: {gap} meters")
print(f"Ann from the sign: {abs(ann)}, Ben from the sign: {abs(ben)}")
print(abs(ben) > abs(ann))
In the lesson you can change this and run it.
The gap is |−350 − 1200| = |−1550| = 1,550 meters: the distance doesn't care who's ahead. Each hiker's distance from the sign is their absolute value: Ann is 350 meters away, Ben 1,200. The last line asks whether Ben is farther, and Python answers True. Notice that comparing abs(ben) with abs(ann) compares distances, while comparing ben with ann would only say who is farther ahead.
Here those two questions get different answers. Ann is now 1,500 meters behind the sign and Ben 900 ahead. Step through one line at a time and watch each name get its value: diff keeps its minus sign, gap drops it, and the two comparisons come out as True or False.
Quick check
What does this print?
print(-7 // 2)
print(abs(-7) // 2)
-4
4
-4
3
-3.5
3
-3
3
Show the answer
-4
3
// always rounds down, toward the left of the number line: −3.5 rounds down to −4. abs(−7) is 7, and 7 // 2 = 3.
Quick check
Which line checks whether two hikers at a and b are less than 100 meters apart?
print(abs(a - b) < 100)
print(abs(a - b) = 100)
print(abs(a) - abs(b) < 100)
print(a - b < 100)
Show the answer
print(abs(a - b) < 100)
The distance is abs(a − b), and < 100 asks whether it's under 100. It works whichever hiker is ahead.
Drill
How far, and which way
A thermometer read −12 at night and 7 at noon. Set change to noon minus night, distance to how far apart the readings are, and warmer to whether noon was warmer than night.
night = -12
noon = 7
change = night - noon
distance = noon - night
warmer = noon < night
print(change, distance, warmer)
In the lesson, your code is checked as you write it.
In your own words
What does abs( ) throw away, and what does it keep?
Write a sentence or two, then compare it with a good answer in the lesson.
In AI and ML: an error has a direction and a size
Purpose. When a model predicts a number, it's usually a little off. Which way it's off, and by how much, are two different facts, and negative numbers carry both.
On the model map below, this lesson lights two parts at the end of the forward row: the targets (the right answers) and the loss (one score for how wrong the predictions are).
For the trail club's tent rentals:
error=prediction−targetabsolute error=∣error∣
In plain words: the error is positive when the model guessed too high, negative when too low. The absolute error is just how far off, either way.
prediction = 38
target = 45
error = prediction - target
print(f"error {error}: too low by {abs(error)} tents")
Negatives show up inside models too. A weight multiplies an input. A positive weight pushes the prediction up as the input grows; a negative weight pushes it down. If rentals drop as rain grows, the weight on "millimeters of rain" will come out negative.
50 − 44 = 6: positive, so the model guessed too high. Its absolute error is also 6, and 6 > 0 is True.
Now put both of those predictions on the number line: 38 against 45 from the cell, and 50 against 44 from the check. An error is a jump from the target to the prediction. Its direction is the sign, and its length is the absolute error. Press Next to step through.
Why the absolute error matters. Suppose a model is 10 tents too high one day and 10 too low the next. Its two errors are +10 and −10, and they add up to zero, as if it were perfect:
e1 = 10
e2 = -10print(e1 + e2, abs(e1) + abs(e2))
The plain sum hides both mistakes, because the signs cancel. The sum of absolute errors, 20, shows how wrong the model really was. That's why a model is judged by the size of its errors, with the sign dropped, while the sign is kept to spot a habit, like always guessing low.
In the real tools. In PyTorch, the library most models are built with, a tensor holds many numbers together, so one line finds the error for every day at once. Here are the two predictions you just drew, and a third day:
Quick check
A model's weight on "millimeters of rain" is −3. What happens to its tent prediction as rain grows?
It goes up by 3 tents per millimeter
It goes down by 3 tents per millimeter
It doesn't change
It becomes negative at once
Show the answer
It goes down by 3 tents per millimeter
The weight multiplies the input: each extra millimeter adds −3 × 1 = −3 to the prediction.
Drill
Errors for three predictions
Three days' tent predictions and true counts are below. Set e1, e2 and e3 to each error (prediction minus target), a1, a2 and a3 to their absolute values, and beats_1 and beats_3 to whether day 2's absolute error is bigger than day 1's and than day 3's.
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