Absolute value

Deep Breakdown

What it tells me (Situation Awareness): Absolute value tells you how far a number is from zero on the number line, completely ignoring direction. It strips the sign away. In DA/DS, you reach for this symbol every time you care about magnitude, not direction: measuring forecast errors, computing distances between data points, checking how far a metric drifted from a target, and calculating MAE (Mean Absolute Error). Whenever a negative and a positive error of the same size should be treated identically, absolute value is the tool.

Flowchart

Real-world use

Where it is used

When you need to measure the size of a discrepancy or distance without caring whether it is above or below a reference point. Sales forecasting, budget variance tracking, model error evaluation, signal deviation detection.

Problem it solves

A sales model predicts 100 units. The actual is 120. Error = 120 - 100 = +20. Another row: predicted 120, actual 100. Error = 100 - 120 = -20. Both errors have the same severity, but opposite signs. If you average +20 and -20 directly you get 0, which is wrong. You need a way to measure "how wrong" without errors canceling each other out.

How it solves the problem

Absolute value forces both errors to be positive: |+20| = 20, |-20| = 20. Now averaging gives 20, which correctly reflects the model's average miss magnitude.

Example in a business

Computing Mean Absolute Error (MAE) for a weekly sales forecast model. You take every (predicted - actual) difference, apply absolute value to each, then average them. The result is your average miss in units. Stakeholders understand "we are off by 20 units on average" far better than a technical loss function value.

The formula, piece by piece

Breaking the notation piece by piece:

• $|$ and $|$ (the two vertical bars): the "absolute value brackets." Whatever is inside gets its sign stripped.
• $x$: any real number. Can be positive, negative, or zero.
• $\geq$: greater than or equal to.

Full piecewise definition:

Two properties worth memorizing:

• $|x| \geq 0$ for all $x$ (result is always non-negative)
• $|x| = |-x|$ (symmetric: distance from -5 to zero equals distance from +5 to zero)

The formula in words

"If x is zero or positive, return x unchanged. If x is negative, flip its sign by multiplying by -1, which makes it positive. The result is always the distance from x to the origin." Manual check: $x = -7$. Since $-7 < 0$, apply $-x = -(-7) = 7$. Result: $|-7| = 7$.

Critical Concepts

1. Distance from zero: $|x|$ is the distance between $x$ and $0$ on the real number line. Distance is always non-negative.
2. Symmetry: $|-x| = |x|$. The value $-5$ and $+5$ are the same distance from $0$.
3. Triangle inequality: $|a + b| \leq |a| + |b|$. The distance of a sum is at most the sum of individual distances. This is used in proofs and in understanding why MAE behaves the way it does.
4. Composition with square root: $|x| = \sqrt{x^2}$. Alternative definition, useful in formulas and proofs.
5. Comparison to squaring: Both $|x|$ and $x^2$ remove sign, but $x^2$ also amplifies large errors.

This is why MAE and MSE behave differently.

Step-by-step calculation (tiny example)

Input: $x = -7$

Step 1: Identify the sign of x $x = -7$. Is $-7 \geq 0$? No. So we use the second branch: output is $-x$.

Step 2: Apply the rule $-x = -(-7) = 7$. Multiplying a negative by $-1$ gives a positive.

Step 3: Verify the result$|-7| = 7$.

Check: Is $7 \geq 0$? Yes.

Is $7$ the distance from $-7$ to $0$ on the number line? Count: $-7, -6, -5, -4, -3, -2, -1, 0$. That is 7 steps. Confirmed.

Bonus: $|+7| = 7$ because $7 \geq 0$, so output is $x$ itself.

Same result: both $-7$ and $+7$ are 7 units from zero.

Code examples

Beginner example (Python)

# abs() is Python's built-in absolute value function
x = -7
result = abs(x)
print(result)   # 7 # Works on floats too
print(abs(-3.14))   # 3.14 # Works on a list via list comprehension
errors = [20, -15, -8, 12]
absolute_errors = [abs(e) for e in errors]
print(absolute_errors)   # [20, 15, 8, 12]

Beginner example (R)

# abs() is R's built-in absolute value function
x <- -7
result <- abs(x)
print(result)  # 7 # Works on a vector automatically (vectorized)
errors <- c(20, -15, -8, 12)
absolute_errors <- abs(errors)
print(absolute_errors)  # 20 15 8 12

Professional example (Python)

import numpy as np
import pandas as pd   # Computing Mean Absolute Error (MAE) for a sales forecast
y_true = np.array([120, 90, 75, 200, 55])
y_pred = np.array([100, 95, 80, 185, 60])   # Step 1: compute raw errors (signed)
raw_errors = y_true - y_pred   # [ 20, -5, -5, 15, -5] # Step 2: apply absolute value to remove sign
absolute_errors = np.abs(raw_errors)   # [20, 5, 5, 15, 5] # Step 3: compute MAE
mae = np.mean(absolute_errors)   # 10.0
print(f"Raw errors: {raw_errors}")
print(f"Absolute errors: {absolute_errors}")
print(f"Mean Absolute Error (MAE): {mae:.2f} units")   # In a DataFrame context (more realistic)
df = pd.DataFrame({"actual": y_true, "predicted": y_pred})
df["abs_error"] = np.abs(df["actual"] - df["predicted"])
print(df)
print(f"\nMAE from DataFrame: {df['abs_error'].mean():.2f}")

Professional example (R)

# Computing Mean Absolute Error (MAE) for a sales forecast
y_true <- c(120, 90, 75, 200, 55)
y_pred <- c(100, 95, 80, 185, 60)  # Step 1: compute raw errors (signed)
raw_errors <- y_true - y_pred  # Step 2: apply absolute value to remove sign
absolute_errors <- abs(raw_errors)  # Step 3: compute MAE
mae <- mean(absolute_errors)
cat("Raw errors: ", raw_errors, "\n")
cat("Absolute errors: ", absolute_errors, "\n")
cat(sprintf("Mean Absolute Error (MAE): %.2f units\n", mae))  # In a data.frame context (more realistic)
df <- data.frame(actual = y_true, predicted = y_pred)
df$abs_error <- abs(df$actual - df$predicted)
print(df)
cat(sprintf("\nMAE from data.frame: %.2f\n", mean(df$abs_error)))

Source: 1300 p. 22

#foundations#maths-symbols#algebra