AI glossary: terms starting with M

the gap between 1 and the next float, 2.22 × 10⁻¹⁶ in float64

Example: np.finfo(float).eps

Choosing a program's parameters from examples, then using it on new examples.

Mean absolute error, average absolute prediction miss in the original unit.

Distance from the fitted centre after accounting for covariance.

The sum of the absolute differences.

Missingness independent of the missing value after conditioning on observed variables.

One variable's distribution after counting across the other variable.

a table of numbers, written as a capital

Example: X, shape (3, 2)

Row-column combination written @

each row of the matrix dotted with the vector, one output per row

Example: Xw = (5, −3, 10)

the derivative changes from positive to negative

Example: x⁴ − 4x² + x at 0.126

Parameter value where the observed-data likelihood is largest.

Missingness probability independent of observed and unobserved values.

The sum of observed values divided by their count.

total absolute error divided by evaluated trips; the average size of the error per trip.

The average of the squared residuals.

One average coordinate for each feature.

the component-by-component mean of a set of points

Example: (4, 3)

the middle value; it minimises the sum of absolute errors

Example: 13.2 for the six hero prices

Storing the training examples' answers instead of learning a rule, as model M does.

A function that belongs to an object, called with a dot: cart.append(...), review.lower().

A small group of training rows used for one update.

the derivative changes from negative to positive

Example: f′(1) = −1, f′(2) = 1

(Σ | dᵢ | ᵖ)^(1/p); Manhattan at p = 1, Euclidean at p = 2; a distance only for p ≥ 1.

Cases whose result cannot be observed under the stated process.

A cell with no recorded value, read by pandas as NaN.

Absent recorded values in a stated field and set of rows: 135,080 lines have no CustomerID.

A possible reason a value is missing; the rows alone cannot say which one applies.

Missingness still depends on unavailable values after conditioning on observed information.

The histogram interval with the highest count under declared edges.

The most frequent observed value under a stated precision (here, to the penny).

The form of the rule, with its parameters left open: "spam when the digit count is at least t".

A Python file that other code imports.

%: the remainder after floor division. 10 % 3 is 1, and -7 % 2 is 1.

Measuring a deployed model's score on new labelled data.

Difference caused by using finitely many random draws under a fixed model: 30.44% simulated against 29.94% exact.

Repeated random draws and outcome calculations under stated assumptions.

The mean of each value and its neighbors; a centered three-week one uses the week and the weeks either side.

Classification with more than two classes, such as the ten digits.

Collinearity among more than two features.

A linear model with more than one feature.

A normal probability model for a vector of jointly varying numeric features.

A type whose objects can be changed after they are created; a list is mutable, and so are sets and dictionaries.

Changing an object in place, which every alias sees: append, remove, sort, assigning to an index.

Two events with no shared outcomes, such as spam and real.