AI glossary: terms starting with R

One minus the model's squared error over the mean's squared error: the fraction of the mean's squared error the model removes.

Create an exception and stop the function with it: raise ValueError(...).

Trying randomly sampled settings.

A random partition of the rows already observed.

A numeric value assigned to each possible outcome; here, a randomly chosen client's assigned cost.

A sequence of whole numbers from a start, up to but not including a stop, with an optional step: range(0, 20, 5) gives 0, 5, 10, 15. It works each number out when the loop asks for it.

the number of dimensions a matrix's output fills

Example: 1 for the singular example

Fitting on random minimal samples and keeping the one most rows agree with.

numerator divided by eligible denominator.

Δy/Δx, the change in the output per unit change in the input

Example: x² from 1 to 2: 3

change in output divided by change in input

Example: 2 everywhere for 2x + 1

An amount divided by the number of complete weeks it represents.

lists of d real numbers

Example: X ∈ ℝ³ˣ²

Binding a name to a different object. The old object is unchanged, and any other name still refers to it.

A function calling itself.

The error Python raises when the call stack goes past its limit, sys.getrecursionlimit(), typically 1000 in our Python session, but configurable.

The part of a recursive function that calls the function on a smaller input: return n * factorial(n - 1).

Combining values along an axis, such as sum or mean

A feature whose information another feature already carries: perimeter_mean beside radius_mean.

The category with no column of its own, which the other categories' coefficients are compared against.

Supervised learning where the answer is a quantity, so being close counts.

A test written from a bug, so that the bug can not come back unnoticed.

Adding a penalty on the weights to the objective.

Learning which actions to take from rewards; outside this track.

Difference divided by the starting value: 22.57% between the weeks starting 2011-10-31 and 2011-11-07.

the error divided by the size of the number

Example: at most 1.11 × 10⁻¹⁶ after rounding

A numerator and denominator plus the rule defining who is included.

Make a bug happen on purpose, on an input you can give again.

Each draw uses the same source values, so repeats are possible.

Interpret the same number of elements with a new shape; a copy is possible for non-contiguous inputs

actual value minus predicted value

Example: the middle third's mean, −1.34

Residuals plotted against fitted values.

Empirical spread of observed minus fitted durations: about 7.71 minutes here.

The value a call gives back. Reaching the end of the body normally without an explicit return gives back None.

A declared test that assigns each row to review or no review, such as PAY_0 > 0.

Linear regression with a penalty on the squared size of the weights.

A distribution with a longer tail toward larger values.

Square root of mean squared prediction error, in the original unit.

Fitting with a loss that a few bad rows can not dominate.

the true positive rate against the false positive rate as the threshold moves

Example: the tiny ward's seven points

Rank-based score comparing positive and negative labels: the share of positive–negative pairs where the positive scores higher, ties counting one half.

b^(1/n), the number whose nth power is b

Example: 2^½ = 1.4142

the difference between a number and the float stored for it

Example: 0.1 is stored 5.55 × 10⁻¹⁸ too high

one row, 1 × d

Example: u.reshape(1, -1)

Marks at observed values, one per recorded duration.

A program that follows a rule a person wrote, such as "spam if it contains free".