AI glossary: terms starting with C

Running a function's body, by writing its name and round brackets: with_tax(12.50).

The frames of every call that has started and not yet returned, the newest on top.

the loss of correct digits when two nearly equal numbers are subtracted

Example: (1 + h)² − 1 at h = 10⁻¹²

An invoice ID with a C prefix; refund status is not established.

A recorded input proposed for later model evaluation.

How closely a model can follow its training rows.

A column whose values name groups, even when they are stored as numbers.

A named group represented by a recorded value.

A claim about what would change if something were changed; the report can not make one from this file.

The outcome change attributable to an intervention under a valid comparison.

Cumulative probability at or below a value.

The largest vertical distance between the two classes' empirical CDFs: 0.812 for concave_points_worst. A threshold placed there separates the classes as well as the feature can.

Under suitable conditions, standardized sample means approach a normal distribution as sample size increases; it gives no fixed sample size.

each column minus its mean, so every column mean is zero

Example: [[0, −1], [−2, 1], [2, 0]]

Subtracting each feature's mean.

the derivative of f(g(x)) is f′(g(x)) · g′(x)

Example: (3 − 2x)² → 2(3 − 2x)(−2)

Calling a method on the value the previous method returned: review.strip().lower().

The largest absolute difference.

a straight line between two points of a curve

Example: x² from −2 to 1

A later test period separated from earlier training dates.

One for a recorded churn label, zero otherwise.

A definition of a new kind of object: class Review:. Calling it, Review(...), creates an instance.

The distribution of a feature within one class: concave_points_worst among the malignant tumors.

Supervised learning where the answer is one of a set of categories.

A formula that gives the minimum directly, without iterating.

A function that retains access to bindings in an enclosing function scope after the outer call returns.

Finding groups of similar examples (Module 12).

Score change when x rises by one.

Every weight plotted against the penalty strength.

Jupyter notebooks in the browser, run on Google's computers. Notebooks are saved to Google Drive, and nothing is installed locally.

A pattern in the data made by how it was collected, not by what it records.

Features that are close to linear functions of each other.

Two keys in the same bucket, such as 46 and 22 in a table of eight buckets.

the average of one column, one number per feature

Example: (2, 2)

the span of a matrix's columns, every possible Ax |

one column, d × 1

Example: u.reshape(-1, 1)

where e₁ and e₂ land

Example: [[2, 1], [1, 3]]: (2, 1) and (1, 3)

Text after #, skipped when the code runs. A comment over several lines starts each line with its own #.

An operator that gives True or False: ==, !=, <, >, <=, >=.

Every outcome in the sample space outside an event.

either the constraint is tight or its multiplier is zero

Example: x + y ≤ 4: multiplier 0

A number with real and imaginary parts, such as 1 + 2j. The j marks the imaginary part; type(1 + 2j) is <class 'complex'>.

one entry of a vector

Example: 3 in (3, 2)

one transformation applied after another

Example: shear, then rotate

a calculation drawn as nodes, one operation each, joined by the values they pass

Example: u = 3a, v = u + b, f = v²

An expression whose truth value an if or a while tests, such as total >= 50 or name.

how much a matrix can magnify a small relative change; 1 is perfect, and large means nearly singular

Example: 25,008 for [[1, 2], [2, 4.001]]

a if condition else b, an expression that gives one of two values: 0.0 if total >= 50 else 4.99.

Within each class, one flag result does not alter another flag's probability. An assumption the construction does not establish.

A fraction calculated inside a restricted group: P(A given B) counts the outcomes in both and divides by all of B. P(spam given flagged) = 9 / 108.

Bounds produced by a sampling procedure.

The intended long-run coverage of a procedure under its assumptions.

The rows within the threshold of the chosen fit.

Similar conditional residual variability across predictor values.

O(1): the same number of operations however large the input, such as values[0] or len(values).

a condition the answer must satisfy, such as g(x, y) = c

Example: x + y = 1

Counts for each combination of two recorded categories.

Ends this iteration; the loop goes on with the next item, or checks its condition again.

Under a model, any value in an interval is possible, such as a review time from 0 to 10 minutes.

a curve joining points where the function has the same value

Example: the ellipse x² + 3y² = 7

Lines of equal loss over the plane of two parameters.

The number the remaining error is multiplied by at each update.

A published population count used here for comparison.

The parameters settling at the minimum.

every chord lies on or above the curve

Example: x², log(1 + eˣ)

How much the fitted values change when one row is left out.

Minimizing over one weight at a time, with the others held fixed.

An array with independent data

covariance divided by both standard deviations, between −1 and 1

Example: 0.8

the covariance matrix of standardised features; ones on the diagonal, correlations off it

Example: [[1, 0.9588], [0.9588, 1]]

the horizontal coordinate of the point at angle θ on the unit circle

Example: cos 60° = 0.5

One minus cosine similarity.

the cosine of the angle between two vectors

Example: 0.5714 for emails 12 and 13

An assumed cost for each action × outcome combination: $1,000 per missed default, $100 per reviewed nondefault, $0 otherwise.

one component per vocabulary word

Example: email 12, (2, 0, 1, 1, 1, 0, 0, 0, 0, 0)

A dictionary that counts; a missing key counts zero. Counter(votes).most_common(1) gives [('tea', 3)]. From collections.

the mean of the products of two centred features; positive when they rise together

Example: 4

axes along the covariance matrix's eigenvectors, semi-axes the square roots of its eigenvalues

Example: 3 and 1

(1/n) XᵀX for centred X; one row and column per feature

Example: [[5, 4], [4, 5]]

Fraction of repeated intervals containing a fixed target.

A rate range containing a stated share of posterior probability under the stated prior and likelihood.

Estimating a model's score by fitting and validating on several splits of the training rows.

A table stored as text, one row per line, values separated by commas. A value with a comma in it is written in quotes.

Probability at or below a threshold.

The count at or below a value divided by all selected observations.

how sharply a curve bends; its sign says up or down

Example: +6 at 1, −6 at −1

How fast the slope of the loss changes as one moves that way.