AI glossary: terms starting with T

A mean interval using sample spread and a t cutoff; its coverage depends on approximation conditions.

Probability beyond a chosen boundary; 2.275% above z = 2.

the line that touches the curve at a point with the curve's slope there

Example: y = 2x − 1 at x = 1

The recorded outcome a future model is intended to predict: next-month default, coded 0 or 1.

A feature carrying label information that is not available at prediction time: duration.

future completed trips under similar conditions; the cases a decision is meant to be used on.

a polynomial built from a function's derivatives at one point, matching it nearby

Example: 1 + x + 1.5x² for x² + eˣ at 0

Revealing, in a labelled teaching step, outcomes that the simulated process hides and a real lender could never recover.

Use of future information during training or selection for an earlier forecast.

A window where you type commands for the computer, one line at a time: Terminal on macOS, PowerShell on Windows.

A function that runs code and asserts what it should give.

Rows used once, to report the chosen model.

A column of free text.

the duration used to decide whether a trip is on time. In 00.2 it is 20 minutes, tested as duration <= 20.

Equal votes for two labels.

A column whose order matters and whose differences mean something, such as the model year.

How the number of operations grows with the size of the input.

The value on a date is associated with values on other dates.

Observations indexed in time order.

Fitting on earlier rows and validating on later ones.

One of the pieces tokenisation produces. Our tokeniser lowercases text, removes only ., ,, ! and ?, then splits on whitespace; other punctuation may remain.

Splitting text into pieces. Language models split into subword pieces; the video's tokeniser splits on whitespace.

A procedure that splits text into tokens.

How small the gradient must be before descent stops.

Python's report of a failed statement. Read it from the bottom: the last line names the error, and the arrow points at the line that failed.

Using more of one resource, memory or time, to use less of the other: a set spends O(m) memory to make each check O(1).

A difference between the data a model was trained on and the data it runs on.

Choosing a model's parameters to make the fewest errors on labelled examples.

The labelled examples used in training: 4,459 messages.

The rows the model is fitted on.

Units of log(1 + pounds), not pounds.

Exchange axes of a two-dimensional array

An axis whose displayed range omits lower values.

Whether a value counts as True or False when it is used as a condition. Falsy built-in values include False, None, zero of any numeric type, and empty strings or collections. Nonzero numbers and nonempty strings are truthy.

Q1 - 1.5 × IQR or Q3 + 1.5 × IQR.

An ordered sequence of values, written with commas, that can not be changed: (12.9716, 77.5946).

Determines what values an object can hold, and what you can do with it. type(x) shows it.

The intended type, written after a colon: items: int = 3. Python does not check it; editors and type checkers do.

A tool that checks every call and return against the type hints, without running the code: mypy.

A note in a function's header of the type a parameter should have, or the type it returns: review: str, -> list[str]. Python does not check it when the code runs.