Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Conformal Predictions

Exchangeability

In this section we discuss the difference between independence and exchangeability.

Definition. Independence.

Two events and from a sample space are said to be independent if .

If , it is equivalent to say , since:

Definition. Exchangeability

Two events and are said to be exchangeable if , which means there is indifference with respect to the order of events.

More generally, exchangeability of a sequence of events means that the joint distribution is unchanged when we permute the order of events:

The simplest way to understand is to use the example of drawing balls from an urn without replacement (example taken from Cordani 2006). Suppose we have an urn with 10 red balls and 5 white balls. Then the following tree shows the draw probabilities at each step:

graph TD;
    A((Start)) --> R1("R1 (10/15)");
    A --> W1("W1 (5/15)");

    R1 --> R2("R2 (9/14)");
    R1 --> W2("W2 (5/14)");

    W1 --> R2_2("R2 (10/14)");
    W1 --> W2_2("W2 (4/14)");

    R2 --> R3("R3 (8/13)");
    R2 --> W3("W3 (5/13)");

    W2 --> R3_2("R3 (9/13)");
    W2 --> W3_2("W3 (4/13)");

    R2_2 --> R3_3("R3 (9/13)");
    R2_2 --> W3_3("W3 (4/13)");

    W2_2 --> R3_4("R3 (10/13)");
    W2_2 --> W3_4("W3 (3/13)");

Suppose we

References