跳到主要内容

京都大学 情報学研究科 知能情報学専攻 2024年8月実施 専門科目 S-2

Author

itsuitsuki

Description (English)

We consider a problem of classifying a three-dimensional vector, where a value of each element is either 0 or 1, into either the class 1 or the class 2. Let be a vector. Assume that each element of of the class () independently follows a Bernoulli distribution, and let () be the probability of . Let be parameters of the class .

Let be a data set consisting of data. Let be the vector of the -th data, and be the class of . We assume that of the class is independently observed from the aforementioned distribution whose parameter is .

(1) Let be a prior probability for the class . We determine an estimated class of by comparing the posterior probabilities. Namely, we set if ; otherwise we set . Show a rule that assigns to an estimated class by using and .

(2) Let be a subset of the data set . By using , derive the maximum likelihood estimate of from the data set .

(3) Assume that a data set is given in Table 1. Compute the values of the maximum likelihood estimates from Table 1.

Table 1: A data set

11
21
32
41
52

(4) Let prior probabilities be and . Compute the estimated class of by substituting computed in (3) for of the rule shown in (1).

(5) Let a prior probability be . We classify by substituting computed in (3) for of the rule shown in (1). Explain the relation between and an estimated class .