1
$\begingroup$

This is a question from a data mining exam:

You are working with a dataset that contains descriptions of toxic and non-toxic substances.

The dataset, which consists of 1000 examples from each of the two classes, is described in terms of a class label and a number of attributes. The dataset is sorted so that the 1000 toxic examples come first, followed by the 1000 non-toxic examples. Someone tells you that they have confirmed that, for this data set, the conditional probability that is gained from knowledge about a specific attribute X is not different from the prior class probability. Assume that they are correct.

The question is to state whether the following statements are correct or incorrect:

g) For each example, the value of attribute X is sampled from {“true”, “false”} with a uniform distribution. h) For each example, the value of attribute X is sampled from {“true”, “false”} such that the probability of selecting “true” is 0.75, while the probability of selecting “false” is 0.25.

both g and h were solved as correct independently but I find this hard to understand. Can anyone explain?

$\endgroup$
  • 1
    $\begingroup$ The information seems pretty garbled to me. E.g., "The conditional probability" that what? And toxic or nontoxic substances are going to have their own attributes: the attributes would not be assigned by some sampling process. $\endgroup$ – rolando2 Sep 29 '12 at 20:53
1
$\begingroup$

Since the sizes of classes are equal, prior class probability is {0.5,0.5}. If the conditional probability that is gained from knowledge about a specific attribute X is not different from the prior class probability - that means that this attribute values have the same mean (or probabilities for nominal attribute) for each class. Uniform discrete distribution with two events has equal "True" and "False" probabilities: {0.5,0.5}. The g) and h) cannot be satisfied simultaneously. Thus I think the question was whether the following statements can be correct or incorrect. The both can be correct: g) is correct when there are 1000 True and 1000 False (uniform distribution), such that class 1 has 500 True and 500 False, and class 2 is the same; h) is correct when there are 1500 True and 500 False (prespecified distribution) such that class 1 has 750 True and 250 False and class 2 is the same.

$\endgroup$
  • $\begingroup$ Yes the question was whether the following statements can be correct or incorrect. But the next statement is: i) For each example, the value of attribute X is sampled from {“true”, “false”} such that if the example is toxic, then the probability of selecting “true” is 0.75, while the probability of selecting “false” is 0.25, and if the example is non-toxic, the probability of selecting “true” is 0.25, while the probability of selecting “false” is 0.75. False .. Why is it not ok to change probabilities in this case? $\endgroup$ – Mona Rifaat Sep 30 '12 at 13:22
  • $\begingroup$ Becouse in this case the conditional probability will be different from prior probability: P(toxic|X=True)=0.75; P(toxic|X=False)=0.25 against prior P(toxic)=0.5 $\endgroup$ – O_Devinyak Sep 30 '12 at 20:18
  • $\begingroup$ The conditional and prior probabilities equivalence is observed when both classes have the same number of __True__s and __False__s. Statements g) and h) agree with this demand, but i) does not. In i) class 1 (toxic) has 1000*0.75=750 True and 250 False, but class 2 (non-toxic) is not the same. Class 2 has vice versa: 250 True and 750 False. $\endgroup$ – O_Devinyak Sep 30 '12 at 20:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.