思维导图

where it is

AHTA题目 -> to finish

Suppose you perform 10,000 statistical tests on random microarray data with a family-wise error rate of 0.05 as the significance threshold. How many “significant” results do you expect from those 10,000 tests if power is not an issue?

FWER = 1-P(no type1 error for 10000 tests) = 0.05

Same as the previous question, but using a false discovery rate significance threshold of 0.05 instead of a family-wise error rate based one. How many “significant” results do you expect then?

FDR = 0.05, it means 5% of signicant result is FP

ML system design

handling skewed data - trade-off precision and recall

precision and recall

precision: 当我们预测得了癌症时候，真正得了癌症的人数
recall:所有得了癌症的人中，被我们正确检测到了得了癌症的数目