ml intro Flashcards
pass exam
The intersection of two convex sets is always convex.
TRUE
The union of two convex sets is always convex.
FALSE
The function f(x)=7+3x1+999x2 is linear.
FALSE
A logistic function determines the threshold according to which the output of a linear model is classified.
FALSE Because YOU decide the threshold.
In order to predict y, x2 is more informative than x1.
FALSE It’s exactly the other way around: X1 is more informative. This is because:
I(x1|y)=1/4log(1/4/(1/4+1/2))+…+…+…=1/4+1/4+1/4+1/4=1
The mutual information between x1 and y is 1.
TRUE Because we have exactly 2 0 and 2 1.
The linear correlation coefficient between xi and yi may change if x values are centered by subtracting their mean value μ in the following manner: xi−μ
FALSE Centering does not affect the linear correlation
The entropy of a uniform probability distribution of n events is log2(n).
TRUE
The use of Chi-square test to deny statistical independence means that, for example, a term in a phrase should not be used as a feature because the square of the number of nearest neighbors is too large.
FALSE No relationship between Chi-square test and nearest neighbors.
If the Pearson correlation coefficient between two data features is zero, then such features are independent.
FALSE Pearson correlation ahd Mutual information are not related. They measures two completely different things.
If the Pearson correlation coefficient between two data features is zero, the Mutual Information between such features is also zero.
FALSE
If the number of input variables is 99, and one starts training from 55 different examples, the parameters of the linear model obtaining zero error on the examples can always be determined.
TRUE
The number of input variables need to be equal or greater than the number of examples for interpolation.
The measurements of two phenomena are different in a statistically significant way if one can demonstrate in a theorem that the two measurements will never be equal.
FALSE
Because we only care of statistically significant results, not theorems.
A result is statistically significant when it is obtained by democratic means, asking for the opinion of the largest possible number of experts.
FALSE
For every k>0k>0, in leave-one-out cross-validation, one of the k partitions is left out as validation data and the other partitions are used as training data.
FALSE