Consider the following methods for evaluating the performance of a learning algorithm in a particular setting.

Method $1.20$ examples are drawn at random. The method of $5-$fold cross validation is applied, giving the

error estimate e$1.$

Method $2.16$ examples are drawn at random. The classifier is trained using these examples. Another set

of $1000$ examples are drawn at random, and the classifier is tested on these examples giving the error

estimate e$2.$

Method $3.1000$ examples are drawn at random. The method of $5-$fold cross validation is applied, giving

the error estimate e$3.$

What is the likely relation between the error estimates? Mark all answers that you believe will hold with

probability greater than $0\mathrm{.}5.$