SPIDER The Spider Objects
GET_MEAN object R=get_mean(A,[L]) An object which calculates the mean(s) R of results with algorithms A, together with the standard error (std. dev / sqrt(number of trials)) optionally calculating the loss with loss type L (see help loss) e.g 'class_loss',.. (NOTE: input is a string, not a loss object). when trained, e.g: train(get_mean(cv(svm)),gen(toy)) Hyperparameters: take_average=0 -- method of taking average over groups 1: average over inside leaf of tree of groups 2: average over outside leaf of tree of groups 0: try to guess which is most appropriate 'take_average' is necessary because if you have a group of groups of algorithms you could wish to take the mean in two different ways: e.g with group({group({a b} group({c d}))}) you may either wish to average a&b and c&d or a&c and b&d. NOTE: The same effect can usually be achieved by doing get_mean(r) and get_mean(r') with a data object r Alternatively, can be used in a feed forward network, e.g: train(chain({ group(cv(svm)) get_mean }),gen(toy)) Can also be called like a function with A=get_mean(D,[L]) calculates the mean for groups of data objects D.