Pocket Algorithm

[wolszhang]

Hi, all, i am currently coding pocket algorithm. However, since we are using it on a non-separable data, the PLA is never going to converge and thus run into infinite loops. Do i just set a break point or it there any other stoping conditions?

[wolszhang]

AND BTW, hw1 specifies that we could only use the training data, but the first problem asks us to plot the train data and test set data.

[gregoryw]

As I understand it, the idea of the pocket algorithm is that you save the best weights, and stop at an arbitrary point, say after 10,000 iterations. Then return that best weight. I think part1 of the homework is saying that you are only supposed to train off the training set, but then plot both sets with the separator.