I was reviewing homework 1 more this evening, and I think I may have found a discrepancy. . In item #7, it says: . Your program should use the distribution command line argument to determine the ''distribution'' that your training and test examples are drawn from. Your program should recognize the values ''bool'' for the uniform boolean distribution and ''sphere'', for the uniform unit spherical distribution. '''In the case where the input file is a nested boolean function, this argument should be ignored and the uniform boolean distribution should be used'''. . In contrast, for item #9, it says: . You should design and '''carry out experiments to see how well the perceptron update rules let you learn a nested boolean formula under the uniform boolean and <u>uniform unit spherical distributions</u>'''. In the perceptron update rule, uniform boolean distribution case in particular, do your experiments jive with our PAC learning results from class? . If the unit spherical mode should not be allowed with the nested boolean function, how should we reconcile these two items?

-- HW01: Task 9 Unit Spherical with Nested Boolean Function

Yes. I think when I wrote the 9th and 10th HW item I was trying to get a more clean split between the type of experiments to be carried out. This probably isn't possible for the 9th item. I rewrote it now so that it is consistent with the earlier remark that you only should use the uniform distribution with nested boolean functions.