On 1/12/16 12:16 PM, Phil Steitz wrote: > On 1/12/16 11:44 AM, Gilles wrote: >> Hi. >> >> This unit test class uses a method "checkNextLongUniform" which in >> turn calls >> "assertChiSquareAccept". When failure is detected, the latter >> produces an output >> like the following: >> >> [...] >> checkNextLongUniform:228 Chisquare test failed p-value = >> 0.003941271563188575 chisquare statistic = 27.439999999999998. >> value expected observed >> 1 0.08 91 >> 2 0.08 75 >> 3 0.08 106 >> 4 0.08 76 >> 5 0.08 71 >> 6 0.08 89 >> 7 0.08 87 >> 8 0.08 96 >> 9 0.08 61 >> 10 0.08 65 >> 11 0.08 80 >> 12 0.08 103 >> >> For comparing, it is a bit awkward to have relative "expected" >> counts vs >> absolute "observed" counts... > Good point. It would be better to display the expected counts > rather than proportions in that column.
Actually, what is going on here is the robustness of our ChiSquareTest impl is compensating for what is arguably an error in the checkNextLongUniform test implementation. The ChiSquareTest rescales the expected array, which is being provided as proportions rather than counts by this test case. It would be better to modify the test to pass the expected (uniform) proportion times the sample size as expected count. So the right fix here is local to checkNextLongUniform. Phil > > Phil >> I also think that the format of the "expected" frequencies should >> display >> more than 2 digits after the decimal point. >> >> >> Regards, >> Gilles >> >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org >> For additional commands, e-mail: dev-h...@commons.apache.org >> >> --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org For additional commands, e-mail: dev-h...@commons.apache.org