Open henryzord opened 7 years ago
When running a N-fold cross-validation, in the plotting of final results, Ardennes is actually plotting the mean accuracy among folds, instead of plotting the correct value. This is exemplified below:
WRONG: (0.93 + 0.87 + 0.95 + 0.88 + 0.75)/5 = 0,876 CORRECT: ((17 0.93) + (17 0.87) + (17 0.95) + (17 0.88) + (21 * 0.75)) / 89 = 0.870
Obviously, when all folds have the same size, this phenomena will not occur.
When running a N-fold cross-validation, in the plotting of final results, Ardennes is actually plotting the mean accuracy among folds, instead of plotting the correct value. This is exemplified below:
WRONG: (0.93 + 0.87 + 0.95 + 0.88 + 0.75)/5 = 0,876 CORRECT: ((17 0.93) + (17 0.87) + (17 0.95) + (17 0.88) + (21 * 0.75)) / 89 = 0.870
Obviously, when all folds have the same size, this phenomena will not occur.