| PRTools contents |
MLRC
W = A*(WU*MLRC)
W = WT*MLRC(B*WT)
D = C*W
| Input | |
| A | Dataset used for training base classifiers as well as combiner |
| B | Dataset used for training combiner of trained base classifiers |
| C | Dataset used for testing (executing) the combiner |
| WU | Set of untrained base classifiers, see STACKED |
| WT | Set of trained base classifiers, see STACKED |
| Output | |
| W | Trained Muli-response Linear Regression Combiner |
| D | Dataset with prob. products (over base classifiers) per class |
Using dataset A that contains the posterior probabilities of each instance belonging to each class predicted by the base classifiers to train a multi-response linear regression combiner. If the original classification problem has K classes, it is converted into K seperate regression problems, where the problem for class c has instances with responses equal to 1 when they have label c and zero otherwise. Put in another way, this function establish a multi-response linear regression model for each class and utilize these models to estimate the probability that the instances belong to each class. Note that in the model for class c, only the probabilities of class c predicted by the set of base classifiers are used.
1. Ting, KM, Witten IH. Issues in stacked generalization, Journal of Artificial Intelligent Research, 1999, 10: 271-289.
2. Dzeroski S, Zenko B. Is combining classifiers with stacking better than selecting the best one? Machine Learning, 2004, 54(3): 255-273.
datasets, mappings, stacked, classc, testd, labeld,
| PRTools contents |