70 likes | 173 Views
Pool-based learning via Weighted Information Gain Measurements. Rafael Augusto Ferreira do Carmo carmorafael@gmail.com Daniel Pinto Coutinho daniel.pintocoutinho@gmail.com Jerffeson Teixeira de Souza jeff@larces.uece.br Universidade Estadual do Ceará
E N D
Pool-based learning via Weighted Information Gain Measurements Rafael Augusto Ferreira do Carmo carmorafael@gmail.com Daniel Pinto Coutinho daniel.pintocoutinho@gmail.com Jerffeson Teixeira de Souza jeff@larces.uece.br Universidade Estadual do Ceará Fortaleza - Brazil
Introduction • Active learning scenario • Binary classification problems • Pool of unlabeled examples • No prior information about class distribution • One labeled example as “seed” for learning
The Task • Select the as few “informative examples” as possible • Minimize the classification costs • Maximize the quality of the model
The Algorithm • What if this example is positive? • What if this example is negative? • Information Gain Ratio • Weight features