One Hot Encoding, I cannot handle words, I only speak in digits. So red, blue and green give me the fidgets. I make three columns, one for every name. A one where it matches, zero for the rest of the game. Simple and clean, no order implied, What encoding technique sits by your side?, Feature Scaling, I punish the large and I shrink down the tall. So no single dimension can dominate all. Euclidean distances weep without my hand. KNN and SVM on me heavily stand., SMOTE, I birth new neighbors from the ones already there. Interpolating between them with geometric care. The minority finds strength in what I create. Synthetic yet purposeful, I balance the weight., Random Forest / Bagging with decision trees, My base models overfit, that is part of the plan. Different bootstrap samples make each tree a different man. I average their sins and the variance drops low. What ensemble strategy puts on this particular show?, AdaBoost, I reweight the samples that the ensemble got wrong. Making the hard examples more prominent and strong. My model weights follow the logit of the error rate. What boosting algorithm uses this reweighting trait?, Stacking, I sit on top and watch the base models perform below. Learning from their predictions how the final answer should go. Linear models make me fast, gradient boosting makes me strong. What ensemble technique has been the plan all along?.

Lab3

由
列印
嵌入

排行榜

視覺風格

選項

切換範本

恢復自動保存: ?