A strong classifier is generated after the number of weak classifiers by a combination of a certain strategy: 1. ensemble learning.
Bagging bagging: a plurality of independent evaluation, a random sampling with replacement. Representative: Random Forests
Enhance law boosting: giving weight to automatically adjust the weights at the end of each round.
stacking
2. Portfolio Strategy: average method (the return value type commonly used), the voting method, learning method (stacking)
3. Integrated algorithm module ensemble
Random Forest classifier
4. Important parameters:. A tree and the same, the control group evaluator
criterion、max_depth、min_samples_leaf、min_samples_split、max_features、min_impurity_decrease
Number b.n_estimators forest trees, i.e. the number the better the group evaluator generally greater effect
c.random_state control pattern generation forests
d.bootstrap control sampling techniques, default true, on behalf of the random sampling with replacement
About 37% will not participate in the training data modeling, data are available outside the bag when n is large enough to test, when the instantiated to true oob_score
5. Important attributes: estimators_ view the status of forest tree
oob_score_ view bag The test result data
feature_importances_
6. Interface: apply, fit, predict, score
predict_proba returning each test sample corresponding to each class is assigned a probability tag
7. used to compose random forest classification tree were to exceed the forecast accuracy rate of 50%.
Random forest regression is
8. important parameters: criterion
9. important attributes: estimators_, oob_score_, feature_importances_
10. Interface: apply, fit, predict, score (no predict_proba)