Statistical Learning Dong Liu Dept EEIS USTC Chapter
- Slides: 36
Statistical Learning Dong Liu Dept. EEIS, USTC
Chapter 8. Decision Tree 1. 2. 3. 4. Tree model Tree building Tree pruning Tree and ensemble 2020/9/30 Chap 8. Decision Tree 2
Taxonomy • How does the biologist determine the category of an animal? • A hierarchy of rules • • • 2020/9/30 Kingdom of Animalia Phylum of Chordata Class of Mammalia Order of Carnivora … Chap 8. Decision Tree 3
Taxonomy as tree 2020/9/30 Chap 8. Decision Tree 4
Decision tree Refund Yes No NO Mar. St Single, Divorced Tax. Inc <= 80 K NO 2020/9/30 Chap 8. Decision Tree Married NO > 80 K YES 5
Using decision tree Start from the root of the tree Refund Yes No NO Mar. St Single, Divorced Tax. Inc <= 80 K NO 2020/9/30 Married NO NO > 80 K YES Chap 8. Decision Tree 6
Tree models • A tree model consists of a set of conditions and a set of base models, organized in a tree • Each internal node represents a condition on input attributes • One condition is a division (split) of input space • Each leaf node represents a base model • Classification: a class (simplest case) or a classifier • Regression: a constant (simplest case) or a regressor 2020/9/30 Chap 8. Decision Tree 7
Chapter 8. Decision Tree 1. 2. 3. 4. Tree model Tree building Tree pruning Tree and ensemble 2020/9/30 Chap 8. Decision Tree 8
Tree induction • Assume we have defined the form of base models • How to find out the optimal tree structure (set of conditions, division of input space)? • Exhaustive search is computationally expensive • Heuristic approach: Hunt’s algorithm 2020/9/30 Chap 8. Decision Tree 9
Hunt’s algorithm • Input: A set of training data • Output: A classification tree or regression tree • Function • If need not or cannot be divided, return a leaf node • Else • Find an attribute of , say , and decide a condition • Divide into , according to the output of • • Let be the children of • Return 2020/9/30 Chap 8. Decision Tree 10
Example of Hunt’s algorithm (1) D= T= Refund Yes T 1=? No T 2=? D need be divided? Yes! D’s class labels have different values D can be divided? Yes! D’s input attributes have different values 2020/9/30 Chap 8. Decision Tree 11
Example of Hunt’s algorithm (2) D 1= T= Refund Yes NO T 1=? No T 2=? D 1 need be divided? No! T 1 is a leaf node 2020/9/30 Chap 8. Decision Tree 12
Example of Hunt’s algorithm (3) D 2= T= Refund Yes No NO Mar. St T 2=? Single, Divorced T 21=? 2020/9/30 Chap 8. Decision Tree Married T 22=? 13
Example of Hunt’s algorithm (4) D 22= T= Refund Yes No NO Mar. St Single, Divorced Tax. Inc <= 80 K NO Married NO T 22=? > 80 K YES D 22 can be divided? No! T 22 is a leaf node 2020/9/30 Chap 8. Decision Tree 14
Find an attribute and decide a condition Mar. St Discrete values: Multi-way or Two-way Which attribute & which condition shall be selected? {Single, Divorced} Single Married Divorced Mar. St Married Single {Married, Divorced} Define a criterion that describes the “gain” of dividing a set into several subsets Continuous values: Two-way or Multi-way 2020/9/30 Chap 8. Decision Tree 15
Purity of set • Purity of set describes how easily the set can be classified • For example, consider two sets with 0 -1 classes: {0, 0, 0, 1} vs. {0, 1, 0, 1} • Measures (using and to denote the percentages of class-0 and class-1) • Entropy: • Gini index: • Misclassification error: 2020/9/30 Chap 8. Decision Tree 16
Criterion to find attribute and decide condition • Information gain where is the entropy • Information gain ratio to suppress too many subsets • Gini index gain where is the Gini index 2020/9/30 Chap 8. Decision Tree 17
Example: Gini index gain • Before split: • After split with {Tax. Inc <= 97}: Training sample s 2020/9/30 Chap 8. Decision Tree 18
Chapter 8. Decision Tree 1. 2. 3. 4. Tree model Tree building Tree pruning Tree and ensemble 2020/9/30 Chap 8. Decision Tree 19
Control the complexity of tree • Using Hunt’s algorithm, we build a tree as accurate as possible, which may incur over-fitting • Two manners to control the complexity (thus to avoid overfitting) • Early termination: stop splitting, if the gain is less than a threshold, or if the tree is too deep, or if the set is too small • Tree pruning: remove branches from the tree so as to minimize the joint cost where is empirical risk (e. g. error rate of training data), is tree complexity (e. g. number of leaf nodes) 2020/9/30 Chap 8. Decision Tree 20
Tree pruning example (1) Refund Yes No NO Mar. St 3 correct Single, Divorced Tax. Inc <= 80 K NO 1 correct 2020/9/30 Chap 8. Decision Tree Married NO > 80 K YES 2 correct 1 error 3 correct 21
Tree pruning example (2) • We have different pruning selections Refund Yes NO 3 correct Single, Divorced Mar. St Single, Divorced Tax. Inc <= 80 K NO 1 correct 2020/9/30 Mar. St No Married NO > 80 K YES 2 correct 1 error 3 correct Chap 8. Decision Tree Tax. Inc <= 80 K NO 1 correct Married NO > 80 K YES 3 correct 1 error 3 correct 2 error 22
Tree pruning example (2) • We have different pruning selections Refund Yes No NO 3 correct Mar. St Single, Divorced Tax. Inc <= 80 K NO 1 correct 2020/9/30 YES No NO Mar. St Married 3 correct NO > 80 K Yes 2 correct 1 error Single, Divorced YES 3 correct 1 error Married NO 2 correct 1 error 3 correct Chap 8. Decision Tree 23
Tree pruning example (2) • We have different pruning selections Refund Yes No NO Mar. St 3 correct Single, Divorced Tax. Inc <= 80 K NO 1 correct 2020/9/30 Married YES No NO Tax. Inc 3 correct <= 80 K NO > 80 K Yes 2 correct 1 error 3 correct 1 error > 80 K YES 3 correct Chap 8. Decision Tree 24
Tree pruning example (2) • Refund Mar. St Single, Divorced Tax. Inc <= 80 K NO Married No NO Tax. Inc NO <= 80 K > 80 K YES NO Refund Yes No NO Mar. St Single, Divorced YES 2020/9/30 Yes Chap 8. Decision Tree > 80 K YES Married NO 25
Tree pruning example (3) • Continue pruning, keep 2 leaf nodes Refund Yes Mar. St Single, Divorced YES 3 correct 3 error Married NO 3 correct 1 error <= 80 K Tax. Inc NO 3 correct 1 error 2020/9/30 Chap 8. Decision Tree No YES 4 correct 3 error > 80 K YES 3 correct 3 error 26
Tree pruning example (4) • Continue pruning, keep 1 leaf node NO 6 correct 4 error 2020/9/30 Chap 8. Decision Tree 27
Tree pruning example (5) Refund • In summary Yes No NO Mar. St Married Single, Divorced Tax. Inc <= 80 K • Therefore, according to , optimal trees are different • • • nodes : four leaf nodes YES Refund Yes No NO Tax. Inc <= 80 K 2020/9/30 > 80 K NO : one leaf node : three leaf NO Chap 8. Decision Tree NO NO > 80 K YES 28
Chapter 8. Decision Tree 1. 2. 3. 4. Tree model Tree building Tree pruning Tree and ensemble 2020/9/30 Chap 8. Decision Tree 29
Decision tree for regression • Consider the simplest case: each leaf node corresponds to a constant • Each time to find attribute and decide condition, is to minimize the (e. g. quadratic) cost • The final regression tree is indeed a piecewise constant function 2020/9/30 Chap 8. Decision Tree 30
Equivalence of decision tree and boosting tree for regression • Hunt’s algorithm: “divide and conquer”, conditions + base models • Boosting: linear combination of base models Model 1 Model 3 Model 1 + Model 2 + Model 3 Model 2 Each model is a constant 2020/9/30 Each model is a decision stump Chap 8. Decision Tree 31
Implementation • ID 3: use information gain • C 4. 5: use information gain ratio (by default), one of most famous classification algorithm • CART: use Gini index (for classification) and quadratic cost (for regression), only 2 -way split • According to , increase gradually to get a series of subtrees • Determine which subtree is optimal according to validation (or cross validation) 2020/9/30 Chap 8. Decision Tree 32
Remarks on tree models • Easy to interpret • Irrelevant/redundant attributes can be filtered out • Good at discrete variables • How to handle complex conditions e. g. ? Oblique tree 2020/9/30 Chap 8. Decision Tree 33
Random forest • Combination of decision tree and ensemble learning • According to bagging, firstly generate multiple datasets (bootstrap samples), each of which gives rise to a tree model • During tree building, consider a random subset of features when splitting 2020/9/30 Chap 8. Decision Tree 34
Chapter summary Dictionary Toolbox • Decision tree • Gini index • Pruning (of decision tree) • CART • C 4. 5 • Hunt’s algorithm • Information gain, ~ ratio • Random forest 2020/9/30 Chap 5. Non-Parametric Supervised Learning 35
Home exercises 2020/9/30 Chap 1. Linear Regression 36
- Dong liu ustc
- Dong liu ustc
- Dong liu ustc
- Ring a ding ding dong christmas song
- Alex liu cecilia liu
- Líu líu lo lo ta ca hát say sưa
- Ustc repeg
- Lunc and ustc
- Mail.ustc.edu.n
- Ustc search
- Cuadro comparativo e-learning b-learning m-learning
- Linguistic determinism ap psychology
- Statistical mechanics of deep learning
- Dept nmr spectroscopy
- Department of agriculture consumer services
- Organizational structure of finance department
- Worcester ma building dept
- Dept. name of organization
- Mn dept of education
- Ms department of finance and administration
- Dept. name of organization
- Employment first ohio
- Affiliation poster presentation
- Vaginal dept
- Gome dept
- Gome dept
- Nyttofunktion
- Gome dept
- Hoe dept
- Firefighter oral interview questions
- Maine department of agriculture conservation and forestry
- Dept of education
- Florida dept of agriculture and consumer services
- Florida dept of agriculture and consumer services
- Dept a
- Central islip fire dept
- Dss rowan county