Overview. Non-Parametrics Models Definitions KNN. Ensemble Methods Definitions, Examples Random Forests. Clustering. k-means Clustering 2 / 8

Transcription

1 Tutorial 3 1 / 8

2 Overview Non-Parametrics Models Definitions KNN Ensemble Methods Definitions, Examples Random Forests Clustering Definitions, Examples k-means Clustering 2 / 8

3 Non-Parametrics Models Definitions Definitions 3 / 8

4 Non-Parametrics Models Definitions Definitions Parametric Models 3 / 8

5 Non-Parametrics Models Definitions Definitions Parametric Models Fixed number of parameters - learned (estimated) from data More data More accurate models. 3 / 8

6 Non-Parametrics Models Definitions Definitions Parametric Models Fixed number of parameters - learned (estimated) from data More data More accurate models. Non-parametric Models 3 / 8

7 Non-Parametrics Models Definitions Definitions Parametric Models Fixed number of parameters - learned (estimated) from data More data More accurate models. Non-parametric Models Number of parameters grows with the amount of data More data More complex models. 3 / 8

8 Non-Parametrics Models Definitions Definitions Parametric Models Fixed number of parameters - learned (estimated) from data More data More accurate models. Non-parametric Models Number of parameters grows with the amount of data More data More complex models. Parametric or Non-parametric? What are the parameters? Decision Trees Naive Bayes KNN Random Forests K-Means Clustering 3 / 8

9 Non-Parametrics Models KNN k-nearest Neighbour 4 / 8

10 Non-Parametrics Models KNN k-nearest Neighbour How does it work? 4 / 8

11 Non-Parametrics Models KNN k-nearest Neighbour How does it work? What is the effect of k with respect to the fundamental tradeoff in machine learning? 4 / 8

12 Non-Parametrics Models KNN k-nearest Neighbour How does it work? What is the effect of k with respect to the fundamental tradeoff in machine learning? What is the runtime of a naive implementation? How could you speed this up? 4 / 8

13 Ensemble Methods Definitions, Examples Ensemble Methods 5 / 8

14 Ensemble Methods Definitions, Examples Ensemble Methods Learning algorithms that take classifiers as input and use the output of each classifier to determine a classification 5 / 8

15 Ensemble Methods Definitions, Examples Ensemble Methods Learning algorithms that take classifiers as input and use the output of each classifier to determine a classification Averaging Take the average of the outputs of each classifier (or mode if categorical) 5 / 8

16 Ensemble Methods Definitions, Examples Ensemble Methods Learning algorithms that take classifiers as input and use the output of each classifier to determine a classification Averaging Take the average of the outputs of each classifier (or mode if categorical) Bagging Each classifier in the ensemble votes on an output with equal weight Each classifier is trained with a random subset of the training set 5 / 8

17 Ensemble Methods Definitions, Examples Ensemble Methods Learning algorithms that take classifiers as input and use the output of each classifier to determine a classification Averaging Take the average of the outputs of each classifier (or mode if categorical) Bagging Each classifier in the ensemble votes on an output with equal weight Each classifier is trained with a random subset of the training set Boosting Incrementally build the ensemble. When training new models higher weight is given to data that was mis-classified by previous models 5 / 8

18 Ensemble Methods Definitions, Examples Ensemble Methods Learning algorithms that take classifiers as input and use the output of each classifier to determine a classification Averaging Take the average of the outputs of each classifier (or mode if categorical) Bagging Each classifier in the ensemble votes on an output with equal weight Each classifier is trained with a random subset of the training set Boosting Incrementally build the ensemble. When training new models higher weight is given to data that was mis-classified by previous models Stacking Train a classifier to combine the predictions of the other classifiers 5 / 8

19 Ensemble Methods Definitions, Examples Ensemble Methods Learning algorithms that take classifiers as input and use the output of each classifier to determine a classification Averaging Take the average of the outputs of each classifier (or mode if categorical) Bagging Each classifier in the ensemble votes on an output with equal weight Each classifier is trained with a random subset of the training set Boosting Incrementally build the ensemble. When training new models higher weight is given to data that was mis-classified by previous models Stacking Train a classifier to combine the predictions of the other classifiers And more! 5 / 8

20 Ensemble Methods Random Forests Random Forests 6 / 8

21 Ensemble Methods Random Forests Random Forests How do they work? How do you train them? 6 / 8

22 Ensemble Methods Random Forests Random Forests How do they work? How do you train them? 1. Create several bootstrap samples of the data 6 / 8

23 Ensemble Methods Random Forests Random Forests How do they work? How do you train them? 1. Create several bootstrap samples of the data 2. Train a random decision tree on each bootstrap sample 6 / 8

24 Ensemble Methods Random Forests Random Forests How do they work? How do you train them? 1. Create several bootstrap samples of the data 2. Train a random decision tree on each bootstrap sample 3. Test by averaging the predictions of each tree 6 / 8

25 Ensemble Methods Random Forests Random Forests How do they work? How do you train them? 1. Create several bootstrap samples of the data 2. Train a random decision tree on each bootstrap sample 3. Test by averaging the predictions of each tree How does the number of trees affect the fundmental tradeoff of machine learning? 6 / 8

26 Ensemble Methods Random Forests Random Forests How do they work? How do you train them? 1. Create several bootstrap samples of the data 2. Train a random decision tree on each bootstrap sample 3. Test by averaging the predictions of each tree How does the number of trees affect the fundmental tradeoff of machine learning? How does the amount of randomness in the trees affect the fundamental tradeoff of machine learning? 6 / 8

27 Clustering Definitions, Examples Clustering 7 / 8

28 Clustering Definitions, Examples Clustering An unsupervised method - not given labels, but want to learn something about the data Specifically the classes, or groups, that the data falls into 7 / 8

29 Clustering Definitions, Examples Clustering An unsupervised method - not given labels, but want to learn something about the data Specifically the classes, or groups, that the data falls into Classes are determined by similarty between data and dissimilarity to other classes 7 / 8

30 Clustering Definitions, Examples Clustering An unsupervised method - not given labels, but want to learn something about the data Specifically the classes, or groups, that the data falls into Classes are determined by similarty between data and dissimilarity to other classes e.g. Types of genes, variants of a disease, topics on Wikipedia, friends on Facebook, etc. 7 / 8

31 Clustering k-means Clustering k-means 8 / 8

32 Clustering k-means Clustering k-means How does it work? 8 / 8

33 Clustering k-means Clustering k-means How does it work? K++ means - what problem does this address? 8 / 8

34 Clustering k-means Clustering k-means How does it work? K++ means - what problem does this address? Label switching problem 8 / 8