Machine LearningMachine Learning

Erica MelisErica Melis

Machine Learning

Introduction to Machine Learning

Decision Trees

Overfitting

A Little Introduction OnlyA Little Introduction Only

Artificial Neuronal Nets

Machine Learning

Why Machine Learning (1)Why Machine Learning (1)

• Growing flood of online data

• Budding industry

• Computational power is available

• progress in algorithms and theory

Machine Learning

Why Machine Learning (2)Why Machine Learning (2)

• Data mining: using historical data to improve decision– medical records ⇒ medical knowledge

– log data to model user

• Software applications we can’t program by hand– autonomous driving

– speech recognition

• Self customizing programs– Newsreader that learns user interests

Machine Learning

Some success storiesSome success stories

• Data Mining, Lernen im Web• Analysis of astronomical data• Human Speech Recognition• Handwriting recognition• Fraudulent Use of Credit Cards• Drive Autonomous Vehicles• Predict Stock Rates• Intelligent Elevator Control• World champion Backgammon• Robot Soccer• DNA Classification

Machine Learning

Problems Too Difficult to Program by HandProblems Too Difficult to Program by Hand

ALVINN drives 70 mph on highways

Machine Learning

Credit Risk AnalysisCredit Risk Analysis

If Other-Delinquent-Accounts > 2, and Number-Delinquent-Billing-Cycles > 1Then Profitable-Customer? = No [Deny Credit Card application]

If Other-Delinquent-Accounts = 0, and (Income > $30k) OR (Years-of-Credit > 3)Then Profitable-Customer? = Yes [Accept Credit Card application]

Machine Learning, T. Mitchell, McGraw Hill, 1997

Machine Learning

Typical Data Mining TaskTypical Data Mining Task

• 9714 patient records, each describing a pregnancy and birth• Each patient record contains 215 features

• Classes of future patients at high risk for Emergency Cesarean Section

Learn to predict:

Given:

Machine Learning, T. Mitchell, McGraw Hill, 1997

Machine Learning

Datamining ResultDatamining Result

IF No previous vaginal delivery, and Abnormal 2nd Trimester Ultrasound, and Malpresentation at admissionTHEN Probability of Emergency C-Section is 0.6

Over training data: 26/41 = .63, Over test data: 12/20 = .60

Machine Learning, T. Mitchell, McGraw Hill, 1997

Machine Learning

Machine Learning

How does an Agent learn?How does an Agent learn?

Priorknowledge

HypothesesKnowledge-basedinductive learning

Observations Predictions

EH

B

Machine Learning

Machine Learning TechniquesMachine Learning Techniques

• Decision tree learning• Artificial neural networks• Naive Bayes• Bayesian Net structures• Instance-based learning• Reinforcement learning• Genetic algorithms• Support vector machines• Explanation Based Learning• Inductive logic programming

Machine Learning

What is the Learning Problem?What is the Learning Problem?

• Improve over Task T• with respect to performance measure P• based on experience E

Learning = Improving with experience at some task

Machine Learning

The Game of CheckersThe Game of Checkers

Machine Learning

Learning to Play CheckersLearning to Play Checkers

• T: Play checkers• P: Percent of games won in world tournament..• E: games played against self..

• What exactly should be learned?• How shall it be represented?• What specific algorithm to learn it?

Machine Learning

A Representation for Learned Function A Representation for Learned Function V’(b)V’(b)

• x1: number of black pieces on board b

• x2 :number of red pieces on board b

• x3 :number of black kings on board b

• x4 number of red kings on board b

• x5 number of read pieces threatened by black (i.e., which can be taken on black’s next turn)

• x6 number of black pieces threatened by red

V’(b)= w0 + w1* x1 + w2* x2+ w3* x3+ w4*x4 + w5* x5+ w6*x6

Target function: V: Board IR

Target function representation:

Machine Learning

Function Approximation Algorithm*Function Approximation Algorithm*

• V(b): the true target function• V’(b): the learned function

• Vtrain(b): the training value

• (b, Vtrain(b)) training example

• Vtrain(b) ← V’(Successor(b)) for intermediate b

One rule for estimating training values:

Machine Learning

Contd: Choose Weight Tuning Rule*Contd: Choose Weight Tuning Rule*

• Select a training example b at random

1. Compute error(b) with current weightserror(b) = Vtrain(b) – V’(b)

2. For each board feature xi, update weight wi:wi ← wi + c * xi * error(b)

c is small constant to moderate the rate of learning

Do repeatedly:

LMS Weight update rule:

Machine Learning

...A.L. Samuel

Machine Learning

Design Choices for Checker LearningDesign Choices for Checker Learning

Machine Learning

Introduction to Machine Learning

Inductive Learning

Decision TreesEnsemble LearningOverfitting

Artificial Neuronal Nets

OverviewOverview

Machine Learning

Supervised Inductive Learning (1)Supervised Inductive Learning (1)

Why is learning difficult?

• inductive learning generalizes from specific examples cannot be proven true; it can only be proven false

• not easy to tell whether hypothesis h is a good approximation of a target function f

• complexity of hypothesis – fitting data

Machine Learning

Supervised Inductive Learning (2)Supervised Inductive Learning (2)

To generalize beyond the specific examples, one needs constraints or biases on what h is best.

• the overall class of candidate hypotheses restricted hypothesis space bias

• a metric for comparing candidate hypotheses to determine whether one is better than another preference bias

For that purpose, one has to specify

Machine Learning

Supervised Inductive Learning (3)Supervised Inductive Learning (3)

Having fixed the bias, learning can be considered as search in the hypothesis space which is guided by the used preference bias.

Machine Learning

Decision Tree Learning Decision Tree Learning Quinlan86, Feigenbaum61Quinlan86, Feigenbaum61

temperature = hot & windy = true & humidity = normal & outlook = sunny PlayTennis = ?

Goal predicate: PlayTennisHypotheses space:Preference bias:

Machine Learning

Illustrating Example Illustrating Example (RusselNorvig)(RusselNorvig)

The problem: wait for a table in a restaurant?

Machine Learning

Illustrating Example: Training DataIllustrating Example: Training Data

Machine Learning

A Decision Tree for A Decision Tree for WillWait (WillWait (SR)SR)

Machine Learning

Path in the Decision Tree

TAFEL

Machine Learning

General ApproachGeneral Approach

• let A1, A2, ..., and An be discrete attributes, i.e. each attribute has finitely many values

• let B be another discrete attribute, the goal attribute

Learning goal:

learn a function f: A1 x A2 x ... x An B

Examples:

elements from A1 x A2 x ... x An x B

Machine Learning

General ApproachGeneral Approach

Restricted hypothesis space bias:

the collection of all decision trees over the attributes A1, A2, ..., An, and B forms the set of possible candidate hypotheses

Preference bias:

prefer small trees consistent with the training examples

Machine Learning

Decision Trees: definition Decision Trees: definition for recordfor record

A decision tree over the attributes A1, A2,.., An, and B is a tree in which

• each non-leaf node is labelled with one of the attributes A1, A2, ..., and An

• each leaf node is labelled with one of the possible values for the goal attribute B

• a non-leaf node with the label Ai has as many outgoing arcs as there are possible values for the attribute Ai; each arc is labelled with one of the possible values for Ai

Machine Learning

Decision Trees: application of tree Decision Trees: application of tree for recordfor record

Let x be an element from A1 x A2 x ... x An and let T be a decision tree.

The element x is processed by the tree T starting at the root and following the appropriate arc until a leaf is reached. Moreover, x receives the value that is assigned to the leaf reached.

Machine Learning

Expressiveness of Decision TreesExpressiveness of Decision Trees

Any boolean function can be written as a decision tree.

001

111

110

000

BA2A1

A1

A2 A2

0 1 0 1

1

11

0

0 0

Machine Learning

Decision Trees Decision Trees

• fully expressive within the class of propositional languages

• in some cases, decision trees are not appropriate

sometimes exponentially large decision trees (e.g. parity function; returns 1 iff an even number of inputs are 1) replicated subtree problem e.g. when coding the following two rules in a tree:

„if A1 and A2 then B“„if A3 and A4 then B“

Machine Learning

Decision Trees Decision Trees

Finding a smallest decision tree that is consistent with a set of examples presented is an NP-hard problem.

smallest „=“ minimal in the overall number of nodes

instead of constructing a smallest decision tree the focus is on the construction of a pretty small one

greedy algorithm

Machine Learning

Inducing Decision Trees Algorithm Inducing Decision Trees Algorithm for recordfor record

function DECISION-TREE-LEARNING(examples, attribs, default) returns a decision treeinputs: examples, set of examples

attribs, set of attributesdefault, default value for the goal predicate

if examples is empty then return default else if all examples have the same classification

then return the classificationelse if attribs is empty then return

MAJORITY-VALUE(examples)else

best CHOOSE-ATTRIBUTE(attribs, examples)tree a new decision tree with root test bestm MAJORITY-VALUE(examplesi)

for each value vi of best doexamplesi {elements of examples with best = vi}subtree DECISION-TREE-LEARNING(examplesi,

attribs – best, m)add a branch to tree with label vi and subtree subtree

return tree

Machine Learning

Machine Learning

Training ExamplesTraining ExamplesDay Outlook Temperature Humidity Wind PlayTennis

D1 Sunny Hot High Weak No

D2 Sunny Hot High Strong No

D3 Overcast Hot High Weak Yes

D4 Rain Mild High Weak Yes

D5 Rain Cool Normal Weak Yes

D6 Rain Cool Normal Strong No

D7 Overcast Cool Normal Strong Yes

D8 Sunny Mild High Weak No

D9 Sunny Cool Normal Weak Yes

D10 Rain Mild Normal Weak Yes

D11 Sunny Mild Normal Strong Yes

D12 Overcast Mild High Strong Yes

D13 Overcast Hot Normal Weak Yes

D14 Rain Mild High Strong No

T. Mitchell, 1997

Machine Learning

Machine Learning

Machine Learning

Entropy Entropy n = 2n = 2

• S is a sample of training examples• p+ is the proportion of positive examples in S• p- is the proportion of negative examples in S• Entropy measures the impurity of S

Entropy(S) ≡ -p+ log2p+ - p- log2 p-

•

Machine Learning

Machine Learning

Machine Learning

Machine Learning

Machine Learning

Example Example WillWait WillWait (do it yourself)(do it yourself)

the problem of whether to wait for a table in a restaurant

Machine Learning

WillWaitWillWait (do it yourself) (do it yourself)

Which attribute to choose?

Machine Learning

Learned TreeLearned Tree WillWait WillWait

Machine Learning

Assessing Decision TreesAssessing Decision Trees

Assessing the performance of a learning algorithm:a learning algorithm has done a good job, if its finalhypothesis predicts the value of the goal attribute

of unseen examples correctly

General strategy (cross-validation)1. collect a large set of examples 2. divide it into two disjoint sets: the training set and the test set3. apply the learning algorithm to the training set, generating a

hypothesis h4. measure the quality of h applied to the test set5. repeat steps 1 to 4 for different sizes of training sets and

different randomly selected training sets of each size

Machine Learning

When is decision tree learning appropriate?When is decision tree learning appropriate?

• Instances represented by attribute-value pairs• Target function has discret values• Disjunctive descriptions may be required• Training data may contain missing or noisy data

Machine Learning

Extensions and Problems Extensions and Problems

• dealing with continuous attributes - select thresholds defining intervals; as a result each

interval becomes a discrete value- dynamic programming methods to find appropriate

split points still expensive

• missing attributes- introduce a new value- use default values (e.g. the majority value)

• highly-branching attributes- e.g. Date has a different value for every example;

information gain measure: GainRatio = Gain/SplitInformation penalizes broad+uniform

Machine Learning

Extensions and Problems Extensions and Problems

• noisee.g. two or more examples with the same description but

different classifications -> leaf nodes report the majority classification for its setOr report estimated probability (relative frequency)

• overfittingthe learning algorithm uses irrelevant attributes to find a

hypothesis consistent with all examples; pruning techniques; e.g. new non-leaf nodes will only be introduced if the information gain is larger than a particular threshold

Machine Learning

Introduction to Machine Learning

Inductive Learning: Decision Trees Overfitting Artificial Neuronal Nets

OverviewOverview

Machine Learning

Overfitting in Decision TreesOverfitting in Decision Trees

Consider adding training example #15:

Sunny, Hot, Normal, Strong, PlayTennis = No

What effect on earlier tree?

Machine Learning

OverfittingOverfitting

Consider error of hypothesis h over

• training data: errortrain(h)• entire distribution D of data: errorD(h)

Hypothesis h ∈ H overfits training data if there is an alternative hypothesis h’ ∈ H such that

anderrortrain(h) < errortrain(h’)

errorD(h) > errorD(h’)

Machine Learning

Overfitting in Decision Tree LearningOverfitting in Decision Tree Learning

T. Mitchell, 1997

Machine Learning

Avoiding OverfittingAvoiding Overfitting

• stop growing when data split not statistically significant• grow full tree, then post-prune

• Measure performance over training data (threshold)

• Statistical significance test whether expanding or pruning at node will improve beyond training set 2

• Measure performance over separate validation data set (utility of post-pruning) general cross-validation

• Use explicit measure for encoding complexity of tree, train MDL heuristics

How to select “best” tree:

Machine Learning

Reduced-Error PruningReduced-Error Pruning

lecture slides for textbook Machine Learning, T. Mitchell, McGraw Hill, 1997

1. Evaluate impact on validation set of pruning each possible node (plus those below it)

2. Greedily remove the one that most improves validation set accuracy

• produces smallest version of most accurate subtree• What if data is limited??

Split data into training and validation set

Do until further pruning is harmful:

Machine Learning

Effect of Reduced-Error PruningEffect of Reduced-Error Pruning

lecture slides for textbook Machine Learning, T. Mitchell, McGraw Hill, 1997

Chapter 6.1 – Learning from Observation

Software that Customizes to UserSoftware that Customizes to User

Recommender systems(Amazon..)