0% found this document useful (0 votes)
5 views36 pages

Machine Learning: Regression & Ensemble Techniques

The document covers key concepts in machine learning, focusing on linear regression, classification, ensemble learning, and model optimization. It discusses techniques such as stochastic gradient descent, boosting, bagging, and regularization methods like L1 and L2. Additionally, it highlights the importance of developing machine learning systems and the theory behind learning from examples.
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
5 views36 pages

Machine Learning: Regression & Ensemble Techniques

The document covers key concepts in machine learning, focusing on linear regression, classification, ensemble learning, and model optimization. It discusses techniques such as stochastic gradient descent, boosting, bagging, and regularization methods like L1 and L2. Additionally, it highlights the importance of developing machine learning systems and the theory behind learning from examples.
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd

CSE 471: MACHINE LEARNING

Learning from Examples (Continued)


Outline
 Linear Regression and Classification
 Ensemble Learning
 Model Selection and Optimization
 The Theory of Learning
 Nonparametric Models
 Developing Machine Learning Systems
Linear Regression &
Classification
Linear Regression
 w∗ = argminw Loss(hw)
 Set partial derivatives to 0 to obtain the minima
Linear Regression
Gradient Descent

  - Learning rate
 Constant
 Decaying
The partial derivatives
The learning rules

For single
example

For Entire
training
set
Stochastic gradient descent
(SGD)
 Use a random sample of examples at
each step
 A minibatch of m out of N samples
 Convergence of minibatch SGD is not
strictly guaranteed
 It can oscillate around the minimum
without settling down
 Decaying learning rate can guarantee
convergence
 Useful in online learning
Multivariable Linear
Regression

Gradient Descent Step

Recall
[for
univariable
case]
Multivariable LR Vectorized
Assuming a dummy input
attribute xj,0 = 1, We can
vectorize the formulation

The
optimization
Problem
Multivariable LR Vectorized
 y hat – prediction on all examples
 X - Data marix. Each row is a multivariable sample
 w - Weight vector

 Loss over training dataset

 Set the gradient to 0

 Normal Equation
Ensemble Learning
Ensemble Learning
 Boosting
 Bootstrap aggregating (Bagging)
 Random forests
 Stacking
Boosting
Boosting Decision Stumps

• A decision stump is a DT with just one test (i.e. only the root)
• Notice that the test set accuracy improves slightly even after the training
accuracy reaches 1, i.e., after the ensemble fits the data exactly.
AdaBoost
AdaBoost

continue
Stacking
 Base models from different model classes
 SVM
 Logistic regression
 Decision Tree
 Augment feature vector with output of the
base models
 Train the ensemble model (say Logistic
regression)
 On the Validation Set
 Cross validation can be done
Stacking
Base
classifier-1

Base New training set


classifier-2 for meta
Meta
classifier
Training classifie
Base consisting of
set r
classifier-3 predictions from
base classifiers
.. . . ..
Base
Final
classifier-n
predicti
on

Taken from MSc thesis defense presentation of Qazi Farah Nawar


Stacking improves class
separation

This is a t-SNE plot from protein carbohydrate binding site prediction


problem. FC-1 to FC-5 are best 5 feature spaces. FC-6 is FC-1 augmented
with prediction values from the base layer (i.e., what the meta layer of
stacking ensemble will see . (Nawar et al.)
Stacking improves class
separation

This is a t-SNE plot from protein carbohydrate binding site prediction


problem. FC-1 to FC-5 are best 5 feature spaces. FC-6 is FC-1 augmented
with prediction values from the base layer (i.e., what the meta layer of
stacking ensemble will see . (Nawar et al.)
Stacking improves class
separation

This is a t-SNE plot from protein carbohydrate binding site prediction


problem. FC-1 to FC-5 are best 5 feature spaces. FC-6 is FC-1 augmented
with prediction values from the base layer (i.e., what the meta layer of
stacking ensemble will see . (Nawar et al.)
Bagging
 K distinct training sets by subsampling
 With replacement
 Reduces variance
 Most commonly used with decision trees
 DTs are very sensitive to changes in
examples
 Can be applied in parallel
Random forests
 A form of decision tree bagging
 Takes extra measures to diversify the base
DTs
 At each split point, attributes are sampled.
 Attribute with highest IG from the sample is
chosen
 Extremely Randomized Tree (ExtraTree)
 For an attribute, we randomly sample several
candidate values from a uniform distribution
over the attribute’s range
 Choose the value that results in highest IG
Random forests
 Can be created efficiently
 Sampled attribute space
 No need for DT pruning
 The trees can be built in parallel
 Out-of-bag error
 Mean error on each example, using only
the trees whose example set didn’t include
that particular example
 Can be used instead of CV error
Multivariable LR
Regularization
Cost(h) = EmpLoss(h) + λ Complexity(h)

 L1 regularization produces sparse model


 Easier for a human to understand
 Less likely to overfit
 You can use L1 loss with L2 regularization and vice versa.

 L1 regularization – LASSO
 L2 regularization - Ridge
L1 vs. L2 regularization

L1 regularization tends to produce a sparse model.


Linear Classification
Linear Classification

 Gradient is discontinuous
 Gradient is zero almost everywhere in weight space
except at those points where and at those points the
gradient is undefined
Perceptron Learning Rule

 If the output is correct (i.e., ) then the weights are not changed.

 If y is 1 but hw(x) is 0, then wi is increased when the corresponding input is


positive and decreased when is negative. This makes sense, because we
want to make w.x bigger so that outputs a 1.

 If is 0 but is 1, then is wi decreased when the corresponding input is positive


and increased when is negative. This makes sense, because we want to
make smaller so that outputs a 0.
Logistic Regression
Logistic Regression
Logistic Regression

Likelihood
of the
parameter
s
Maximizing the log
Likelihood

For single data point


Update rule
 j = j + (y - h(x))xj
 Identical to the update rule in linear
regression

You might also like