Ensemble Learning

[email protected]
472C20Y4RY

Proprietary content. ©Great Learning.

This AllforRights
file is meant Reserved.
personal Unauthorizedonly.
use by [email protected] use or distribution prohibited
Sharing or publishing the contents in part or full is liable for legal action.
 Ensemble Methods
• Ensembles are machine learning methods for combining predictions
from multiple separate models.

• The central motivation is rooted under the belief that a committee of

experts working together can perform better than a single expert.

Training Data

Model-1 Model-2 Model-3 … Model-n

[email protected]
472C20Y4RY

Test Data

Model-1 Model-2 Model-3 … Model-n

Prediction-1 Prediction-2 Prediction-3 … Prediction-n

Combined Prediction

2
Proprietary content. ©Great Learning.
This AllforRights
file is meant Reserved.
personal Unauthorizedonly.
use by [email protected] use or distribution prohibited
Sharing or publishing the contents in part or full is liable for legal action.
 • Both Regression and Classification can be done using Ensemble learning

• Combining the individual predictions can be done by using either voting or averaging

• The individual ensemble learners need to be:

• Different from each other (independent errors)

• Can be weak (slightly better than random): Because of the number of models in
an Ensemble method, computational requirements are much higher than that of
[email protected] evaluating a single model. So ensembles are a way to compensate for poor
472C20Y4RY
models by performing a lot of extra computation.

3
Proprietary content. ©Great Learning.
This AllforRights
file is meant Reserved.
personal Unauthorizedonly.
use by [email protected] use or distribution prohibited
Sharing or publishing the contents in part or full is liable for legal action.
 Common Ensemble Techniques
• Bagging (Bootstrap Aggregation)

• Reduced chances of over fitting by training each model only with a randomly
chosen subset of the training data. Training can be done in parallel.

• Essentially trains a large number of “strong” learners in parallel (each model is

an over fit for that subset of the data)

• Combines (averaging or voting) these learners together to "smooth out"

predictions.

•
[email protected]
472C20Y4RY
Boosting

• Trains a large number of "weak" learners in sequence. A weak learner is a

simple model that is only slightly better than random (eg. One depth decision
tree).

• Miss-classified data weights are increased for training the next model. So
training has to be done in sequence.

• Boosting then combines all the weak learners into a single strong learner.

Bagging uses complex models and tries to "smooth out" their predictions, while
Boosting uses simple models and tries to "boost" their aggregate complexity.
4
Proprietary content. ©Great Learning.
This AllforRights
file is meant Reserved.
personal Unauthorizedonly.
use by [email protected] use or distribution prohibited
Sharing or publishing the contents in part or full is liable for legal action.
 Boosting Methods

• AdaBoosting (Adaptive Boosting)

• In AdaBoost, the successive learners are created with a focus on the

ill fitted data of the previous learner

• Each successive learner focuses more and more on the harder to fit
[email protected]
472C20Y4RY
data i.e. their residuals in the previous tree

• Gradient Boosting

• Each learner is fit on a modified version of original data (original data is

replaced with the x values and residuals from previous learner

• By fitting new models to the residuals, the overall learner gradually

improves in areas where residuals are initially high

5
Proprietary content. ©Great Learning.
This AllforRights
file is meant Reserved.
personal Unauthorizedonly.
use by [email protected] use or distribution prohibited
Sharing or publishing the contents in part or full is liable for legal action.
 + _ + _ + _
+ + +
+ _ + _ + _
+ _ _ + _ _ + _ _

[email protected]
472C20Y4RY

+ _
+
+ _
+ _ _

Proprietary content. ©Great Learning.

This AllforRights
file is meant Reserved.
personal Unauthorizedonly.
use by [email protected] use or distribution prohibited
Sharing or publishing the contents in part or full is liable for legal action.