Machine Learning Top Trumps with David Tarrant

8/20/2019 Machine Learning Top Trumps with David Tarrant

1/22

Machine learning and classification


2/22

Explain the core aspects of Machine LearningApply classification to a set of data to create a decision

Write a machine learning algorithm

O


3/22

Machine Learning

“construction and study of systems that can learn from data”

Can be seen as building blocks to make computers learn to

behave more intelligently

There are various techniques with various implementations.

Slide ideas


4/22

Question?

What instances of machine learninghave you come across?

What types are they?


5/22

Use-Cases

•

Spam Email Detection•Machine Translation (Language Translation)

•Image Search (Similarity)

•Clustering (KMeans) : Amazon Recommendatio

•Classification : Google News


6/22

Use-Cases (contd.)

•Text Summarization - Google News

•Rating a Review/Comment: Yelp

• Fraud detection : Credit card Providers

•Decision Making : e.g. Bank/Insurance sector

• Sentiment Analysis

• Speech Understanding – iPhone with Siri

• Face Detection – Facebook’s Photo tagging


7/22

Not a Spam

Not a Spam

Not Spam


8/22

NER (Named Entity Recognition)

http://nlp.stanford.edu:8080/ner/process

http://nlp.stanford.edu:8080/ner/process


9/22

Similar/Duplicate ImagesRemember

Features ? (Feature Extraction)

Can be :

• Width

• Height

• Contrast

• Brightness

• Position

• Hue

• Colors

Credit: https://www.google.co.in/

Check this :

LIRE (Lucene Image REtrieval) library -

https://code.google.com/p/lire/

https://www.google.co.in/https://code.google.com/p/lire/https://www.google.co.in/


10/22

Recommendations

http://www.webdesignerdepot.com/2009/10/an-analysis-of-the-amazon-shopping-experience/

http://www.webdesignerdepot.com/2009/10/an-analysis-of-the-amazon-shopping-experience/http://www.webdesignerdepot.com/2009/10/an-analysis-of-the-amazon-shopping-experience/


11/22

TerminologyFeatures

The number of features or distinct traits that can be used to describe eachitem in a quantitative manner.

SamplesA sample is an item to process (e.g. classify). It can be a document, a picture, asound, a video, a row in database or CSV file, or whatever you can describe witha fixed set of quantitative traits.

Feature vectoris an n-dimensional vector of numerical features that represent some object.

Feature extractionPreparation of feature vector

Transforms the data in the high-dimensional space to a space offewer dimensions.

Training/Evolution setSet of data to discover potentially predictive relationships.

Slide ideas


12/22

Learning (Training)

Features:

1. Color:

Radish/Red

2. Type :

Features:

1. Sky

Blue

2. Logo

Features:

1. Yellow

2. Fruit

3. ShapeSlide ideas


13/22

Workflow

Slide ideas


14/22

Property locations?

Each table has a set of “Top Trump”training-set cards.

Build a decision tree to sort them into “New

York” and “San Francisco”.

You cannot use the target to sort them.


15/22

Categories

Supervised Learning

Unsupervised Learning

Semi-Supervised Learning

Reinforcement Learning

Slide ideas


16/22

Supervised Learning

The correct labels of the training data are known

Credit: http://us.hudson.com/legal/blog/postid/513/predictive-analytics-artificial-intelligence-science-fiction-e-discovery-truth

Slide ideas

http://us.hudson.com/legal/blog/postid/513/predictive-analytics-artificial-intelligence-science-fiction-e-discovery-truthhttp://us.hudson.com/legal/blog/postid/513/predictive-analytics-artificial-intelligence-science-fiction-e-discovery-truth


17/22

Unsupervised Learning

The correct labels of the training data are not knowndiscovery and pattern generation.


Slide ideas

http://us.hudson.com/legal/blog/postid/513/predictive-analytics-artificial-intelligence-science-fiction-e-discovery-truthhttp://us.hudson.com/legal/blog/postid/513/predictive-analytics-artificial-intelligence-science-fiction-e-discovery-truth


18/22

Semi-Supervised LearningA Mix of Supervised and Unsupervised learning.

Some labels are known, but majority are not.


http://us.hudson.com/legal/blog/postid/513/predictive-analytics-artificial-intelligence-science-fiction-e-discovery-truth


19/22

Reinforcement LearningSoftware agent learns based on feedback and/or reward.


http://us.hudson.com/legal/blog/postid/513/predictive-analytics-artificial-intelligence-science-fiction-e-discovery-truth


20/22

Techniques

classification

predict class from observations

clustering

group observations into “meaningful” groups

regression (prediction):

predict value from observations


21/22

r2d3.us


22/22

Frameworks/Tools

R

Weka

Carrot2

Gate

OpenNLP

LingPipe

Stanford NLP

Mallet – Topic Modelling

Gensim – Topic Modelling (Python) Apache Mahout

MLib – Apache Spark

scikit-learn - Python

LIBSVM : Support Vector Machines

and many more…

Documents

Machine Learning Top Trumps with David Tarrant