Introduction to Machine Learning

Lets Get Rolling - Student Pre-Read

Before this lesson , we recommend you go through

• Read through the scikit-learn quick-start.
• Take a peak at the Machine Learning Cheat Sheet (for scikit-learn)

Learning Objectives

After this lesson, you'll be able to

• Understand broad categories of Machine Learning Algorithms
• Understand the Machine Learning Workflow
• Work with Data

Agenda

• Merits of 3 Pass System

• What is Machine Learning?

• History of Machine Learning

• Types of Machine Learning

• The Machine Learning Workflow

• Scientific Evolution 01

  • Difference between Principal & Law
  • Newton came up with a Principal
  • J. D. Murray
  • Statisticians
  • Modern Data Science

• Management Evolution 02

  • Henry Ford - Personal Charisma - Key Lieutenants
  • Toyota - Process - 6 Sigma
  • Michael Porter - 5 Forces - Consulting Practices
  • Evolution of Decision Support Systems - Investment Banking
  • Thomas Davenport - Analytics 3.0
  • Tactical Support - Mass Personalization

• Algorithmic Evolution

  • Originally prosposed in 1963 - SVM - Vapnik
  • Kernel Mathematically implemented in 1995
  • SVM first won on Kaggle in 2013
  • History of Machine Learning

• Overview of key Algorithms

  Deep Learning & Neural Nets

  • 1.1 No. of Layers
  • 1.2 No. of Nodes
  • 1.3 Weights
  • 1.4 Activation Function
  • 1.5 Learning Rate - Algorithms to vary Learning Rates
  • 1.6 Other Hyper Parameters

  Ordinary Least Square

  • 2.1 Regularization
  • 2.2 Shrinkage
  • 2.3 Kernels

  Decision Tree - Entropy based + Marketing Example

  • 3.1 Tree Depth
  • 3.2 Prune
  • 3.3 Stopping Criteria

  Maximum Likelihood - Video1 + Video2 + Video3

  • 4.1 This is a Base Algorithm

  PCA

  • 5.1 No. of Components
  • 5.2 Learning Rate
  • 5.3 Random Seed

  Stochastic Gradient Descent

  • 6.1 Learning rate
  • 6.2 Type of Loss Function
  • 6.3 Penalty

  Clustering - Centroid Based

  • 7.1 No. of clusters
  • 7.2 Algorithm -
  • 7.3 N Jobs - Random Seed

  Ensembles

  • 8.1 Everything in Oridinary Least Square/ Decision Tree and Stochastic Gradient Descent more

• Thinking about Data

• Where does Data Come from?

• How to Access Data?

  • Data from the Web 1 (Web Scraping)
  • Data from the Web 2 (APIs)

Slides

@gslides

Instructors code alongs

• Thinking about Data
• Introduction to Machine Learning

Assignments

Complete this setup before attempting the assignments

& many more inside commit.live.

Resources & Post Reads

• Check out the intro to scikit-learn video series from SciPy2013.

• Learn more about sklearn by reading API design for machine learning software: experiences from the scikit-learn project.

• Check out Datalicious Notebookmania – My favorite 7 IPython Notebooks

• Find out more about Nicholas Nassim Taleb

• He has written following landmark books:

  • Fooled by Randomness
  • The Black Swan
  • Anti Fragile
  • Feedly
  • Curate your personal feed for ML and DS

• Feedly

  • Algorithmia
  • Algorithms for the Masses - Julian m Bucknall
  • Algorithms Weekly by Petr Mitrichev
  • Artificial Intelligence News -- ScienceDaily
  • Daniel Lemire's blog
  • Data Science 101
  • DataTau
  • Everyone's Blog Posts
  • FastML
  • Geeking with Greg
  • GeeksforGeeks
  • Journal of Machine Learning Research
  • KDnuggets
  • Machine Learning
  • Machine Learning (Theory)
  • Machine Learning Mastery
  • Neural Networks
  • Newest questions tagged scikit-learn - Stack Overflow
  • No Free Hunch
  • Nuit Blanche
  • Simply Statistics
  • Statistical Modeling, Causal Inference, and Social Science
  • mathbabe

• Curate your personal feed for ML and DS

• Add the relevant channels on Youtube

  • Abu Yaser Mostafa

  • Victor Lavrenko

  • Nat & Friends

  • Hugo Larochelle

  • Ben Lambert

  • Mathematical Monk

  • Answer Questions:

• Can search based algorithms be applied for Linear Regression?

• Create an Excel sheet

  • Can you build a regression model in MS Excel? Solution
  • Can you build a regression model in MS Excel using closed form equations?
  • Can you build a regression model in MS Excel using SGD?