Introduction to Machine Learning

Monday, July 8th

Today

  • What is Machine Learning
  • Supervised Machine Learning
  • Linear Regressions (Revisited)
  • Logistic Regression

Warm-Up

https://shorturl.at/2KYoI

Introduction to Machine Learning

Introduction to Machine Learning

Introduction to Machine Learning

Machine learning is a subfield of computer science that is concerned with building algorithms which, to be useful, rely on a collection of examples of some phenomenon. These examples can come from nature, be handcrafted by humans or generated by another algorithm.

The Hundred-Page Machine Learning Book

Introduction to Machine Learning

An algorithm that builds a statistical model based on a dataset.

Introduction to Machine Learning

Types of Learning

  • Supervised Learning: The dataset is a collection of labeled examples.
  • Unsupervised Learning: The dataset is a collection of unlabeled examples.
  • Semi-Supervised Learning: The dataset contains both labeled and unlabeled examples.
  • Reinforcement Learning: The machine interacts with an environment to maximize rewards.
Introduction to Machine Learning

Supervised Learning

  • The goal is to produce a model that takes a feature vector (x) as input and outputs a label.
  • Most things we consider "Machine Learning" fall into this category.
Introduction to Machine Learning

Supervised Learning Examples

  • Credit Card Fraud Detection

    • Features: Vendor, location, time, distance from last transaction
    • Labels: Chargebacks on previous transactions

  • Gmail Inbox Categories

    • Features: Terms in the subject, body, origin email address
    • Label: "Primary", "Social", "Promotions", "Updates", "Forums"

  • Stock Market Prediction

    • Features: Stock price from Feb 1st to March 1st
    • Labels: Stock price on March 7th.
Introduction to Machine Learning
Introduction to Machine Learning

Unsupervised Learning

  • The goal is to create a model that transforms input vectors or generates values to solve practical problems.
Introduction to Machine Learning

Unsupervised Learning Examples

  • Grocery Store Customer Segmentation

    • Input: Credit card numbers, purchase history
    • Output: Clusters of shopping behaviors

  • TFIDF

    • Input: Term frequencies in documents
    • Output: Term-vectors representing documents

  • Word Embeddings (Word2Vec)

    • Input: A corpus of words used in a language.
    • Output: Embeddings that can be used in Supervised Learning.
Introduction to Machine Learning
Introduction to Machine Learning

Model-Based vs Instance-Based Learning

  • Clustering is the most common example of Unsupervised Machine Learning.

  • Clustering is an example of Instance-Based Learning.

  • Model-Based Learning involves creating a model that can be used to make predictions.

  • Instance-Based Learning involves storing examples and making predictions based on similarity to stored examples.

Introduction to Machine Learning

Semi-Supervised Learning

  • Combines labeled and unlabeled data to improve model performance.
  • Adding unlabeled examples can provide more information about the data distribution.
Introduction to Machine Learning

fit center

Introduction to Machine Learning

Reinforcement Learning

  • Involves a machine "living" in an environment, perceiving states, and executing actions to maximize rewards.
  • Example applications: Game playing, robotics, resource management.
Introduction to Machine Learning

fit center

Introduction to Machine Learning

Supervised Machine Learning

Introduction to Machine Learning

Classification

  • In classification, our label is something from a finite set of classess.
  • When only two classes (True/False), we call it Binary Classification.
  • When more than two, we call it Multiclass Classification.

Regression

  • In regression, our label is a numerical real value. We're often approximating some target.
Introduction to Machine Learning

Supervised Learning Examples

  • Credit Card Fraud Detection ← Binary Classification

    • Features: Vendor, location, time, distance from last transaction
    • Labels: Chargebacks on previous transactions

  • Gmail Inbox Categories: ← Multiclass Classification

    • Features: Terms in the subject, body, origin email address
    • Label: "Primary", "Social", "Promotions", "Updates", "Forums"

  • Stock Market Prediction: ← Regression

    • Features: Stock price from Feb 1st to March 1st
    • Labels: Stock price on March 7th.
Introduction to Machine Learning

How Supervised Learning Works

  1. Data Collection: Gather data with a clear set of inputs and outputs.
  2. Feature Extraction: Convert inputs into feature vectors.
  3. Model Building: Choose an algorithm, it's parameters, and train a model.
  4. Model Testing: Validate your model using a test set that's separate from your of your training data and evaluate its performance.
  5. Prediction: Apply the model to new data to make predictions.

TODO, we should do this long before ML

TODO: Add a graphic here