Machine Learning Tasks

Supervised, Unsupervised, and Reinforcement Learning

David Dalpiaz

Machine Learning Tasks

Big Picture

  • Supervised Learning
    • Classification
    • Regression
  • Unsupervised Learning
    • Density Estimation
    • Clustering
    • Novelty and Outlier Detection
    • Dimension Reduction
  • Reinforcement Learning

Data Interlude: Palmer Penguins

Palmer Penguins: Background

Palmer Penguins: Tabular View

species bill_length_mm bill_depth_mm flipper_length_mm
306 Chinstrap 40.9 16.6 187.0
341 Chinstrap 49.6 18.2 193.0
291 Chinstrap 50.5 19.6 201.0
102 Adelie 37.7 16.0 183.0
289 Chinstrap 52.0 18.1 201.0
267 Gentoo 55.1 16.0 230.0
125 Adelie 40.6 19.0 199.0

Palmer Penguins: Graphical View

Supervised Learning

Classification

Given data, learn a function that will output (predict) a categorical response (target) variable given input feature variables.

With the penguins:

  • Input: Bill length and depth
  • Output: Penguin species

Question: How do we know if a learned function is doing a good job?

species bill_length_mm bill_depth_mm
306 Chinstrap 40.9 16.6
341 Chinstrap 49.6 18.2
291 Chinstrap 50.5 19.6
102 Adelie 37.7 16.0
289 Chinstrap 52.0 18.1
267 Gentoo 55.1 16.0
125 Adelie 40.6 19.0

Regression

Given data, learn a function that will output (predict) a numeric response variable (target) given input feature variables.

With the penguins:

  • Input: Bill length (and Species)
  • Output: Bill Depth

Question: How do we know if a learned function is doing a good job?

bill_depth_mm species bill_length_mm
306 16.6 Chinstrap 40.9
341 18.2 Chinstrap 49.6
291 19.6 Chinstrap 50.5
102 16.0 Adelie 37.7
289 18.1 Chinstrap 52.0
267 16.0 Gentoo 55.1
125 19.0 Adelie 40.6

Classification and Regression

Commonality: When learning (training) we will have access to a response or target variable.

Importantly, in both cases, we want to make predictions for unseen data!

How will we evaluate how well we predict data we haven’t seen?!

Unsupervised Learning

Clustering

Given data, learn potential categorizations (groupings) of the samples.

With the penguins:

  • Input: Bill length and depth
  • Output: Categories

Question: How do we know if a learned good categories?

Note: These categories have no meaning!

Reinforcement Learning

Reinforcement Learning

A computer program interacts with a dynamic environment in which it must perform a certain goal (such as driving a vehicle or playing a game against an opponent). As it navigates its problem space, the program is provided feedback that’s analogous to rewards, which it tries to maximize. (Bishop, C. M. (2006), Pattern Recognition and Machine Learning)

  • Reinforcement learning is outside the scope of CS 307.

But… How?

Also… Who Cares?