Day 41 - More course overview

A master list of algorithms, sorted by response/predictor type:

	Response	Predictor	Visualization	Sorting	Abstraction	Model
1num	n	-	histogram, boxplot	-	beyond,within	-
1cat	c	-	hierarchy	-	balance	-
contingency table	c	-	mosaic	correspondence analysis	predictive (homogeniety)	binomial,lift
1aov	n	1c	boxplots	by median	predictive 1. same variance 2. diff't variance	normal
regression tree	n	c	-	-	predictive	tree
classification tree	c	c	-	-	predictive	tree
response table	n	2c	profile	by effects	-	additive,bilinear,normal
linear regression	n	n	scatter	-	-	linear,additive,bilinear,normal
classification	c	n	cplot (colored scatterplot)	-	discriminative projection	logistic regression, nearest neighbor
multivariate	n	-	star, parallel	PCA	PCA projection, clustering	multiv.normal

Classification methods

Histogram method For classifying batches `L(pop) = sum_x n_x log p(x \| pop)` Anomaly detection: `L(pop) < t` Fast to train and test Special case of linear classifier Highly constrained	Tree Recursive partitioning Flexible but crude Fast Concise Unstable
Logistic regression Slow to train, fast to test Linear classifier Somewhat constrained Feature expansion weakens constraints	K-nearest neighbor Flexible Slow Not concise Stable Hurt by irrelevants

Last modified: Tue Dec 04 13:18:56 Eastern Standard Time 2001