Neural Networks with Excel

Notes on “A comparison of logistic regression and naive Bayes”
Download FINAL Dataset

Andrew Y. Ng and Michael I. Jordan had a classic paper on the comparison between logistic regression and naive Bayes. The main contribution of the paper is a theoretical analysis of how logistic regression and naive Bayes might perform and an experimental comparison to support this idea.

Several important points made by the paper. The first is that:

  • The asymptotic error made by logistic regression is no more than that made by naive Bayes (Proposition 1 in the paper).

This conclusion provides a basis for what seems to be the widely held belief that discriminative classifiers are better than generative ones. The main conclusions of the paper is about the sample complexity of both classifiers. Sample complexity is the number of examples needed to approach the asymptotic error. For logistic regression, such sample complexity is:
\[
m = \Omega(n)
\]which means that the sample complexity is linear in \( n \) (Proposition 2 in the paper). For naive Bayes, we have:
\[
m = O(\log n)
\]which means that the sample complexity is logarithmic in \( n \) (Lemma 3 and Corollary 6 in the paper). All these conclusions imply that even though naive Bayes converges to a higher asymptotic error compared to logistic regression, it may also approach it significantly faster — after \( O(\log n) \), rather than \( O(n) \), training examples.

Source: Hong, LiangJie

You might also like:

(ML 8.3) Bayesian Naive Bayes (part 1)
(ML 8.3) Bayesian Naive Bayes (part 1)
Missing Values using RapidMiner
Missing Values using RapidMiner
RapidMiner Tutorial (part 7/9) Naïve Bayes Classification
RapidMiner Tutorial (part 7/9) Naïve Bayes Classification

Short answer

2001-06-13 12:10:13 by datamouse

An article in Rolling Stone (October 20, 1994) by Adam Miller called J. Philippe Rushton a 'professor of hate,' someone who 'takes money from an organization with a terrible past' (the Pioneer Fund, a foundation said to have an orientation toward eugenics). He is accused of being 'obsessed with intelligence and genetics' to the point of having 'racist' attitudes by Jeffrey Rosen and Charles Lane in The New Republic symposium on IQ (October 31, 1994). They single out Rushton for linking ethnocentricism to genetic factors; this in turn subjects him to the broad brush of being, along with Richard J. Herrnstein and Charles Murray, 'Neo-Nazis' Newsweek (October 24, 1994). In The Chronicle of Higher Education (October 26, 1994) critiquing Herrnstein and Murray's The Bell Curve it is clear...

Palisade's NeuralTools replaces dangerous contrast dye for brain tumour scans  — RealWire
As a sophisticated neural networks tool that works directly in Microsoft Excel, NeuralTools could create accurate new predictions based on patterns in known data that were easily accessible, and simple to read. Historical magnetic values of tissue ...

Two Rutgers Seniors Awarded Gates Scholarships for Graduate Study at …  — News from Rutgers
Her research there involved testing drugs that promote the division of neural embryonic stem cells. Four other Rutgers students have since ... Lambert credits Rutgers for providing options and opportunities to excel in the sciences. “I would not have ...

(ML 8.1) Naive Bayes classification
(ML 8.1) Naive Bayes classification

Related posts:

  1. Neural Networks with Weka
  2. Neural Networks in the Cockpit
  3. Neural Networks bias node
  4. Neural Networks in Perl
  5. Neural Networks in Excel