|
|
Guests are welcome to view our materials. To subscribe, edit, view raw markup, etc., you'll need to register for an account. Accounts are free (and will always be free) - your involvement helps us directly and indirectly (by demonstrating that our work matters to our funders...) StartingPoints has more info.
Pattern Recognition
Planning a project
Methods used for classification are fundamentally different than those used for signal representation, or at least they should be (although many practitioners still (mis)apply signal representation techniques to classification. Why is this such an important distinction? Because the difference in task fundamentally affects what aspects of the data are useful.
-
- We can capture our intuitive interpretation of what we see in the above image by considering the difference between variance and scatter. Variance is a measure of how tightly data clusters around it's own mean, while scatter is a measure of how tightly data clusters around the other class's mean.
- There are many other examples given in the papers listed in PatternRecognitionHighLevelViews
Exploratory Data Analysis
The point of the previous was to help you hang on to why, even though in the first round of exploratory data analysis we often work with unlabeled data, we still limit ourselves to classification oriented techniques.
I cannot emphasize this point enough. It's not ok to fall back on principle components analysis (PCA), etc., which are based solely on variance, in the first step, when planning on doing classification.
Why do exploratory data analysis? Why treat data as unlabeled, or use as few assumptions as possible in this stage? The answer is that you want to avoid assuming your results. Humans are hard-wired to hyper-detect patterns (all prey animals are.)
The goals of exploratory data analysis are two-fold (ideally):
- to get an estimate of the potential for classification of the current feature set (another way to put this is "how much information is there in those features, really?)
- to get an estimate of the complexity of the classification problem you are facing (another way to put this is "how big a hammer do I need?")
SelfOrganizingMaps are an excellent first pass at providing some information about both of those questions.
Topics elsewhere in this wiki that discuss Pattern Recognition
Found 2 topics
| FreeOpenSrc Web |
Changed |
By |
|
RisCool
|
27 May 2009 - 18:47
|
HilaryHolz
|
| R is Cool R is an open source "free software environment for statistical computing and graphics" (the FOSS answer to packages such as S, hence the name) which runs ... |
Found 1 topics
Found 2 topics
| Mutualdiscdev Web |
Changed |
By |
Found 2 topics
| Mutualdiscovery Web |
Changed |
By |
|
DataSets
|
10 Feb 2009 - 15:19
|
HilaryHolz
|
| Data Sets for use in PatternRecognitionWinter09 The R datasets package is part of the core, but can be hard for a beginner to find. The Irvine Machine Learning ... |
|
SomLabSpr09
|
29 Apr 2009 - 18:21
|
HilaryHolz
|
| Exploring Exploratory Data Analysis with Self Organizing Maps Introduction The SOM lab serves several roles at once. It's our exploratory data analysis lab, as well ... |
|
SprLab
|
08 Apr 2009 - 18:15
|
HilaryHolz
|
| Statistical Pattern Recognition Steps to get Started on R 1 Sandbox.IntroSPRbyKavi Overview of Pattern Recognition 1 People.SPRMaterialsKavi ... |
|
SvMLabSpr09
|
01 Jun 2009 - 16:49
|
HilaryHolz
|
| Support Vector Machine Lab (Supervised learning) Introduction ( You are in a maze of twisty little passages ... ) or, in this case, we start out with a seemingly ... |
|
SvmLab
|
15 Jun 2009 - 11:15
|
SowjanyaBandaru
|
| Intro Kernel based techniques (such as support vector machines, Bayes point machines, kernel principal component analysis, and Gaussian processes) represent a major ... |
Found 56 topics
|
NateDelgado
|
14 Oct 2009 - 18:32
|
HilaryHolz
|
| Nate's Corner Located in the Greater Twiki Universe CSUEB classes Pattern Recognition NotesOnExpertPythonProgramming WebSoftwareJournals WumpusU Assignments ... |
|
ThuonChen
|
22 Jan 2009 - 12:05
|
ThuonChen
|
| 22JAN2009THU Costco San Leandro Gas Station hours: M F 6am to 9:30pm Sat Sun 6am 7pm 20JAN2009TUE I am very excited about StatisticalPatternRecognition because ... |
Found 34 topics
Found 10 topics
-- HilaryHolz - 21 Apr 2009
|
|
Copyright 1999-2009 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.