remove unnecessary text

Author: escherba

clustering_metrics/metrics.py

@@ -1,92 +1,6 @@
 # -*- coding: utf-8 -*-
 
 """
-
-Motivation
-----------
-
-The motivation behind the given re-implementation of some clustering metrics is
-to avoid the high memory usage of equivalent methods in Scikit-Learn.  Using
-sparse dictionary maps avoids storing co-incidence matrices in memory leading to
-more acceptable performance in multiprocessing environment or on very large data
-sets.
-
-A side goal was to investigate different association metrics with the aim of
-applying them to evaluation of clusterings in semi-supervised learning and
-feature selection in supervised learning.
-
-Finally, I was interested in the applicability of different association metrics
-to different types of experimental design. At present, there seems to be both
-(1) a lot of confusion about the appropriateness of different metrics, and (2)
-relatively little attention paid to the type of experimental design used. I
-believe that, at least partially, (1) stems from (2), and that different types
-of experiments call for different categories of metrics.
-
-Contingency Tables and Experimental Design
-------------------------------------------
-
-Consider studies that deal with two variables whose respective realizations can
-be represented as rows and columns in a table.  Roughly adhering to the
-terminology proposed in [1]_, we distinguish four types of experimental design
-all involving contingency tables.
-
-========= ===================================
-Model O   all margins and totals are variable
-Model I   only the grand total is fixed
-Model II  one margin (either row or column totals) is fixed
-Model III both margins are fixed
-========= ===================================
-
-Model O is rarely employed in practice because researchers almost always have
-some rough total number of samples in mind that they would like to measure
-before they begin the actual measuring. However, Model O situation might occur
-when the grand total is not up to researchers to fix, and so they are forced to
-treat it as a random variable. An example of this would be astronomy research
-that tests a hypothesis about a generalizable property such as dark matter
-content by looking at all galaxies in the Local Group, and the researchers
-obviously don't get to choose ahead of time how many galaxies there are near
-ours.
-
-Model I and Model II studies are the most common and usually the most confusion
-arises from mistaking one for the other. In psychology, interrater agreement is
-an example of Model I approach. A replication study, if performed by the
-original author, is a Model I study, but if performed by another group of
-researchers, becomes a Model II study.
-
-Fisher's classic example of tea tasting is an example of a Model III study [2]_.
-The key difference from a Model II study here is that the subject was asked to
-call four cups as prepared by one method and four by the other. The subject was
-not free to say, for example, that none of the cups were prepared by adding milk
-first. The hypergeometric distribution used in the subsequent Fisher's exact
-test shares the assumption of the experiment that both row and column counts are
-fixed.
-
-Choosing an Association Metric
-------------------------------
-
-Given the types of experimental design listed above, some metrics seem to be
-more appropriate than others. For example, two-way correlation coefficients
-appear to be inappropriate for Model II studies where their respective regression
-components seem more suited to judging association.
-
-Additionally, if there is implied causality relationship, one-sided measures
-might be preferred. For example, when performing feature selection, it seems
-logical to measure the influence of features on the class label, not the other
-way around.
-
-Using Monte Carlo methods, it should be possible to test the validity of the
-above two propositions as well as to visualize the effect of the assumptions
-made.
-
-References
-----------
-
-.. [1] `Sokal, R. R., & Rohlf, F. J. (2012). Biometry (4th edn). pp. 742-744.
-       <http://www.amazon.com/dp/0716786044>`_
-
-.. [2] `Wikipedia entry on Fisher's "Lady Tasting Tea" experiment
-       <https://en.wikipedia.org/wiki/Lady_tasting_tea>`_
-
 """
 
 import warnings