Changeset 13:f98f0417ceb6 in orange-reliability


Ignore:
Timestamp:
07/11/12 20:58:41 (22 months ago)
Author:
Matija Polajnar <matija.polajnar@…>
Branch:
default
Message:

Documentation of reference method for classification.

Files:
2 edited

Legend:

Unmodified
Added
Removed
  • _reliability/__init__.py

    r12 r13  
    383383 
    384384class ReferenceExpectedError: 
    385  
     385    """ 
     386 
     387    :rtype: :class:`Orange.evaluation.reliability.ReferenceExpectedErrorClassifier` 
     388 
     389    Reference reliability estimation method for classification as used in Evaluating Reliability of Single 
     390    Classifications of Neural Networks, Darko Pevec, 2011. 
     391 
     392    :math:`O_{ref} = 2 (\hat y - \hat y ^2) = 2 \hat y (1-\hat y)` 
     393 
     394    where :math:`\hat y` is the estimated probability of the predicted class. 
     395 
     396    Note that for this method, in contrast with all others, a greater estimate means lower reliability (greater 
     397    expected error). 
     398 
     399    """ 
    386400    def __init__(self, name="reference"): 
    387401        self.name = name 
  • docs/rst/Orange.evaluation.reliability.rst

    r5 r13  
    1010########################################################## 
    1111 
    12 ************************************* 
    13 Reliability Estimation for Regression 
    14 ************************************* 
     12******************************************************** 
     13Reliability Estimation for Regression and Classification 
     14******************************************************** 
    1515 
    1616Reliability assessment statistically predicts reliability of single 
     
    4949=================== 
    5050 
    51 For regression, all the described measures can be used. Classification domains 
    52 are supported by the following methods: BAGV, LCV, CNK and DENS. 
     51For regression, all the described measures can be used, except for the :math:`O_{ref}`. Classification domains 
     52are supported by the following methods: BAGV, LCV, CNK and DENS, :math:`O_{ref}`. 
    5353 
    5454Sensitivity Analysis (SAvar and SAbias) 
     
    8787 
    8888.. autoclass:: ParzenWindowDensityBased 
     89 
     90Reference Estimate for Classification (:math:`O_{ref}`) 
     91------------------------------------------------------- 
     92 
     93.. autoclass:: ReferenceExpectedError 
    8994 
    9095Reliability estimation wrappers 
Note: See TracChangeset for help on using the changeset viewer.