[CADIHash.h Index] [CADIHash.h Hierarchy]
a universally usable class for weighting More...
#include <../libInvertedFile/include/CWeightingFunction.h>
a universally usable class for weighting. The rationale is, that we want to implement very flexible feature weighting, so flexible that it does not necessarily have anything to do with the initial concept of feature weighting.
The scenario: We have query/feedback from the user. That is: We have a list of images with either positive or negative relevance for the user. Each of these images contains a list of features.
Now we want to combine the different features. What we usually need to know is: how many images where in the positive feedback, and how many images where in the negative feedback.
The query factor contains all informatio which depends only on the query. (for efficiency)
Calculate the QueryFactor.
Constructor:
inQueryNormalizer: | FIXME |
inThisNormalizer: | FIXME |
inAccessor: | the accessor containing this \ (for getting information about single documents) |
Set the accessor (should not be needed)
Set the query normalizer (should not be needed);
Set the ID of this feature
get the ID of this feature
Setting the relevance sum, that is the sum of the absolute values of all relevance levels of all documents of the query
for rocchio we need this sum to be once calculated for positive, once for negative feedback images
Add a feature. A feature will be a document frequency element out of the feature file.
Giving a pseudo term frequency for normalization
Apply this on a document frequency
Preparing the right parameters for subApply: Apply this on another feature
Preparing the right parameters for subApply: Apply this on itself
For the "prototype pattern": Kinda virtual copy constructor:
For the "prototype pattern": Kinda virtual copy constructor: Gamma/Helm/Johnson/Vlissides p. 107
This is a new version of constructNew which manages to live without the setID function
This value depends on all the elements of the query, whichhave a positive user-assigned relevance.
i.e. not necessarily only on the elements which contain this feature, so it has to be explicitly set using the function setRelevanceSum from the outside.
If relevance=+-1 then this is the number of positive images in the query
This value depends on all the elements of the query, whichhave a negative user-assigned relevance.
i.e. not necessarily only on the elements which contain this feature, so it has to be explicitly set using the function setRelevanceSum from the outside.
If relevance=+-1 then this is the number of negative images in the query
At present the feature description is a simple integer which expresses if the feature is a histogram feature or not and if it is a texture feature or not. We probably will use more elaborate descriptions in the future.
The accessor for the file on which the query is performed
FIXME THERE IS A GOOD DESCRIPTION MISSING FOR THE NEXT
A normalizer for apply on self
The query usually will separate into a part which is equal for all the query and a part which is to be calculated for each document. The part which stays constant
| Kdoc |