-
Section 6.1.1 - Equi - vs - Freq Diagnosis
Equi does
consistently better than Freq. Here are a list of
sample leaves and figures of, index of the feature
where sync has been placed sync - vs - probability
of that feature.
Leaf ID : 67576 - Equi allocation did better
Equi inner allocation: |
Freq inner allocation: |
 |
 |
Leaf ID : 752280 - Equi allocation did better
Equi inner allocation: |
Freq inner allocation: |
 |
 |
Leaf ID : 1146201 - Equi allocation did better
Equi inner allocation: |
Freq inner allocation: |
 |
 |
Leaf ID : 374897 - Freq allocation did better
Equi inner allocation: |
Freq inner allocation: |
 |
 |
Leaf ID : 593508 - Freq allocation did better
Equi inner allocation: |
Freq inner allocation: |
 |
 |
Back to the top
-
Section 6.1.4 - Feature re-numbering diagnosis
Global FeatHits (in the below images represented as Freq) ordering
is in excellent agreement with some of the costliest leaves.
Below are some of the plots of the costliest leaves. Each plot
corresponds to the index of a feature - vs - its probability of
occurence in the workload.
Leaf ID : 16312 - Feature probabilities before and after permutation
Default ordering: |
FeatHit ordering: |
 |
 |
Leaf ID : 16561 - Feature probabilities before and after permutation
Default ordering: |
FeatHit ordering: |
 |
 |
Leaf ID : 16873 - Feature probabilities before and after permutation
Default ordering: |
FeatHit ordering: |
 |
 |
Back to the top
-
Section 6.1.5 - Real time speed - vs - Estimated cost
The following graph proves the excellent agreement between our
estimated cost and the actual time taken to annotate.
Back to the top
-
Section 6.1.6 - Sample document list
Here
is the list of documents used for comparison with
other systems.
(1.6 MB)
- Number of documents: 1001
- File name: www-testdata.tar.gz
- Hyperlink:
www-testdata.tar.gz
- md5sum: dde033291aa31ff076dc93d3b215a340
Back to the top