Changes for MDL-Extension
Submit rating
MDL-Extension
This is a project for the implementation of MDL (Minimum Description Length) based extensions. The MDL principle can be applied to get a shorter description of a dataset, using its regularities in reference to a suitable compression. So, the best description is seen as the one that compresses the dataset best. The MDL extension currently includes an operator for the implementation of KRIMP algorithm, which might be used to prune a set of frequent patterns.
This is a project for the implementation of MDL (Minimum
Description Length) based extensions. The MDL principle can be
applied to get a shorter description of a dataset, using its
regularities in reference to a suitable compression. So, the best
description is seen as the one that compresses the dataset best.
The MDL extension currently includes an operator for the
implementation of KRIMP algorithm, which might be used to prune a
set of frequent patterns. In particular, gaining a descriptive
selection of frequent patterns is considerably due to the high
redundancy and tremendous amount of patterns, that are generated by
frequent pattern mining algorithms with loosened constraints (e.g.
a lower support threshold). However, straightening these
constraints results mostly in an output with few patterns that
reveal nothing but common knowledge. KRIMP, or MDL based approaches
in general, can be applied here to get a reasoning and
understandable output of frequent patterns from a database. (For
further informations visit patternsthatmatter.org.)
Additionally, algorithms based on the MDL principle are naturally guarded from overfitting because the length of the computed model is taken into account as much as the length of the encoded dataset when the overall size of the compressed dataset is computed.
Additionally, algorithms based on the MDL principle are naturally guarded from overfitting because the length of the computed model is taken into account as much as the length of the encoded dataset when the overall size of the compressed dataset is computed.
Product Details
Version | 5.0.0 |
File size | 167 kB |
Downloads | 7301 (2 Today) |
Vendor | AI Group - TU Dortmund |
Category | Machine Learning |
Released | 9/3/13 |
Last Update | 9/3/13 2:44 PM |
License | AGPL |
Product web site | http://www-ai.cs.uni-dortmund.de/auto?self=$dttc2lvzsw |
Rating | (0)
|
Comments
Log in to post comments.