aboutsummaryrefslogtreecommitdiffstats
path: root/src/classifiers
Commit message (Collapse)AuthorAgeFilesLines
* Explicitly set in_class field while learning.Vsevolod Stakhov2013-12-011-0/+1
|
* Fix bayes algorithm.Vsevolod Stakhov2013-11-151-4/+4
|
* Parse classifiers and statfiles in ucl.Vsevolod Stakhov2013-11-072-2/+2
|
* Detect exp2l as well. Update to 0.5.6.Vsevolod Stakhov2013-06-101-0/+2
|
* Test for expl function presence.Vsevolod Stakhov2013-06-091-0/+4
|
* Try to avoid exp overflowing.Vsevolod Stakhov2013-05-271-3/+3
|
* Fix probability calculations.Vsevolod Stakhov2013-05-231-1/+1
|
* Fix call to get statfile's revision.Vsevolod Stakhov2013-05-231-3/+4
|
* Use revisions to get total learn numbers.Vsevolod Stakhov2013-05-231-2/+2
|
* Fix calculations.Vsevolod Stakhov2013-05-231-3/+8
|
* New chi2square based bayes normalizer.Vsevolod Stakhov2013-05-231-160/+110
|
* * Use murmur hash for all hashes as it is more efficient and provides more ↵Vsevolod Stakhov2012-10-081-9/+29
| | | | | | uniform distribution as glib's default one. * Fix probability renormalization while using advanced classification.
* * Add meta-classification example.Vsevolod Stakhov2012-10-041-4/+7
| | | | | Many changes to advanced statistic and meta-classification logic. Add example of complex meta-classification.
* Another debian license fix.Vsevolod Stakhov2012-09-103-3/+3
| | | | | Add apache license for regexp that were delivered from SpamAssassin project. Fix debian/copyright for src/dns.c.
* Update copyright (required by debian).Vsevolod Stakhov2012-09-043-9/+9
|
* * Rework thread pools locking logic to avoid global lua mutex usage.Vsevolod Stakhov2012-08-223-35/+15
| | | | | | Fixed several memory leaks with modern glib. Fixed memory leak in dkim code. Fixed a problem with static global variables in shared libraries.
* * More things to be thread-safe:Vsevolod Stakhov2012-02-132-0/+21
| | | | | | | - pool allocator is now thread-safe - lua subsystem now holds lock to avoid lua stack corruption - events subsystem now using conditional variables to wait for async_threads - insert_result is thread-safe now
* * Add configuration utils for kvstorageVsevolod Stakhov2011-10-171-3/+3
|
* Fix signness in arithmetic operations.Vsevolod Stakhov2011-08-042-5/+5
|
* * Fix build under CentOS 5 with old glib 2.12Vsevolod Stakhov2011-07-294-18/+15
| | | | | * Fix build of rspamd with CMAKE_BINARY_DIR differs from CMAKE_SOURCE_DIR Rework include style.
* * Add max_tokens options to avoid classifying and learning with too much ↵Vsevolod Stakhov2011-07-261-7/+47
| | | | | | tokens from one message. Fix stupid memory leakage on client's timeout.
* Write error about too few tokens during learning.Vsevolod Stakhov2011-07-221-0/+5
|
* Create statfiles using learn_spam function for bayes classifier.Vsevolod Stakhov2011-07-181-0/+18
| | | | Fix call of pre callbacks for a classifier.
* * Add learn_spam/learn_ham interface to librspamdclient and to rspamcVsevolod Stakhov2011-07-141-1/+1
| | | | | * Improve logic of io dispatcher restoration Remove correction factor from bayes as it leads to classify errors.
* * First commit to implement multi-statfile filter system with new learning ↵Vsevolod Stakhov2011-07-124-18/+103
| | | | mechanizm (untested yet)
* * Add correcting factor to statistics.Vsevolod Stakhov2011-06-281-15/+33
| | | | | | Now learning increments version of a statfile. Avoid learning and classifying of similar text parts if a message has 2 text parts. Several fixes to statistics.
* Fix statshow utility.Vsevolod Stakhov2011-06-271-2/+2
|
* * Welcome 0.4.0Vsevolod Stakhov2011-06-241-16/+15
| | | | | | | | | | | | | | | | | | Uncompatible changes: - Statistics is uncompatible in utf8 mode Major changes: - Improved utf8 mode - Convert all characters to lowercase in statistics - Skip URL's in statistics - Improve speed of bayes classifier by using integer arithmetics - Fixed statfiles synchronization that was broken for a long time - Synchronization is now configurable Minor changes: - Bugfixes - Removed some of legacy code - Types polishing
* * Major cleanup of cmake build systemVsevolod Stakhov2011-05-062-1/+7
| | | | | | * Add initial version of statshow utility for statfiles debugging * Add debugging for statistics * Remove unused utilities
* Fixes in classifying for small messages.Vsevolod Stakhov2011-01-252-6/+24
|
* Implement checking options for modulesVsevolod Stakhov2010-12-131-0/+1
| | | | | | Implement checking for classifier options Fix redirector to handle timeouts and invalid replies properly Fix surbl module not to check each url
* Remove G_INLINE_FUNC definitions as I misunderstood its purposes.Vsevolod Stakhov2010-10-152-2/+2
|
* Fixes bugs found with clang-static analyser.Vsevolod Stakhov2010-10-112-2/+14
| | | | | Strictly follow c99 standart. Turn on pedantic c99 checks.
* * Fix races in fuzzy storageVsevolod Stakhov2010-08-261-1/+3
|
* * Fix normalization for systems that have not tanhl functionVsevolod Stakhov2010-08-181-2/+2
|
* * Remove normalizer as it is winnow specific thing, so all statistic ↵Vsevolod Stakhov2010-08-131-4/+18
| | | | algorithms now returns value from 0 to 1
* * Add bayesian classifier (initial version)Vsevolod Stakhov2010-08-134-11/+409
|
* * One more try to improve accuracy of winnow algorithmVsevolod Stakhov2010-08-061-10/+19
|
* * Fixes to winnow learningVsevolod Stakhov2010-08-052-47/+114
|
* * Fix some logic errors in learningVsevolod Stakhov2010-08-031-6/+8
|
* * Improve logic of learning messages: do not learn more than specific thresholdVsevolod Stakhov2010-08-021-8/+82
| | | | * Fix inserting results for symbols that were incorrectly (for example more than 1 time) defined in config file
* * Change metric logicVsevolod Stakhov2010-06-161-1/+1
| | | | | | | * Completely remove lex/yacc readers for config * Make common sense of metric/action and symbols * Sync changes with all plugins TODO: add this to documentation
* * Fix strict aliasing while compiling with optimizationVsevolod Stakhov2010-05-311-1/+1
| | | | | * Fix tanhl detection for platforms that have not implementation of it * Remove several compile warnings
* * Fix orderVsevolod Stakhov2010-05-271-3/+3
|
* * In classify normalize result after comparing, not beforeVsevolod Stakhov2010-05-271-3/+3
|
* * Convert statistic sums to use long double for countersVsevolod Stakhov2010-05-272-24/+18
| | | | * Use hyperbolic tangent for internal normalizer
* * Implement new learning system, now rspamd should be much more intelligent ↵Vsevolod Stakhov2010-05-271-32/+125
| | | | while learning messages
* * Fix awfull bug in classifying when first statfile has twice weight than ↵Vsevolod Stakhov2010-05-141-2/+2
| | | | | | second... * Fix undisclosed recipients detection
* * Bugfixes:Vsevolod Stakhov2010-04-201-1/+1
| | | | | | | | | | - handle '\' characters in lua strings correctly - fix lua initialization - avoid of using global lua state (global L) - fix listen sockets hash to allow multiply workers of same type but on different listen sockets - fix modules options inserting to allow multiply options of the same name - fix parsing of lua options - fix lua rules
* * Add option min_tokens to classifier that allows to skip too short messages ↵Vsevolod Stakhov2010-03-221-1/+34
| | | | | | from statistic check, format: min_tokens = "10"; (for 10 words minimum)
generated by cgit v1.2.3 (git 2.39.1) at 2025-04-04 20:39:35 +0000