- removed contest mode from LearningAlgorithmsEvaluation
Merge branch 'master' of https://git.code.sf.net/p/mlpc/code
fixed swapped music<->speech parameters for RF
Merge branch 'master' of https://git.code.sf.net/p/mlpc/code
tool to export
rename to export_predictions
training folds
ore contest mode
more flexibility
added NL
print result for checking
use contest data
parameters set to contest mode
- fixed small errors in ensemble and optimized the code
Merge branch 'master' of https://git.code.sf.net/p/mlpc/code
added parametrisation
Merge branch 'master' of ssh://git.code.sf.net/p/mlpc/code
ensemble now always combines prediction levels and assumes e.g. (0,1,0) probabilities for speech&music
- LearningAlgorithmsEvaluation now supports systematic testing of all folds (13 + 1 rotating)
- binary_to_ternary_classification now supports systematic testing of all folds (13 + 1 rotating)
Merge branch 'master' of https://git.code.sf.net/p/mlpc/code
print results read to check parameters etc.
same tick setup on both charts
made more user friendly and nicer output
Fixed check for available labels
improved comments/docu
restored LearningAlgorithmsEvaluation which was previously overwritten by an old version of itself on during a commit
Merge branch 'master' of https://git.code.sf.net/p/mlpc/code
re-added Sravanis class prediction functionality
added smoothing status to chart title
add smoothing
make easily callable
first untrustworthy version
safer arg count handling
Merge branch 'master' of ssh://git.code.sf.net/p/mlpc/code
highlight improvement over no-factors
explore shifting time
factored label retrieval into separate class
aggregate labels, too
combine probabilities into joint probability
added unit
more accurate title
make command-line executable
added parameter want_data_too to use less memory
not a good db normalization, but it is practical to store "true" labels within the results
Merge branch 'master' of ssh://git.code.sf.net/p/mlpc/code
not a good db normalization, but it is practical to store "true" labels within the results
Class Thresholds are implemented
setting random folds to true again, to avoid accidental troubles due to mis-runs
- fixed some issues with directory names and lengths when using ensemble
- ensemble can now be executed to merge results from disk
- predict_levels will now be stored for all results, not just combi_results (allows for combining multiple results into a combi_result and still having predict_levels there)
Merge branch 'master' of https://git.code.sf.net/p/mlpc/code
add unit
- Small fixes and cleanups
Merge branch 'master' of https://git.code.sf.net/p/mlpc/code
adapt to changes in combi_result and qualitative x-axis
Merge branch 'master' of ssh://git.code.sf.net/p/mlpc/code
- Fixed lin_chart to correctly calculate the criterions again
Merge branch 'master' of https://git.code.sf.net/p/mlpc/code
more common syntax
adapt to change in CombiResult, adapt to affine class_weight coordinates
- Some bugfixes
Merge branch 'master' of https://git.code.sf.net/p/mlpc/code
linear charts of average earning rates
specify zero class weight in subdir, even folds only for training.
factored common constants into their own file
Some fixes for combining multiple results
Merge branch 'master' of https://git.code.sf.net/p/mlpc/code
added loading of contest folds (15…29)
added store method
display class 0 weight from data
bug-fix to deal with attribute name change
- Added first version of combination evaluation (one set of algorithms examines music, the other one speech, then they are combined)
fix break by abstract result
- Because internal fold names are now numbers instead of strings, we lost the information whether a result is for speech or music (if it is not both). Therefore, such a field was introduced into the result classes
Inserted our best parameter values instead of the default(empty) ones
Small fixes
- Unified many things (variable names, output, structure) between "old" binary classification and new ternary
fix exception when array not complete; code cosmetics
added earning rate to earnings statistics
mode x
initial baseline exploration with gain matrix
fix yticks
more experimentation
Merge branch 'master' of ssh://git.code.sf.net/p/mlpc/code
fixed small error during previous merge
- Adapted fold descriptions to numbers instead of strings (as requested by Bernhard)
parameters replace paths, create a colormap figure
Merge branch 'master' of ssh://git.code.sf.net/p/mlpc/code
added inner parameter testing loop
experiment with RandomForest
correct result column headers
also show parameters
added earning_rate(gain_matrix)
initial working tool
experiment with class_weight
initial ranking tool
store algorithm parameters instead of gain matrix
store all pickles of one timestamp into one file