milk 0.5.3 - Changelog

What's new in milk 0.5.3:

June 20th, 2013

· This version adds fixes compilation on a few older compilers and adds some new functionality on Multidimensional scaling.

What's new in milk 0.5.1:

January 18th, 2013

· Add subspace projection kNN
· Export pdist in milk namespace
· Add Eigen to source distribution
· Add measures.curves.roc
· Add mds_dists function
· Add verbose argument to milk.tests.run

What's new in milk 0.5:

November 6th, 2012

· Add coordinate-descent based LASSO
· Add unsupervised.center function
· Make zscore work with NaNs (by ignoring them)
· Propagate apply_many calls through transformers
· Much faster SVM classification with means a much faster defaultlearner() [measured 2.5x speedup on yeast dataset!]

What's new in milk 0.4.3:

September 21st, 2012

· Add select_n_best & rank_corr to featureselection
· Add Euclidean MDS
· Add tree multi-class strategy
· Fix adaboost with boolean weak learners (issue #6, reported by audy (Austin Richardson))
· Add axis arguments to zscore()

What's new in milk 0.4.2:

January 17th, 2012

· Make defaultlearner able to take extra arguments
· Make ctransforms_model a supervised_model (adds apply_many)
· Add expanded argument to defaultlearner
· Fix corner case in SDA
· Fix repeated_kmeans
· Fix parallel gridminimise on Windows
· Add multi_label argument to normaliselabels
· Add multi_label argument to nfoldcrossvalidation.foldgenerator
· Do not fork a process in gridminimise if nprocs == 1 (makes for easier debugging, at the cost of slightly more complex code).
· Add milk.supervised.multi_label
· Fix ext.jugparallel when features is a Task
· Add milk.measures.bayesian_significance

What's new in milk 0.4.1:

August 26th, 2011

· Fix important bug in multi-process gridsearch

What's new in milk 0.4.0:

August 25th, 2011

· Use multiprocessing to take advantage of multi core machines (off by default).
· Add perceptron learner
· Set random seed in random forest learner
· Add warning to milk/__init__.py if import fails
· Add return value to gridminimise
· Set random seed in precluster_learner
· Implemented Error-Correcting Output Codes for reduction of multi-class to binary (including probability estimation)
· Add multi_strategy argument to defaultlearner()
· Make the dot kernel in svm much, much, faster
· Make sigmoidal fitting for SVM probability estimates faster
· Fix bug in randomforest (patch by Wei on milk-users mailing list)

What's new in milk 0.3.10:

May 11th, 2011

· Add ext.jugparallel for integration with jug
· parallel nfold crossvalidation using jug
· parallel multiple kmeans runs using jug
· cluster_agreement for non-ndarrays
· Add histogram & normali(z|s)e options to milk.kmeans.assign_centroid
· Fix bug in sda when features were constant for a class
· Add select_best_kmeans
· Added defaultlearner as a better name than defaultclassifier
· Add measures.curves.precision_recall
· Add unsupervised.parzen.parzen

What's new in milk 0.3.9:

March 16th, 2011

· Add folds argument to nfoldcrossvalidation
· Add assign_centroid function in milk.unsupervised.nfoldcrossvalidation
· Improve speed of k-nearest neighbour (10x on scikits-learn benchmark)
· Improve kmeans on newer numpy (works for larger datasets too)
· Faster kmeans by coding centroid recalculation in C++
· Fix gridminize for low count labels
· Fix bug with non-integer labels for tree learning

What's new in milk 0.3.6:

December 18th, 2010

· Unsupervised (1-class) kernel density modeling
· Fix for when SDA returns empty
· weights option to some learners
· stump learner
· Adaboost (result of above changes)