clj-ml History

v0.4.0 - 2013-08-07

Supported algorithms:

• Filters

  • Discretization (supervised, unsupervised, PKI)
  • Nominal to binary (supervised, unsupervised)
  • Numeric to nominal
  • String to word vector
  • Attribute manipulation (reorder, add, remove range, remove percentage, etc.)
  • Resample (supervised, unsupervised)
  • Replace missing values with mean (numeric attributes) or mode (nominal attributes)

• Classifiers

  • k-Nearest neighbor
  • Decision trees: C4.5/J4.8, Boosted stump, Random forest, Rotation forest, M5P
  • Naive Bayes
  • Multilayer perceptrons
  • Support vector machines (grid-based training), SMO, Spegasos

• Regression

  • Linear
  • Logistic
  • Pace
  • Additive gradient boosting

• Clusterers

  • k-Means
  • Cobweb
  • Expectation-maximization

Older clj-ml History

v0.2.5 - 2013-01-04

• Upgraaded to clojure 1.4 and lein2.

v0.2.4 - 2012-07-09

• More Maven appeasement....

v0.2.3 - 2012-07-06

• Use new RELEASE version of Incanter now that its available.

v0.2.2 - 2012-03-27

• Maven repo changes.

v0.2.1 - 2012-03-26

• Weight information of an instance is attached as metadata when converting to a map.

v0.2.0 - 2012-02-06

• Upgraded to clojure 1.3.

v0.1.5 - 2012-01-23

• MongoDB support was removed (the original author has never communicated with me and I have no intention on maintaining that functionality)

New Features

• adds attribute-value-fn helper fn

Bug Fixes

• fixes bug in add-attribute filter, name can now be specified.

v0.1.4 - 2011-12-14

New Features

• new filter wrappers: remove-range, remove-percentage
• new helpers: split-dataset, do-split-dataset, take-dataset, randomize-dataset!, randomize-dataset, map-fast-vec

v0.1.3 - 2011-10-07

New Features

• added new attribute selection namespace

v0.1.2 - 2011-06-14

New Features

• added dataset-weights function

Bug Fixes

• Does not create new ClojureInstances with dataset-seq, just uses weka's built in enumerator.

v0.1.1 - 2011-06-14

New Features

• added FastRandomForests which are more space and time efficient that Weka's default implementation.

Bug Fixes

• More typehinting for filter creation.

v0.1.0 - 2011-04-11

New Features

• New filter wrappers added: RemoveUseless as :remove-useless, Add as :add-attribute
• New classifier wrappers added: PaceRegression, RandomForest, M5P Trees and boosted stumps (LogitBoost), AdditiveRegression, Gradient Boosted Decision Trees, RotationForest, SPegasos
• Increased the number of options that can be specified for classifiers.
• Adds :clj-streamable and :clj-batch filters which allow for custom functions to be provided for filtering the dataset.
• More idiomatic way and using/applying filters that allows threading.
• Attribute names can now be specified in the filters and other accessor functions instead of requiring the columns index.
• New utility functions: into-fast-vec, dataset-replace-attribute, dataset-class-values, dataset-nominal?, make-apply-filters, classifer-copy-and-train, keyword-name, headers-only, dataset-class-name, attribute-labels-as-strings, dataset-name
• Speed improvement in dataset-as-maps

Bug Fixes

• is-dataset? reports falses correctly now.
• A large ammount of type hinting was added to many areas where it was slowing down real use-cases.
• Loading and saving instances functions now work with streams (not just files or file names).
• nil values are allowed in datasets (represented as NaNs in weka)

v0.0.4 - 2010-10-28

New Features

• Upgraded to Clojure 1.2. (Ben Mabey)
• Upgraded deps and swithced to modular Incanter dependencies. (Ben Mabey)

v0.0.3 - 2010-02-28

• Initial release by Antonio Garrote.