Linear Learner

orange.LinearLearner is a learner that uses the LIBLINEAR library backend that is very fast on large datasets.

LinearLearner

Linear learner learnes the attribute weights using one of the four possible methods.

Attributes

solver_type

Specifiys whitch method to use. Can be one of the folowing:

orange.LinearLearner.L2_LR (L2-regularized logistic regression, default)


	orange.LinearLearner.L2LOSS_SVM_DUAL
	orange.LinearLearner.L2LOSS_SVM
	orange.LinearLearner.L1LOSS_SVM_DUAL


	Note that only L2_LR supports probabilty esstimations.


  eps
  Stopping criteria (default 0.01)
  C
  Regularization parameter (default 1.0)




LinearClassifeir
Linear classifiers that uses one class vs. rest strategy for multi-class classification. It supports probability esstimation only if it was build with L2-regularized logistic regression learner.
 Attributes

  weights
  A list of computed weight vectors for all one class vs. rest classifiers


Examples
Part of linear-learner.py
data = orange.ExampleTable("iris")
classifier = orange.LinearLearner(data)

for i, cls_name in enumerate(data.domain.classVar.values):
    print "Attribute weights for %s vs. rest classification:\n\t" % cls_name,
    for attr, w in  zip(data.domain.attributes, classifier.weights[i]):
        print "%s: %.3f " % (attr.name, w),
    print


Produces the output:

Attribute weights for Iris-setosa vs. rest classification:
	sepal length: 0.463  sepal width: 1.464  petal length: -2.251  petal width: -1.025 
Attribute weights for Iris-versicolor vs. rest classification:
	sepal length: 0.566  sepal width: -1.482  petal length: 0.548  petal width: -1.415 
Attribute weights for Iris-virginica vs. rest classification:
	sepal length: -1.862  sepal width: -1.640  petal length: 2.474  petal width: 2.587 
...




References

R.-E. Fan, K.-W. Chang, C.-J. Hsieh, X.-R. Wang, and C.-J. Lin. LIBLINEAR: A Library for Large Linear Classification, Journal of Machine Learning Research 9(2008), 1871-1874. Software available at http://www.csie.ntu.edu.tw/~cjlin/liblinear