zensols.model.classifier — com.zensols.ml/model 0.0.18

A utility library that wraps Weka library. This library works with zensols.model.weka do the following:

• Cross validate models
• Manage and sort results (i.e. cross validations)
• Train models
• Read/write ARFF files

This namspace uses the resource location system to configure the location of files and output analysis files. For more information about the configuration specifics see model-read-resource and [[analysis-dir]], which both use resource-path.

You probably don't want to use this library directly. Please look at zensols.model.eval-classifier and zensols.model.execute-classifier.

*arff-file* clj

File to read or write from for any operation regarding file system access to a/the ARFF file(s).

*best-result-criteria* clj

Key used to sort results by their most optimal performance statistic. Valid values are: :accuracy, :wprecision, :wrecall, :wfmeasure, :kappa, :rmse

*class-feature-meta* clj

The class feature metadata (see zensols.model.weka/create-attrib).

*classifier-class* clj

Class name for the classifier used. This defaults to J48.

*create-classifier-fn* clj

Function used to create a classifier. Takes as input a weka.core.Instances.

*cross-fold-count* clj

The default number of folds to use during cross fold validation (see [[cmpile-results]]).

*cross-val-fns* clj

If this is non-nil then two pass validation is used. This is a map with the following keys:

• :train-fn a function that is called during training for each fold to stitch in partial feature-sets to get better results; almost always set to zensols.model.eval-classifier/two-pass-train-instances

• :test-fn just like :train-fn but called during testing; almost always set to zensols.model.eval-classifier/two-pass-train-instances

See [[zensols.model.eval-classifier/two-pass-config]]

*get-data-fn* clj

A function that generates a weka.core.Instances for cross validation, training, etc.

*operation-write-instance-fns* clj

A map with valus of functions that are called that return a java.util.File for an operation represented by the respective key. An ARFF file is created at the file location. The keys are one of:

• :train-classifier called when the classifier is training a model
• :test-classifier called when the classifier is testing a model

*output-class-feature-meta* clj

Default attribute name for the predicted label.

*rand-fn* clj

A function that returns a java.util.Random used to randomize the train/test dataset.

analysis-report-resource clj

(analysis-report-resource)

Return the model directory on the file system as defined by the :analysis-report. See namespace documentation on how to configure.

classifier-name clj

(classifier-name classifier-instance)

Return a decent human readable name of a classifier instance.

classify-instance clj

(classify-instance classifier unlabeled return-keys)

Make predictions for all instances.

• classifier instance of weka.classifiers.Classifier
• unlabeled contains feature set data with an empty class label as a weka.core.Instances
• return-keys what data to return
  • :label the classified label
  • :distributions the probability distribution over the label

compile-results clj

(compile-results results)

Return an easier to use map of result data given from cross-validate-tests. The map returns all the performance statistics and:

• :feature-metadata feature metadatas
• :result weka.core.Evaluation instance
• all-results a sorted list of weka.core.Evaluation instances

See cross-validate-tests for where the results data is created.

cross-validate-tests clj

(cross-validate-tests classifier attributes feature-metadata)

Run the cross validation for classifier and attributes (symbol set).

excel-results clj

(excel-results sheet-name-results out-file)

Save the results in Excel format.

excel-results-precision clj

An integer specifying the length of the mantissa when creating the results spreadsheet in excel-results.

filter-attribute-data clj

(filter-attribute-data unfiltered attributes)

Create a filtered data set (weka.core.Instances) from unfiltered Instances. Paramater attributes is a set of string attribute names.

initialize clj

(initialize)

Initialize model resource locations.

This needs the system property clj.nlp.parse.model set to a directory that has the POS tagger model english-left3words-distsim.tagger(or whatever you configure in [[zensols.nlparse.stanford/create-context]]) in a directory called pos.

See the source documentation for more information.

model-exists? clj

(model-exists? name)

Return whether a the model exists with name.

See model-read-resource.

model-read-resource clj

(model-read-resource name)

Return a file pointing to model with name using the the :model-read resource path (see [[zensols.actioncli.resource/resource-path]]).

model-write-resource clj

(model-write-resource name)

Return a file pointing to model with name using the the :model-write resource path (see [[zensols.actioncli.resource/resource-path]]).

print-eval-results clj

(print-eval-results eval)

Print the results, confusion matrix and class details to standard out of a weka.core.Evalution.

print-results clj

(print-results results & {:keys [title]})

Print the results, confusion matrix and class details to standard out of a single or sequence of weka.core.Evalutions.

read-arff clj

(read-arff input-file)

Return a weka.core.Instances from an ARFF file.

read-model clj

(read-model name & {:keys [fail-if-not-exists?] :or {fail-if-not-exists? true}})

Get a saved model (classifier and attributes used). If name is a string, use model-read-resource to calculate the file name. Otherwise, it should be a file of where the model exists.

See model-read-resource.

Keys

• :fail-if-not-exists? if true then throw an exception if the model file is missing

test-classifier clj

(test-classifier classifier attributes train-data test-data)

Test/evaluate classifier (weka.classifiers.Classifier).

train-classifier clj

(train-classifier classifier attributes)

Train classifier (weka.classifiers.Classifier).

train-test-classifier clj

(train-test-classifier classifier
                       feature-meta-sets
                       feature-metadata
                       train-instances
                       test-instances)

write-arff clj

(write-arff instances)

Write a weka.core.Instances to an ARFF file and return that file.

write-model clj

(write-model name model)

Get a saved model (classifier and attributes used). If name is a string, use model-write-resource to calculate the file name. Otherwise, it should be a file of where to write the model.

See model-read-resource