llama (version 0.7.2)

classifyPairs: Classification model for pairs of algorithms

Description

Build a classification model for each pair of algorithms that predicts which one is better based on the features of the problem. Predictions are aggregated to determine the best overall algorithm.

Usage

classifyPairs(classifier = NULL, data = NULL,
    pre = function(x, y=NULL) { list(features=x) }, combinator = NULL)

Arguments

classifier

the mlr classifier to use. See examples.

data

the data to use with training and test sets. The structure returned by trainTest or cvFolds.

pre

a function to preprocess the data. Currently only normalize. Optional. Does nothing by default.

combinator

The classifier function to predict the overall best algorithm given the predictions for pairs of algorithms. Optional. By default, the overall best algorithm is determined by majority vote.

Value

• predictions
  a list of lists of data frames with the predictions for each test set. Each data frame has columns algorithm and score and is sorted according to preference, with the most preferred algorithm first. The score corresponds to the number of times the respective algorithm was predicted to be faster. If stacking is used, each data frame contains simply the best algorithm with a score of 1.
• predictor
  a function that encapsulates the classifier learned on the entire data set. Can be called with data for the same features with the same feature names as the training data to obtain predictions.
• models
  the models for each pair of algorithms trained on the entire data set. This is meant for debugging/inspection purposes and does not include any models used to combine predictions of individual models.

Details

classifyPairs takes the training and test sets in data and processes it using pre (if supplied). classifier is called to induce a classifier for each pair of algorithms to predict which one is better. If combinator is not supplied, the best overall algorithm is determined by majority vote. If it is supplied, it is assumed to be a classifier with the same properties as the other one. This classifier is passed the original features and the predictions for each pair of algorithms.

Which algorithm is better of a pair is determined by comparing their performance scores. Whether a lower performance number is better or not is determined by what was specified when the LLAMA data frame was created.

The evaluation across the training and test sets will be parallelized automatically if a suitable backend for parallel computation is loaded.

If the given classifier supports case weights, the performance difference between the best and the worst algorithm is passed as a weight for each instance.

Training this model can take a very long time. Given n algorithms, choose(n, 2) models are trained and evaluated. This is significantly slower than the other approaches that train a single model or one for each algorithm.

References

Xu, L., Hutter, F., Hoos, H. H., Leyton-Brown, K. (2011) Hydra-MIP: Automated Algorithm Configuration and Selection for Mixed Integer Programming. RCRA Workshop on Experimental Evaluation of Algorithms for Solving Problems with Combinatorial Explosion, 16--30.

See Also

classify, cluster, regression, regressionPairs

Examples

data(satsolvers)
folds = cvFolds(satsolvers)

res = classifyPairs(classifier=makeLearner("classif.J48"), data=folds)
# the total number of successes
sum(successes(folds, res))
# predictions on the entire data set
res$predictor(subset(satsolvers$data, TRUE, satsolvers$features))

# combine predictions using J48 induced classifier instead of majority vote
res = classifyPairs(classifier=makeLearner("classif.J48"),
                    data=folds,
                    combinator=makeLearner("classif.J48"))