recommenderlab (version 1.0.6)

evaluationScheme: Creator Function for evaluationScheme

Description

Creates an evaluationScheme object from a data set. The scheme can be a simple split into training and test data, k-fold cross-evaluation or using k independent bootstrap samples.

Usage

evaluationScheme(data, ...)
# S4 method for ratingMatrix
evaluationScheme(data, method="split",
    train=0.9, k=NULL, given, goodRating = NA)

Value

Returns an object of class "evaluationScheme".

Arguments

data

data set as a ratingMatrix.

method

a character string defining the evaluation method to use (see details).

train

fraction of the data set used for training.

k

number of folds/times to run the evaluation (defaults to 10 for cross-validation and bootstrap and 1 for split).

given

single number of items given for evaluation or a vector of length of data giving the number of items given for each observation. Negative values implement all-but schemes. For example, given = -1 means all-but-1 evaluation.

goodRating

numeric; threshold at which ratings are considered good for evaluation. E.g., with goodRating=3 all items with actual user rating of greater or equal 3 are considered positives in the evaluation process. Note that this argument is only used if the ratingMatrix is a of subclass realRatingMatrix!

...

further arguments.

Details

evaluationScheme creates an evaluation scheme (training and test data) with k runs and one of the following methods:

"split" randomly assigns the proportion of objects specified by train to the training set and the rest is used for the test set.

"cross-validation" creates a k-fold cross-validation scheme. The data is randomly split into k parts and in each run k-1 parts are used for training and the remaining part is used for testing. After all k runs each part was used as the test set exactly once.

"bootstrap" creates the training set by taking a bootstrap sample (sampling with replacement) of size train times number of users in the data set. All objects not in the training set are used for testing.

For evaluation, Breese et al. (1998) introduced the four experimental protocols called Given 2, Given 5, Given 10 and All-but-1. During testing, the Given x protocol presents the algorithm with only x randomly chosen items for the test user, and the algorithm is evaluated by how well it is able to predict the withheld items. For All-but-x, the algorithm sees all but x withheld ratings for the test user. given controls x in the evaluations scheme. Positive integers result in a Given x protocol, while negative values produce a All-but-x protocol.

If a user does not have enough ratings to satisfy given, then the user is dropped from the evaluation with a warning.

References

Kohavi, Ron (1995). "A study of cross-validation and bootstrap for accuracy estimation and model selection". Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, pp. 1137-1143.

Breese JS, Heckerman D, Kadie C (1998). "Empirical Analysis of Predictive Algorithms for Collaborative Filtering." In Uncertainty in Artificial Intelligence. Proceedings of the Fourteenth Conference, pp. 43-52.

See Also

getData, evaluationScheme, ratingMatrix.

Examples

data("MSWeb")

MSWeb10 <- sample(MSWeb[rowCounts(MSWeb) >10,], 50)
MSWeb10

## simple split with 3 items given
esSplit <- evaluationScheme(MSWeb10, method="split",
        train = 0.9, k=1, given=3)
esSplit

## 4-fold cross-validation with all-but-1 items for learning.
esCross <- evaluationScheme(MSWeb10, method="cross-validation",
        k=4, given=-1)
esCross