cv_libFM: Cross validation for rank and initialization standard...

In andland/libFMexe: Factorization Machines with R

Description

Cross validation for rank and initialization standard deviation with libFM factorization machines

Usage

cv_libFM(x, ...)

## S3 method for class 'data.frame'
cv_libFM(x, formula, validation, grouping,
  cv_verbosity = 0, ...)

## S3 method for class 'matrix'
cv_libFM(x, y, validation, y_validation, grouping,
  cv_verbosity = 0, ...)

## S3 method for class 'dgCMatrix'
cv_libFM(x, y, validation, y_validation, grouping,
  cv_verbosity = 0, ...)

## Default S3 method:
cv_libFM(x, dims = c(0, 8), init_stdevs = 0.1,
  folds = 5, validation, grouping, task = c("c", "r"), loss_function,
  cv_verbosity = 0, ...)

Arguments

x	data.frame, (sparse) matrix, or character vector used for cross validation
...	arguments passed to libFM
formula	formula of response and covariates included
validation	validation data.frame, (sparse) matrix, or character vector used for adaptive SGD
grouping	logical scalar or integer vector
cv_verbosity	how much feedback to give on the progress of cross validation
y, y_validation	numeric vectors of responses when train and validation are matrices
dims	vector of the two-way interaction dimensions to try
init_stdevs	vector of the initialization standard deviations to try
folds	number of cross validation folds or integer vector giving the fold ids for each row
task	classifcation or regression
loss_function	loss function to evaluate. See details for more information

Details

This function attempts to speed things up by converting all the data to libFM format first, so that there are not repetitive and computationally-costly data conversions.

If unspecified, loss_function will be mean squared error for regression or mean Bernoulli deviance for classification. You can specify your own function also. loss_function must be a function that takes two vector arguments and returns a scalar numeric. The first argument is the predicted value / probability and the second is the true value.

Value

A matrix of the cross validated loss_function. There are length(dims) rows and length(init_stdevs) columns.

Methods (by class)

• data.frame:

• matrix:

• dgCMatrix:

• default:

See Also

libFM

Examples

## Not run: 
data(movie_lens)
mses = cv_libFM(movie_lens, Rating ~ User + Movie,
                task = "r", dims = c(0, 5, 10), cv_verbosity = 1)
mses

# with a sparse matrix
movie_mat = Matrix::sparse.model.matrix(Rating ~ User + Movie - 1, movie_lens)
mses = cv_libFM(movie_mat, movie_lens$Rating,
                task = "r", dims = c(0, 5, 10), cv_verbosity = 1)
mses

## End(Not run)

andland/libFMexe documentation built on May 12, 2019, 2:41 a.m.