slb.load.datasets: Load Datasets from Available Repositories

In neurodata/slbR: Statistical Learning Benchmarks

Description

A function to load a specified dataset from the PMLB dataset.

Usage

slb.load.datasets(repositories = NULL, datasets = NULL, tasks = NULL,
  clean.invalid = TRUE, clean.ohe = FALSE, verbose = TRUE, ...)

Arguments

repositories	the name of the repository you would like to query a dataset from. Defaults to NULL, which will return datasets matching from all repositories. NULL Load datasets from all repositories matching other queries. "pmlb" Load datasets from the Penn Machine-Learning Benchmarks. "uci" Load datasets from the University of California - Irvine Machine Learning Repository. "mnist" Load datasets from the MNIST dataset. c("repo1", "repo2", ...) Load data from the indicated repositories.
datasets	the name of the dataset you wish to load. Defaults to NULL. NULL Load all the datasets without specifying a specific name matching the desired query. 'datasetid' Returns the dataset with the desired id matching the desired query. c("datasetid1", "datasetid2", ...) Load data from the indicated datasets.
tasks	the type of the task, either "classification" or "regression". Defaults to NULL. NULL Return all datasets matching the desired query. 'classification' Load all classification datasets matching the desired query. 'regression' Load all regression datasets matching the desired query. c("taskid1", "taskid2", ...) Load data for the indicated tasks.
clean.invalid	whether to remove samples with invalid entries. Defaults to TRUE. TRUE Remove samples that have features with NaN entries or non-finite. FALSE Do not remove samples that have features with NaN entries or are non-finite..
clean.ohe	options for whether to one-hot-encode columns. Defaults to FALSE. clean.ohe < 1 Converts columns with < thr*n unique identifiers to one-hot encoded. is.integer(clean.ohe) Converts columns with < thr unique identifiers to one-hot encoded. FALSE Do not one-hot-encode any columns.
verbose	whether to print messages to the console if a repository or dataset is being ignored. defaults to TRUE.
...	trailing args.

Value

A list of lists, where each element is a key-worded list for a particular benchmark dataset, containing at least the following:

X	[n, d] array with the n samples in d dimensions.
Y	[n] vector or [n, r] array with responses for each of the n samples.

Author(s)

Eric Bridgeford

Examples

library(slb)
# request 1 specific dataset from the pmlb dataset
test <- slb.load.datasets(repositories="pmlb", datasets="ESL", clean.invalid=FALSE, clean.ohe=FALSE)
length(test$ESL$Y) == 488 # a known example from the pmlb dataset

# request all of the pmlb classification datasets
## Not run: 
test <- slb.load.datasets(repositories="pmlb", tasks="classification")
length(test) <- 166  # validates that we loaded all of the classification datasets from pmlb

## End(Not run)

neurodata/slbR documentation built on May 22, 2019, 2:41 p.m.