smartPARE_train: Train the smartPARE CNN model

In kristianHoden/smartPARE: An R package for efficient identification of true mRNA cleavage sites

Description

Build the CNN model based on directories with cleavage images; train/goodUp (true cleavages on 5' strand), train/goodDown (true cleavages on 3' strand) and train/bad (false cleavages)

Usage

smartPARE_train(
  homePath1 = paste0(system.file("example/", package = "smartPARE"), "/"),
  pixels1 = 28,
  search_bound = list(denseLoop2 = c(0, 4), epochs2 = c(100, 300), batch_size2 = c(32,
    128), dropout2 = c(0, 0.3), validation_split2 = c(0.1, 0.4), convolutionalLoop2 =
    c(1, 4), NO_pooling2 = c(1, 2)),
  n_iter = 100
)

Arguments

homePath1	Path to directory containing a directory with the following subdirs train/goodUp (true cleavages on 5' strand), train/goodDown (true cleavages on 3' strand) and train/bad (false cleavages)
pixels1	Number of pixels to convert each image to
search_bound	List of min and max values for the following variables: denseLoop2, epochs2, batch_size2, validation_split2, convolutionalLoop2 and NO_pooling2
n_iter	Number of iterations to run the Bayesian optimization

Examples

smartPARE_train(homePath1 = "example/",
pixels1 = 28,
search_bound = list(denseLoop2 = c(0,4),
                    epochs2 = c(100, 300),
                    batch_size2 = c(32,128),
                  dropout2 = c(0, 0.3),
                    validation_split2 = c(0.1,0.4),
                 convolutionalLoop2 = c(1,4),
                 NO_pooling2 = c(1,2)
),
n_iter = 100)

kristianHoden/smartPARE documentation built on July 3, 2021, 7:10 p.m.