Autoencoder for pseudoBulk

The present function compress data using autoencoder partially connected creating pseudoBulk matrix

autoencoder4pseudoBulk(
  group = c("sudo", "docker"),
  scratch.folder,
  file,
  separator,
  permutation,
  nEpochs,
  patiencePercentage = 5,
  seed = 1111,
  projectName,
  bN,
  lr = 0.01,
  beta_1 = 0.9,
  beta_2 = 0.999,
  epsilon = 1e-08,
  decay = 0,
  loss = "mean_squared_error",
  regularization = 10,
  version = 2
)

Arguments

group,

a character string. Two options: sudo or docker, depending to which group the user belongs

scratch.folder,

a character string indicating the path of the scratch folder

file,

a character string indicating the path of the file, with file name and extension included

separator,

separator used in count file, e.g. '\t', ','

permutation,

number of permutations to perform the pValue to evaluate clustering. Suggested minimal number of permutations 10

nEpochs,

number of Epochs for neural network training

patiencePercentage,

number of Epochs percentage of not training before to stop.

seed,

important value to reproduce the same results with same input

projectName,

might be different from the matrixname in order to perform different analysis on the same dataset

bN,

path to the clustering.output file

lr,

learning rate, the speed of learning. Higher value may increase the speed of convergence but may also be not very precise

beta_1,

look at keras optimizer parameters

beta_2,

look at keras optimizer parameters

epsilon,

look at keras optimizer parameters

decay,

look at keras optimizer parameters

loss,

loss of function to use, for other loss of function check the keras loss of functions.

regularization,

this parameter balances between reconstruction loss and enforcing a normal distribution in the latent space.

version,

version 1 implements static batchsize, version 2 implements adaptive batchsize

Author

Luca Alessandri, alessandri [dot] luca1991 [at] gmail [dot] com, University of Torino

Examples

if (FALSE) {
}