d_H2OAE: Autoencoder using H2O

In egenn/rtemis: Machine Learning and Visualization

Autoencoder using H2O

Description

Train an Autoencoder using h2o::h2o.deeplearning Check out the H2O Flow at ⁠[ip]:[port]⁠, Default IP:port is "localhost:54321" e.g. if running on localhost, point your web browser to localhost:54321

Usage

d_H2OAE(
  x,
  x.test = NULL,
  x.valid = NULL,
  ip = "localhost",
  port = 54321,
  n.hidden.nodes = c(ncol(x), 3, ncol(x)),
  extract.layer = ceiling(length(n.hidden.nodes)/2),
  epochs = 5000,
  activation = "Tanh",
  loss = "Automatic",
  input.dropout.ratio = 0,
  hidden.dropout.ratios = rep(0, length(n.hidden.nodes)),
  learning.rate = 0.005,
  learning.rate.annealing = 1e-06,
  l1 = 0,
  l2 = 0,
  stopping.rounds = 50,
  stopping.metric = "AUTO",
  scale = TRUE,
  center = TRUE,
  n.cores = rtCores,
  verbose = TRUE,
  save.mod = FALSE,
  outdir = NULL,
  ...
)

Arguments

x	Vector / Matrix / Data Frame: Training set Predictors
x.test	Vector / Matrix / Data Frame: Testing set Predictors
x.valid	Vector / Matrix / Data Frame: Validation set Predictors
ip	Character: IP address of H2O server. Default = "localhost"
port	Integer: Port number for server. Default = 54321
n.hidden.nodes	Integer vector of length equal to the number of hidden layers you wish to create
extract.layer	Integer: Which layer to extract. For regular autoencoder, this is the middle layer. Default = ceiling(length(n.hidden.nodes)/2)
epochs	Integer: How many times to iterate through the dataset. Default = 5000
activation	Character: Activation function to use: "Tanh" (Default), "TanhWithDropout", "Rectifier", "RectifierWithDropout", "Maxout", "MaxoutWithDropout"
loss	Character: "Automatic" (Default), "CrossEntropy", "Quadratic", "Huber", "Absolute"
input.dropout.ratio	Float (0, 1): Dropout ratio for inputs
hidden.dropout.ratios	Vector, Float (0, 2): Dropout ratios for hidden layers
learning.rate	Float: Learning rate. Default = .005
learning.rate.annealing	Float: Learning rate annealing. Default = 1e-06
l1	Float (0, 1): L1 regularization (introduces sparseness; i.e. sets many weights to 0; reduces variance, increases generalizability)
l2	Float (0, 1): L2 regularization (prevents very large absolute weights; reduces variance, increases generalizability)
stopping.rounds	Integer: Stop if simple moving average of length stopping.rounds of the stopping.metric does not improve. Set to 0 to disable. Default = 50
stopping.metric	Character: Stopping metric to use: "AUTO", "deviance", "logloss", "MSE", "RMSE", "MAE", "RMSLE", "AUC", "lift_top_group", "misclassification", "mean_per_class_error". Default = "AUTO" ("logloss" for Classification, "deviance" for Regression)
scale	Logical: If TRUE, scale input before training autoencoder. Default = TRUE
center	Logical: If TRUE, center input before training autoencoder. Default = TRUE
n.cores	Integer: Number of cores to use
verbose	Logical: If TRUE, print summary to screen.
save.mod	Logical: If TRUE, save all output to an RDS file in outdir save.mod is TRUE by default if an outdir is defined. If set to TRUE, and no outdir is defined, outdir defaults to paste0("./s.", mod.name)
outdir	Path to output directory. If defined, will save Predicted vs. True plot, if available, as well as full model output, if save.mod is TRUE
...	Additional arguments to pass to h2p::h2o.deeplearning

Value

rtDecom object

Author(s)

E.D. Gennatas

See Also

decom

Other Decomposition: d_H2OGLRM(), d_ICA(), d_Isomap(), d_KPCA(), d_LLE(), d_MDS(), d_NMF(), d_PCA(), d_SPCA(), d_SVD(), d_TSNE(), d_UMAP()

Other Deep Learning: s_H2ODL(), s_TFN()

egenn/rtemis documentation built on Oct. 28, 2024, 6:30 a.m.