train_model: Trains a VAE or autoencoder model. This acts as a wrapper for...

In ML2Pvae: Variational Autoencoder Models for IRT Parameter Estimation

Trains a VAE or autoencoder model. This acts as a wrapper for keras::fit().

Description

Trains a VAE or autoencoder model. This acts as a wrapper for keras::fit().

Usage

train_model(
  model,
  train_data,
  num_epochs = 10,
  batch_size = 1,
  validation_split = 0.15,
  shuffle = FALSE,
  verbose = 1
)

Arguments

model	the keras model to be trained; this should be the vae returned from build_vae_independent() or build_vae_correlated
train_data	training data; this should be a binary num_students by num_items matrix of student responses to an assessment
num_epochs	number of epochs to train for
batch_size	batch size for mini-batch stochastic gradient descent; default is 1, detailing pure SGD; if a larger batch size is used (e.g. 32), then a larger number of epochs should be set (e.g. 50)
validation_split	split percentage to use as validation data
shuffle	whether or not to shuffle data
verbose	verbosity levels; 0 = silent; 1 = progress bar and epoch message; 2 = epoch message

Value

a list containing training history; this holds the loss from each epoch which can be plotted

Examples

data <- matrix(c(1,1,0,0,1,0,1,1,0,1,1,0), nrow = 3, ncol = 4)
Q <- matrix(c(1,0,1,1,0,1,1,0), nrow = 2, ncol = 4)
models <- build_vae_independent(4, 2, Q)
vae <- models[[3]]
history <- train_model(vae, data, num_epochs = 3, validation_split = 0, verbose = 0)
plot(history)

ML2Pvae documentation built on May 23, 2022, 9:05 a.m.