Tucker_model: Factorization object for 3D Tucker models.
In nathanlazar/BaTFLED3D: Bayesian Tensor Factorization Linked to External Data

Description Usage Format Members Methods Examples

Tucker_model objects are 'R6' objects so that their values can be updated in place. The object is treated like an environment and components are accessed using the $ operator. When creating a new Tucker_model object it will be populated with default values and empty matrices. To initialize a Tucker_model call the initialize() method.

1	Tucker_model

An R6Class generator object

iter: Integer showing the number of iterations that have been run on this object.
early.stop: Stop if the lower bound increases by less than this value.
lower.bnd: Vector storing the lower bound values during training.
RMSE: Vector of the root mean squared error of the predictions during training.
H.RMSE: vector of the root mean squared error of predictions made by multiplying the H matrices.
exp.var: Vector of the explained variance of predictions during training.
p.cor: Vector of the Pearson correlation of predictions during training.
s.cor: Vector of the Spearman correlation of predictions during training.
times: Vector of the time taken for each update iteration.
core.mean: Mean parameters of the q Gaussian distributions in the core tensor.
core.var: Variance parameters of the q Gaussian distributions in the core tensor.
core.lambda.shape: Prior for the shape parameter of the gamma distribution on the core precision.
core.lambda.scale: Prior for the scale parameter of the gamma distribution on the core precision.
resp: array storing the predicted response tensor.
delta: binary array indicating whether the response is observed.
core.var: variance parameters of the q Gaussian distributions in the core tensor.
m1Xm1X: Product of mode1.X with itself stored to avoid recalculating.
m2Xm2X: Product of mode2.X with itself stored to avoid recalculating.
m3Xm3X: Product of mode3.X with itself stored to avoid recalculating.
mode1.lambda.shape: Matrix storing the shape parameters for the gamma distributions on the mode 1 projection (A) matrix.
mode1.lambda.scale: Matrix storing the scale parameters for the gamma distributions on the mode 1 projection (A) matrix.
mode2.lambda.shape: Matrix storing the shape parameters for the gamma distributions on the mode 2 projection (A) matrix.
mode2.lambda.scale: Matrix storing the scale parameters for the gamma distributions on the mode 2 projection (A) matrix.
mode3.lambda.shape: Matrix storing the shape parameters for the gamma distributions on the mode 3 projection (A) matrix.
mode3.lambda.scale: Matrix storing the scale parameters for the gamma distributions on the mode 3 projection (A) matrix.
mode1.A.mean: Matrix storing the mean parameters for the normal distributions on the mode 1 projection (A) matrix.
mode1.A.cov: Array storing the covariance parameters for the normal distributions on the mode 1 projection (A) matrix.
mode2.A.mean: Matrix storing the mean parameters for the normal distributions on the mode 2 projection (A) matrix.
mode2.A.cov: Array storing the covariance parameters for the normal distributions on the mode 2 projection (A) matrix.
mode3.A.mean: Matrix storing the mean parameters for the normal distributions on the mode 3 projection (A) matrix.
mode3.A.cov: Array storing the covariance parameters for the normal distributions on the mode 3 projection (A) matrix.
mode1.H.mean: Matrix storing the mean parameters for the normal distributions on the mode 1 latent (H) matrix.
mode1.H.var: Matrix storing the variance parameters for the normal distributions on the mode 1 latent (H) matrix.
mode2.H.mean: Matrix storing the mean parameters for the normal distributions on the mode 2 latent (H) matrix.
mode2.H.var: Matrix storing the variance parameters for the normal distributions on the mode 2 latent (H) matrix.
mode3.H.mean: Matrix storing the mean parameters for the normal distributions on the mode 3 latent (H) matrix.
mode3.H.var: Matrix storing the variance parameters for the normal distributions on the mode 3 latent (H) matrix.
sigma2: Variance for the response tensor.
m1.sigma: Variance for the mode 1 latent (H) matrix.
m2.sigma: Variance for the mode 2 latent (H) matrix.
m3.sigma: Variance for the mode 3 latent (H) matrix.
m1.alpha: Prior shape parameter for the gamma distribution on the precision of the mode 1 projection (A) matrix.
m1.beta: Prior scale paramet for the gamma distribution on the precision of the mode 1 projection (A) matrix.
m2.alpha: Prior shape parameter for the gamma distribution on the precision of the mode 2 projection (A) matrix.
m2.beta: Prior scale paramet for the gamma distribution on the precision of the mode 2 projection (A) matrix.
m3.alpha: Prior shape parameter for the gamma distribution on the precision of the mode 3 projection (A) matrix.
m3.beta: Prior scale paramet for the gamma distribution on the precision of the mode 3 projection (A) matrix.
core.alpha: Prior shape parameter for the gamma distribution on the precision of the core tensor.
core.beta: Prior scale parameter for the gamma distribution on the precision of the core tensor.
core.0D.alpha: Prior shape parameter for the gamma distribution on the precision of the 0D subset of the core tensor.
core.0D.beta: Prior scale parameter for the gamma distribution on the precision of the 0D subset of the core tensor.
core.1D.alpha: Prior shape parameter for the gamma distribution on the precision of the 1D subset of the core tensor.
core.1D.beta: Prior scale parameter for the gamma distribution on the precision of the 1D subset of the core tensor.
core.2D.alpha: Prior shape parameter for the gamma distribution on the precision of the 2D subset of the core tensor.
core.2D.beta: Prior scale parameter for the gamma distribution on the precision of the 2D subset of the core tensor.
core.3D.alpha: Prior shape parameter for the gamma distribution on the precision of the 3D subset of the core tensor.
core.3D.beta: Prior scale parameter for the gamma distribution on the precision of the 3D subset of the core tensor.

new(data, params): Creates a new Tucker_model object with matrices sized accoring to the matrices in data.
rand_init(params): Initializes the Tucker_model with random values accoring to params.

data.params <- get_data_params(c('decomp=Tucker'))
toy <- mk_toy(data.params)
train.data <- input_data$new(mode1.X=toy$mode1.X[,-1],
                             mode2.X=toy$mode2.X[,-1],
                             mode3.X=toy$mode3.X[,-1],
                             resp=toy$resp)
model.params <- get_model_params(c('decomp=Tucker'))
toy.model <- mk_model(train.data, model.params)
toy.model$rand_init(model.params)