test_results: Get RMSE & explained variance for warm and cold test results

In BaTFLED3D: Bayesian Tensor Factorization Linked to External Data

Description

Get RMSE & explained variance for warm and cold test results

Usage

test_results(m, d, test.results = numeric(0), verbose = T,
  warm.resp = numeric(0), test.m1 = numeric(0), test.m2 = numeric(0),
  test.m3 = numeric(0), test.m1m2 = numeric(0), test.m1m3 = numeric(0),
  test.m2m3 = numeric(0), test.m1m2m3 = numeric(0))

Arguments

m	a CP_model or Tucker_model object
d	an input data object created with input_data
test.results	an object generated by this function that will combined with the new results
verbose	Logical indicating whether to print the resulting prediction measures (TRUE)
warm.resp	True responses for warm test data (optional).
test.m1	True responses for mode 1 cold test data (optional).
test.m2	True responses for mode 2 cold test data (optional).
test.m3	True responses for mode 3 cold test data (optional).
test.m1m2	True responses for mode 1/2 combination cold test data (optional).
test.m1m3	True responses for mode 1/3 combination cold test data (optional).
test.m2m3	True responses for mode 2/3 combination cold test data (optional).
test.m1m2m3	True responses for mode 1/2/3 combination cold test data (optional).

Value

list of results TODO: add more here

Examples

data.params <- get_data_params(c('decomp=Tucker'))
toy <- mk_toy(data.params)

# Make training data object excluding the first two samples for modes 1 & 2.
train.data <- input_data$new(mode1.X=toy$mode1.X[-(1:2),-1],
                             mode2.X=toy$mode2.X[-(1:2),-1],
                             mode3.X=toy$mode3.X[,-1],
                             resp=toy$resp)
# Remove some responses for warm prediction
warm.ind <- sample(1:prod(dim(train.data$resp)), 20)
warm.resp <- train.data$resp[warm.ind]
train.data$resp[warm.ind] <- NA

# Make testing objects 
m1.test.data <- input_data$new(mode1.X=toy$mode1.X[1:2,-1],
                               mode2.X=toy$mode2.X[-(1:2),-1],
                               mode3.X=toy$mode3.X[,-1],
                               resp=toy$resp[1:2,-(1:2),])
m2.test.data <- input_data$new(mode1.X=toy$mode1.X[-(1:2),-1],
                               mode2.X=toy$mode2.X[1:2,-1],
                               mode3.X=toy$mode3.X[,-1],
                               resp=toy$resp[-(1:2),1:2,])
m1m2.test.data <- input_data$new(mode1.X=toy$mode1.X[1:2,-1],
                                 mode2.X=toy$mode2.X[1:2,-1],
                                 mode3.X=toy$mode3.X[,-1],
                                 resp=toy$resp[1:2,1:2,])
                               
model.params <- get_model_params(c('decomp=Tucker'))
toy.model <- mk_model(train.data, model.params)
toy.model$rand_init(model.params)
toy.model$iter <- 1

test.results <- numeric(0)
test_results(m=toy.model, d=train.data, warm.resp=warm.resp, 
             test.m1=m1.test.data, test.m2=m2.test.data, 
             test.m1m2=m1m2.test.data)

BaTFLED3D documentation built on May 2, 2019, 2:38 p.m.