model.control: Auxiliary for Controlling FM Fitting

In evanwang1990/FMwR: Factorization Machines with R

Description

Auxiliary function for fm.train fitting.

Usage

model.control(task = "CLASSIFICATION", solver = "TDAP")

Arguments

task	A character string naming the task for FM model, such as "CLASSIFICATION", "REGRESSION"
...	a list of parameters for controlling the fitting process
solver	A character string naming an optimization routine of FM model, the default solver is "TDAP"

Details

FMwR provide many optimization routines including:

• "SGD": stochastic gradient descent, supporting ℓ1 or ℓ2 SGD.solver

• "FTRL": FTRL-Proximal FTRL.solver

• "TDAP": TDAP.solver

• "MCMC": MCMC.solver

• "ALS": ALS.solver

The other parameters including:

keep.w0

whether to keep intercept, default TRUE

L2.w0

regularization of intercept, default 0.0

keep.w1

whether to keep 1-way interaction, default TRUE

L1.w1

ℓ1 regularization of 1-way interaction, default 0.0

L2.w1

ℓ2 regularization of 1-way interaction, default 0.0

factor.number

number of 2-way latent factors, default 2L

v.init_mean

mean for initialization of 2-ways latent factors, default 0.0

v.init_stdev

stdev for initialization of 2-ways latent factors, default 0.01

L1.v

ℓ1 regularization of 2-way interaction, default 0.0

L2.v

ℓ2 regularization of 2-way interaction, default 0.0

Examples

model.control() # the default settings

evanwang1990/FMwR documentation built on May 16, 2019, 9:38 a.m.