autoRaschOptions function

autoRasch Optimization Parameters Setting

Returns and updates the default settings used by the functions in autoRasch package.

autoRaschOptions(x = NULL)

Arguments

• x: A name of single parameter setting that is wanted to be shown. NULL means returns all parameters.

Returns

• fixed_par: A vector of names of the parameter types that are set to be fixed. It means that these parameters are not going to be estimated.

• fixed_theta: A vector of theta values when theta are listed in the fixed_par. If it is not set, it will be set to zero.

• fixed_beta: A vector of beta values when beta are listed in the fixed_par. If it is not set, it will be set to zero.

• fixed_gamma: A vector of gamma (natural logarithm of discrimination parameters, $\alpha = exp(\gamma)$) values when gamma are listed in the fixed_par. If it is not set, it will be set to zero.

• fixed_delta: A vector of delta values when delta are listed in the fixed_par. If it is not set, it will be set to zero.

• isPenalized_theta: It is a logical parameter whether, in the estimation procedure, theta is penalized or not.

• isPenalized_gamma: It is a logical parameter whether, in the estimation procedure, gamma is penalized or not.

• isPenalized_delta: It is a logical parameter whether, in the estimation procedure, delta is penalized or not.

• groups_map: A matrix $n x f$ to map the subject into DIF groups, where $n$ is number of subjects and $f$ is number of focal groups.

• optz_method: Options of the optimization method used. The default is optim which implies on applying the PJMLE which is implemented using optim(). When it is set to mixed means that it applies the coordinate descent.

• optim_control: A list of setting parameters of the optim(). For complete settings can be seen in stats::optim().

• lambda_theta: The regularization parameter to the theta. The default value is 0.05

• lambda_in: The regularization parameter to the gamma in the included itemset. The default value is 50.

• lambda_out: The regularization parameter to the gamma in the excluded itemset. The default value is 1.

• lambda_delta: The regularization parameter to the delta. The default value is 10.

• randomized: A logical parameter whether the initial values of the estimated parameters are randomized or not.

• random.init.th: A threshold value to limit the range of the initial values. The default value is 1e-2, means that the initial values range between [-0.01,0.01]

• isHessian: A logical parameter whether, in the estimation procedure, need to return the Hessian matrix or not. The default value is TRUE, which means the Hessian matrix will be computed.

• cd_control: A list of coordinate descent optimization setting.

• mode: An option setting to use "DIF" or "DSF" mode.

• isTraced: A logical value whether the progress need to be tracked or not.

Details

cd_control lists the parameters used to control the coordinate descent optimization procedure. The paramaters are:

• init.stepInitial value of the delta parameters updating step. The default is 1.
• scale.downA constant value to scale down the updating step. The default is 0.5.
• maxit.cd.higherMaximum iteration in the higher level coordinate descent. The default is 500.
• maxit.cd.lowerMaximum iteration for every coordinate optimization in the lower level coordinate descent. The default is 500.
• abs.tolThe convergence tolerance. The algorithm stops if it is unable to reduce the negative log likelihood value by the given tolerance. The default is 1e-12.
• max.dif.parThe convergence tolerance. The algorithm stops if it is unable to update all of the parameters' value by the given tolerance. The default is 1e-8.

Examples

### To show the default values
autoRaschOptions()
autoRaschOptions(x = "isHessian")

### To change the default values
adj_setting <- autoRaschOptions()
adj_setting$isHessian <- TRUE
pcm_res <- pcm(shortDIF, setting = adj_setting)