1. Parameters <trainModel>
    (1) trainOriginal; # a data frame consists of 'data' and 'y', for training, required
    (2) testOriginal; # same as trainOriginal, for testing, default NULL
    (3) pv_thresh; # p-value threshold for selecting a variable/terminating, default 0.01
    (4) max; # maximum variables to select for the model, default minimun of 25 and varaible amount
    (5) var_thresh; # variance (minimum) threshold for considering a variable, default 0.01
    (6) project; # decide whether to project variables, TURE for fspls and FALSE for fs, default TRUE
    (7) info; # dummy parameter from previous version, default empty list
    (8) pivot; # pivot for (multi) logistic regression, default 0
    (9) fit_coeff <new added>; # decide whether to fit coefficients, set it as FALSE to keep consistent with the previous version that only select variables, default TRUE
    (10) log <new added>; # log file name, run program in a silence mode with a log, default NULL
    (11) append <new added>; # decide whether to append the outputs to the log file, default FALSE
    (12) refit <new added>; # decide whether to refit the model at the end, default FALSE

2. Bug Fixed
    (1) add 'means = means[variables]' in line 529 and 530 (using default in previous versions)
    (2) replace 'const_term=rep(0,length(indices))' with 'const_term=constant_term' in line 502
    
3. Change Log
    (1) For multinomial problem, instead of solving (K-1) binomial problems, R package mlogit was employed to calculate the likelihood of each variable similar 
        to the function glm in binomial problem. Thus lead to corresponding modifications in function preprocess find_best_glm, convertToBinary, fitModelShrink, 
        fitModel getOffset, refit, evalModels, and pred.
    (2) For output, a unified format was defined. In every step, indices of the variable selected (ind), RMSE (rms), AUC (auc) and accurary (acc) for both train
        and test dataset (if not NULL) will be printed. AUC and accurary for linear regressions will be NA, AUC and RMSE for multinomial problems will be NA.