#DATE MODIFIED: 04/07/2014            

extract_table_term_factor <- function(i,list_param){

  #This function generate a linear model for proportion of hold out effect on accuracy

  #Add option to choose MAE later

  #First parse arguments

  interpolation_method <- list_param$list_interp_method[[i]]

  tb_mv <- list_param$tb_mv

  #Begin script:

  list_pred_mod_name <- unique(tb_mv$pred_mod)

  list_prop_cat <- unique(tb_mv$prop_month)  

  list_pred_mod_name <- grep(c("mod_kr"),list_pred_mod_name,value=TRUE,invert=TRUE)

  list_mod_table <- vector("list",length=length(list_pred_mod_name))

  tb_dat<- subset(tb_mv,tb_mv$method_interp==interpolation_method) 

  for(j in 1:length(list_mod_table)){

      mod <-lm(rmse~ as.factor(prop_month),

               data=subset(tb_dat,tb_dat$pred_mod==list_pred_mod_name[[j]]))      

      term_table <- summary(mod)$coefficients

      list_mod_table[[j]] <- term_table  

  }

  names(list_mod_table) <- list_pred_mod_name

  tx <- lapply(list_mod_table,function(x){x[,c(1,4)]})

  tx <-lapply(tx, round,digit=3)

  #tx <- format(tx,digits=3)

  #column_tab <- paste(format(tx[[1]][,1],digits=3)," ",

  #             "(",format(tx[[1]][,2],digits=3),")",sep="")

  #column_tab <-lapply(tx,function(x){paste(format(x[,1],digits=3)," ",

  #             "(",format(x[,2],digits=3),")",sep="")})

  column_tab <-lapply(tx,function(x){as.data.frame(paste(format(x[,1],digits=3)," ",

               "(",format(x[,2],digits=3),")",sep=""))})

  table_method <- as.data.frame(do.call(cbindX,column_tab))

  names(table_method) <- list_pred_mod_name

  table_method$method_interp <- rep(interpolation_method,nrow(table_method))

  table_method$prop <- list_prop_cat

  mod_table_obj<- list(list_mod_table,table_method)

  names(mod_table_obj) <- c("list_mod_table","table_method")

  return(mod_table_obj)

}  

calc_stat_from_tb_list <-function(i,list_param){

  #Calculate statistics from validation and training out of raster_prediction_obj

  #If training is TRUE, then using training dataset

  tb <- list_param$list_tb[[i]] #raster prediction object to use

  stat <- list_param$stat

  training <- list_param$training

  #extract relevant information

  if(training==TRUE){

    rownames(tb)<-NULL #remove row names

  }

  #Now summarize

  t<-melt(tb,

          measure=c("mae","rmse","r","me","m50"), 

          id=c("pred_mod"),

          na.rm=T)

  stat_tb<-cast(t,pred_mod~variable,stat)

  return(stat_tb)

}