Fix coco.predict.

NAMESPACE

@@ -27,6 +27,7 @@ export(cocoResid)
 export(cocoScore)
 export(cocoSim)
 export(cocoSoc)
+export(dGP1)
 export(installJuliaPackages)
 export(setJuliaSeed)
 importFrom(Rcpp,sourceCpp)

R/cocoForecastKSteps.R

@@ -2,8 +2,10 @@ cocoForecastKSteps <- function(fit, k=3, number_simulations=1000, alpha=0.05, co
                                decimals=4, 
                                julia=FALSE){
   if (julia){
+
     forecasts <- cocoForecastKStepsJulia(fit, k=k, number_simulations = number_simulations,
                                    covariates=covariates)
+
   } else {
     forecasts <- cocoForecastKStepsRCPP(fit, k=k, number_simulations = number_simulations,
                                   covariates=covariates)
@@ -13,6 +15,13 @@ cocoForecastKSteps <- function(fit, k=3, number_simulations=1000, alpha=0.05, co
     densities <- forecasts[[i]][,"frequency"]
     x <- as.numeric(forecasts[[i]][,"value"])
 
+    if (min(x) > 0){
+      x_fill <- 0:(min(x)-1)
+      dens <- rep(0, (min(x)))
+      x <- c(x_fill, x)
+      densities <- c(dens, densities)
+    }
+
     mode <- match(max(densities), densities) - 1
     distribution_function <- cumsum(densities)
     median <- min(which(distribution_function >= 0.5)) - 1

R/cocoForecastKStepsRCPP.R

@@ -5,30 +5,24 @@ cocoForecastKStepsRCPP <- function(fit, k=1, number_simulations=500, covariates=
     init <- fit$ts[(length(fit$ts)-1):length(fit$ts)]
   }
 
-  if (is.null(covariates)){
-    run_sim <- function(i){
-      cocoSim_base(
-        type = fit$type, order = fit$order, par = fit$par, size = k+length(init),
-        seasonality = c(1,2), init = init)$data
-    }
-  } else {
-    run_sim <- function(i){  
-      output <- cocoSim_cov(
-        type = fit$type, order = fit$order, par = fit$par, size =  k+length(init),
-        xreg = covariates,
-        seasonality = c(1,2), init = init, link_function=fit$link_function)$data
-    }
+  run_sim <- function(i){ 
+    output <- cocoSim(type = fit$type, order = fit$order, par = fit$par, length =  k+length(init),
+            xreg = covariates,
+            julia=FALSE, link_function=fit$link_function)
   }
+
   if (k > 1) {
     out_matrix <- t(sapply(1:number_simulations, run_sim))
   } else if (k==1){
-    out_matrix <- as.matrix(sapply(1:number_simulations, run_sim))
+    out_matrix <- as.matrix(sapply(1:number_simulations, run_sim)[1, ])
   }
 
   freq_table <- function(i){
     df <- as.data.frame(table(out_matrix[,i]) / number_simulations)
-    colnames(df) <- c("value", "frequency")
+    colnames <- c("value", "frequency")
+    colnames(df) <- colnames
     df["value"] <- as.numeric(levels(df[, "value"]))
+
     return(df)
   }
 

R/cocoForecastOneStep.R

@@ -19,7 +19,11 @@ cocoForecastOneStep <- function(coco, max=NULL, epsilon=1e-12, xcast=NULL,
 
   if (!is.null(coco$julia_reg) & julia){
     addJuliaFunctions()
-    coco_forecast <- JuliaConnectoR::juliaGet( JuliaConnectoR::juliaCall("cocoPredictOneStep", coco$julia_reg, 0:max_use, xcast))
+    if ((k == 1) & (is.matrix(xcast))) {
+      xcast <- c(xcast)
+    }
+    coco_forecast <- JuliaConnectoR::juliaGet( JuliaConnectoR::juliaCall("cocoPredictOneStep",
+                                                                         coco$julia_reg, 0:max_use, xcast))
     densities <- coco_forecast$values[[4]]
     if (is.null(max)){
       cumulative <- cumsum(densities)

R/dGP1.R

@@ -1,3 +1,4 @@
+#' @export
 dGP1 <- function(x, y, par) {
   if ((round(y) != y) | (y < 0)) {
     stop("y must be a non-negative integer")

R/plot.cocoForecast.R

@@ -2,13 +2,14 @@
 #' @export
 ggplot2::autoplot
 #' @export
-autoplot.cocoForecast <- function(object, ...){
+autoplot.cocoForecast <- function(object, breaks=NULL, width=0.1, ...){
 
   pl <- ggplot2::ggplot(mapping = ggplot2::aes(x = object$x, y = object$densities_plot)) +
-      ggplot2::geom_bar(stat="identity", position="dodge", width=0.04) + 
+      ggplot2::geom_bar(stat="identity", position="dodge", width=width) + 
       ggplot2::labs(title = "Probability mass forecast", x = "Support", y = "Probability mass") +
-      ggplot2::scale_x_continuous(breaks=object$x) +
+    ggplot2::xlim(c(0, max(object$x))) +
       ggplot2::theme_bw() + ggplot2::theme(text = ggplot2::element_text(size = 20)) 
+
   pl
 }
 

R/predict.coco.R

@@ -34,6 +34,10 @@ predict.coco <- function(object, k=1, number_simulations=1000,
                          decimals = 4,
                          julia=FALSE,...) {
 
+  if ((k == 1) & (!is.matrix(xcast))){
+    xcast <- matrix(xcast, nrow=1)
+  }
+
   if ((k==1) & (!simulate_one_step_ahead)){
     return(cocoForecastOneStep(object, max=max, epsilon=epsilon, xcast=xcast,
                                alpha=alpha,