Applications on Real Data

In kernopt: Estimating Count Data Distributions with Discrete Optimal Symmetric Kernel

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(kernopt)

SIMTAP project aimed to develop sustainable aquaculture production production system that, in particular, contributes to reduce fish feed inputs and resources consumption.
We used data from an experiment in which gilthead seabream (sparus aurata) were stocked in 1.6 m$^{3}$ tanks at a density of 1.5 kg$\cdot$m$^{-3}$. Fish were reared for 46 days in a single recirculating aquaculture system composed of three rearing tanks. At the beginning of the experiment, a number $n=200$ of the fish were individually weighed (dg), and their length at the caudal fork (mm) was measured.

The nonparametric kernel estimator was applied to provide smoothed count distributions of weight and length of fish measured. The performance of the nonparametric kernel estimator for modeling the empirical frequencies $f_{0}$ was evaluated by using Integrated Squared Error (ISE) given by $$ISE(h_{cv})=\sum_{x\in\mathbb{N}}(\widehat{f}{K,h{cv}}(x) - f_{0}(x))^{2}. $$

```r$) from the cross-validation procedure. Integrated Squared Error (ISE) was also calculated.'}

Data

data("fish_data", package = "kernopt") y <- fish_data$weight data1 <- as.data.frame(table(y)) x <- as.numeric(as.character(data1[, 1])) freq_weight <- as.numeric(as.character(data1[, 2])) f0 <- freq_weight / sum(freq_weight)

Optimal kernel

H <- seq((max(y) - min(y)) / 200, (max(y) - min(y)) / 2, length.out = 50) k <- 1 hcv_opt_k1 <- cv_bandwidth(kernel = "optimal", y, H, k = 1) fn_opt_k1 <- estim_kernel(kernel = "optimal", x, hcv_opt_k1, y, k = 1) ISE_opt_k1 <- sum((fn_opt_k1 - f0)^2)

Epanechnikov

H <- seq(2, 10, 1) hcv_epanech <- cv_bandwidth(kernel = "epanech", y, H) fn_epanech <- estim_kernel(kernel = "epanech", x, hcv_epanech, y, k = NULL) ISE_epanech <- sum((fn_epanech - f0)^2)

Graph

oldpar <- par(mfrow = c(1, 2)) plot( x, f0, col = "black", axes = F, lwd = 3, ylab = "Probability", xlab = "Weight (dg)", ylim = c(0, 0.06), xlim = c(41, 132), type = "h", main = "(a) Optimal k=1 - ISE=0.002, (hcv=0.94)", cex.axis = 1.70, cex.lab = 1.70 ) axis(1, at = x, cex.axis = 1.70, cex.lab = 1.70 ) axis(2) box() points( x + 0.4, fn_opt_k1, lwd = 3, col = "grey", lty = 1, type = "h" ) plot( x, f0, col = "black", axes = F, lwd = 3, ylab = "Probability", xlab = "Weight (dg)", ylim = c(0, 0.06), xlim = c(41, 132), type = "h", main = "(b) Epanechnikov - ISE=0.0033 (hcv=9)", cex.axis = 1.70, cex.lab = 1.70 ) axis(1, at = x, cex.axis = 1.70, cex.lab = 1.70 ) axis(2) box() points( x + 0.4, fn_epanech, lwd = 3, col = "grey", lty = 1, type = "h" )

Restore option

par(oldpar) ```

Acknowledgments

We thank the SIMTAP project (Self-sufficient Integrated Multi-Trophic AquaPonic systems, https://www.simtap.eu) supported by the PRIMA program (Partnership for Research and Innovation in the Mediterranean Area) under grand (18110-2) for providing the dataset.

kernopt documentation built on April 3, 2025, 9:34 p.m.