In pokyah/agrometeoR.extras: complementary functions for agrometeoR destined to dataviz

knitr::opts_chunk$set(echo = TRUE, message = FALSE, warning = FALSE)
devtools::load_all()
library(dplyr)
library(sf)

---

title: "presentation SC april 2019 : The agrometeoR package" author: "Thomas Goossens" date: "4/23/2019" always_allow_html: yes output: revealjs::revealjs_presentation: css: - "~/Documents/code/pokyah/craw-reveal-template/theme/craw.css" center: no highlight: pygments incremental: yes self_contained: true slide_level: 2 theme: default transition: slide fig_caption: false reveal_options: previewLinks: false md_document: toc: no toc_depth: 6 variant: markdown_github word_document: toc: no toc_depth: '6' pdf_document: default odt_document: fig_height: 5 fig_width: 7 html_document: theme: default toc: yes toc_depth: 6 toc_float: collapsed: yes smooth_scroll: yes

---

Defining the best spatialization method

idea : test various "explorative constructions"

1 EC = 1 unique combination of

• multiple learners
• multiple potential hyper-parameters
• multiple explanatory variables
• multiple data sources (stations) : test the interest of networks integration

Table of EC

explorative_constructions = read.csv2("./explorative_constructions.csv", sep = ";") %>%
  dplyr::mutate_all(.funs = as.factor)
knitr::kable(head(explorative_constructions))

Why EC

• deliver data with metadata
• which best EC among which tested EC's
• easy shortcut name to precisely know what we are talking about

Testing EC's

• Conduction of batches of benchmarks experiments

Benchmarking of EC's

Function to conduct benchmark

The agrometeoR package offers a function for this : makeBatchOfBenchExp

# example
bmrsResults = makeBatchOfBenchExp(
  tasks = tasks,
  learners = agrometeorLearners,
  measures = list(rmse, mae, mse),
  keep.pred = TRUE,
  models = FALSE,
  grouping = 100,
  level = "mlr.benchmark",
  resamplings = "LOO",
  cpus = 4
) 

benchmark conduction difficulties

• memory intensive => split the results in chunks written to disks to free RAM
• paralellel computing to speed up the computations.
• run on dedicated server using bash in a tmux session
• use scp to retrieve the results from server
• hourly 2 years 5 learners => 24h of computation

Conducted benchmarks :

01/01/2016 -> 31/12/2017, 5 learners (KED, OK, NN1, IDW, MultiReg), with elevation as explanatory for KED, OK, and multiple linear reg.

• daily tsa max pameseb
• daily tsa max pameseb + IRM
• hourly tsa pameseb
• hourly tsa pameseb + IRM

summary statistics of hourly tsa Pameseb : RMSE

hourly_tsa_pameseb_rmse = read.csv2("~/Desktop/sc_april_2019/tsa hourly Pameseb summary stats rmse.csv", sep = ",")
knitr::kable(hourly_tsa_pameseb_rmse)

summary statistics of hourly tsa Pameseb & IRM : RMSE

hourly_tsa_pameseb_IRM_rmse = read.csv2("~/Desktop/sc_april_2019/tsa hourly Pameseb_IRM summary stats rmse.csv", sep = ",")
knitr::kable(hourly_tsa_pameseb_IRM_rmse)

summary statistics of hourly tsa Pameseb : Residuals

hourly_tsa_pameseb_residuals = read.csv2("~/Desktop/sc_april_2019/tsa hourly Pameseb summary stats residuals.csv", sep = ",")
knitr::kable(hourly_tsa_pameseb_residuals)

summary statistics of hourly tsa Pameseb & IRM : Residuals

hourly_tsa_pameseb__IRM_residuals = read.csv2("~/Desktop/sc_april_2019/tsa hourly Pameseb_IRM summary stats residuals.csv", sep = ",")
knitr::kable(hourly_tsa_pameseb__IRM_residuals)

summary statistics of daily tsa max Pameseb : RMSE

daily_tsamax_pameseb_rmse = read.csv2("~/Desktop/sc_april_2019/tsamax daily Pameseb summary stats rmse.csv", sep = ",")
knitr::kable(daily_tsamax_pameseb_rmse)

summary statistics of daily tsa max Pameseb & IRM : RMSE

daily_tsamax_pameseb_IRM_rmse = read.csv2("~/Desktop/sc_april_2019/tsamax daily Pameseb_IRM summary stats rmse.csv", sep = ",")
knitr::kable(daily_tsamax_pameseb_IRM_rmse)

summary statistics of daily tsa max Pameseb : Residuals

daily_tsamax_pameseb_residuals = read.csv2("~/Desktop/sc_april_2019/tsamax daily Pameseb summary stats residuals.csv", sep = ",")
knitr::kable(daily_tsamax_pameseb_residuals)

summary statistics of daily tsa max Pameseb & IRM : Residuals

daily_tsamax_pameseb_IRM_residuals = read.csv2("~/Desktop/sc_april_2019/tsamax daily Pameseb_IRM summary stats residuals.csv", sep = ",")
knitr::kable(daily_tsamax_pameseb_IRM_residuals)

pokyah/agrometeoR.extras documentation built on May 27, 2019, 2:07 p.m.