5 v 9 Chunk Size

1 Objective

How does changing the chunk size effect the matching results? Does chunking, matching, and stitching data based on periods of 5 or 9 years matter?

2 Set Up

library(dplyr)
library(tidyr)
library(ggplot2)
library(knitr)
library(kableExtra)
# Note this will need to changed to run on your local machine 
BASE_DIR <- "/Users/dorh012/Documents/2021/stitches/notebooks/stitches_dev"
# Load the functions that we will use! 
source(file.path(BASE_DIR, "nearest_neighbor_matching.R"))
source(file.path(BASE_DIR, "stitching_functions.R"))
# Import the archive and target data from the two different periods. 
archive_data_5 <- read.csv(file.path(BASE_DIR, "inputs", "archive_data_5yrs.csv"), stringsAsFactors = FALSE) %>% 
  dplyr::filter(!experiment %in% c("ssp534-over", "ssp245"))
target_data_5 <- read.csv(file.path(BASE_DIR, "inputs", "target_data_5yrs.csv"), stringsAsFactors = FALSE)
archive_data_9 <- read.csv(file.path(BASE_DIR, "inputs", "archive_data_9yrs.csv"), stringsAsFactors = FALSE)  %>% 
  dplyr::filter(!experiment %in% c("ssp534-over", "ssp245"))
target_data_9 <- read.csv(file.path(BASE_DIR, "inputs", "target_data_9yrs.csv"), stringsAsFactors = FALSE)
# Import the data to be stiched together. 
tgav_data <- read.csv("inputs/main_raw_pasted_tgav_anomaly_all_pangeo_list_models.csv", 
                      stringsAsFactors = FALSE) %>% dplyr::select(-X)
  
  #read.csv(file.path(BASE_DIR, "inputs", "tgav_data.csv"), stringsAsFactors = FALSE)

3 Compare the inputs

What is the difference between the fx and dx relationship for the two different chunks?

ggplot() + 
  geom_point(data = target_data_5, aes(fx, dx, color = "5 yr chunks")) + 
  geom_point(data = target_data_9, aes(fx, dx, color = "9 yr chunks")) + 
  labs(title = "Target Data", subtitle = "comparing 5 and 9 yr chunks", 
       y = "dx (rate of change per chunk)", x = "fx (median value per chunk)") + 
  theme_bw()

ggplot() + 
  geom_point(data = archive_data_5, aes(fx, dx, color = "5 yr chunks")) + 
  geom_point(data = archive_data_9, aes(fx, dx, color = "9 yr chunks")) + 
  labs(title = "Archive Data", subtitle = "comparing 5 and 9 yr chunks", 
       y = "dx (rate of change per chunk)", x = "fx (median value per chunk)") + 
  theme_bw()

3.1 Match archive and target data

# Match the target and archive data with one another. 
match_5 <- match_nearest_neighbor(target_data = target_data_5, archive_data = archive_data_5)
match_9 <- match_nearest_neighbor(target_data = target_data_9, archive_data = archive_data_9)
match_5$chunk <- "5"
match_9$chunk <- "9"

# Join the data frames together 
d <- dplyr::bind_rows(match_5, match_9)
# Plot 
ggplot() + 
  # Add the data from the 5 yr 
  geom_point(data = d, aes(archive_fx, archive_dx,  
                                             color = "matched archive data")) + 
  geom_point(data = d, aes(target_fx, target_dx,  
                                             color = "target data"), alpha = 0.4) + 
  geom_segment(data = d, aes(x = target_fx, y = target_dx, 
                                               xend = archive_fx, yend = archive_dx), alpha = 0.4) +
  
  # Some asthetics 
  scale_color_manual(values = c("matched archive data" = "red", 
                               "target data" = "blue")) +
  facet_wrap("chunk") +
  theme_bw() + 
  labs(y = "dx (rate of change per chunk)", 
       x = "fx (value of median time per chunk)", 
       title = "Compare target with matched archive values", 
       subtitle = "comparing 5 and 9 yr chunks" )

pivot_longer(d, cols = c("dist_dx", "dist_fx", "dist_l2"), names_to = "distance", values_to = "value") %>% 
  ggplot() + 
  geom_dotplot(aes(value, fill = chunk), alpha = 0.5) + 
  theme_bw() + 
  facet_wrap("distance") + 
  labs(title = "Compare Match Error", x = "distance")

4 Match Data