analyses/deep/06_prediction_performance.R
In nmouquet/RLS_AESTHE: Research Compendium
###################################################################################################
#' Vizualize the performances of the deep learning model
#'
#'Produces Figure 1 panel b
#'
#' @author Juliette Langlois, \email{juliette.a.langlois@@gmail.com},
#' Nicolas Mouquet, \email{nicolas.mouquet@@cnrs.fr},
#' Valentine Fleure \email{valentine.fleure@@gmail.com},
#' Julien Renoult \email{jurenoult@@gmail.com}
#'
#'
#' @date 2021/06/29 first created
##################################################################################################
# Load data and functions ----
source(here::here("R", "functions_elo.R"))
all_table <- read.csv(here::here(res_dir_elo, "02_esthe_focus.csv"))
learn_table <- read.csv(here::here(res_dir_elo, "02_deep_train.csv"))
pred_table <- read.csv(here::here(res_dir_deep, "04_esthe_pred50.csv"))
# variation of the MSE across the epochs
run1 <- read.csv(here::here(res_dir_deep, "training", "run1.csv"), header = FALSE)
run2 <- read.csv(here::here(res_dir_deep, "training", "run2.csv"), header = FALSE)
run3 <- read.csv(here::here(res_dir_deep, "training", "run3.csv"), header = FALSE)
run4 <- read.csv(here::here(res_dir_deep, "training", "run4.csv"), header = FALSE)
run5 <- read.csv(here::here(res_dir_deep, "training", "run5.csv"), header = FALSE)
# performances on the validation sets
val1 <- read.csv(here::here(res_dir_deep, "validation_score", "validation_score1.txt"))
val2 <- read.csv(here::here(res_dir_deep, "validation_score", "validation_score2.txt"))
val3 <- read.csv(here::here(res_dir_deep, "validation_score", "validation_score3.txt"))
val4 <- read.csv(here::here(res_dir_deep, "validation_score", "validation_score4.txt"))
val5 <- read.csv(here::here(res_dir_deep, "validation_score", "validation_score5.txt"))
# ----
# R squared on the validation sets ----
colnames(val1)
val1$image_name <- gsub(".png", "", val1$image_name)
for(i in 1:nrow(val1)){
val1$evaluated_score[i] <- learn_table$esthe_all[which(learn_table$name_worms %in% val1$image_name[i])]
}
mod1 <- lm(val1$predicted_score~val1$evaluated_score)
sum1 <- summary(mod1)
Rsq1 <- sum1$r.squared
val2$image_name <- gsub(".png", "", val2$image_name)
for(i in 1:nrow(val2)){
val2$evaluated_score[i] <- learn_table$esthe_all[which(learn_table$name_worms %in% val2$image_name[i])]
}
mod2 <- lm(predicted_score~evaluated_score, data = val2)
sum2 <- summary(mod2)
Rsq2 <- sum2$r.squared
val3$image_name <- gsub(".png", "", val3$image_name)
for(i in 1:nrow(val3)){
val3$evaluated_score[i] <- learn_table$esthe_all[which(learn_table$name_worms %in% val3$image_name[i])]
}
mod3 <- lm(predicted_score~evaluated_score, data = val3)
sum3 <- summary(mod3)
Rsq3 <- sum3$r.squared
val4$image_name <- gsub(".png", "", val4$image_name)
for(i in 1:nrow(val4)){
val4$evaluated_score[i] <- learn_table$esthe_all[which(learn_table$name_worms %in% val4$image_name[i])]
}
mod4 <- lm(predicted_score~evaluated_score, data = val4)
sum4 <- summary(mod4)
Rsq4 <- sum4$r.squared
val5$image_name <- gsub(".png", "", val5$image_name)
for(i in 1:nrow(val5)){
val5$evaluated_score[i] <- learn_table$esthe_all[which(learn_table$name_worms %in% val5$image_name[i])]
}
mod5 <- lm(predicted_score~evaluated_score, data = val5)
sum5 <- summary(mod5)
Rsq5 <- sum5$r.squared
mean(Rsq1, Rsq2, Rsq3, Rsq4, Rsq5)
sd(x = c(Rsq1, Rsq2, Rsq3, Rsq4, Rsq5))
# ----
# FIGURE 1b side left ----
val1$set <- "1"
val2$set <- "2"
val3$set <- "3"
val4$set <- "4"
val5$set <- "5"
val_all <- rbind(val1, val2, val3, val4, val5)
perfplot <-
ggplot2::ggplot(data = val_all, ggplot2::aes(x = evaluated_score, y = predicted_score, col = set)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8) +
ggplot2::theme_light() +
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val_all$predicted_score), y = range(val_all$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
labels = c("1250", "1500", "1750", "2000")) +
ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = 14, family = "serif"),
axis.text = ggplot2::element_text(size = 9, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
# ----
# Clean data ----
# remove path from image name for predicted values
pred_table$image_name <- gsub(pattern = "../../../../data/images/png/", replacement = "",
x = pred_table$image_name)
pred_table$image_name <- gsub(pattern = ".png", replacement = "", x = pred_table$image_name)
colnames(pred_table) <- c("name_worms", "esthe_pred")
# change column names of table of evaluated values
eval_table <- learn_table[, c("name_worms", "esthe_all")]
colnames(eval_table) <- c("name_worms", "esthe_pred")
eval_table$method <- "evaluated"
# remove the images evaluated from the predicted table
pred_table <- pred_table[-which(pred_table$name_worms %in% eval_table$name_worms),]
range(pred_table$esthe_pred)
pred_table$method <- "predicted"
# combine the evaluated and predicted tables
scores_table <- rbind(pred_table, eval_table)
# Add predicted score to the table with everything and save
all_table <- merge(all_table, scores_table, by = "name_worms")
write.csv(all_table, here::here(res_dir_deep, "06_esthe_focus.csv"), row.names = FALSE)
# ----
# Predicted score VS the evaluated (FIGURE S1 M) -----
#(for five folds cross validation and the 481 images used in the questionaires)
textsize <- 2
axissize1 <- 8
axissize2 <- 4
perfplot_val1 <-
ggplot2::ggplot(data = val1, ggplot2::aes(x = evaluated_score, y = predicted_score)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=1) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val1$predicted_score), y = range(val1$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("a) Cross-validation fold #1 (n=",dim(val1)[1],")"),
color="black",hjust=0,size=textsize)+
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
perfplot_val2 <-
ggplot2::ggplot(data = val2, ggplot2::aes(x = evaluated_score, y = predicted_score)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=2) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val2$predicted_score), y = range(val2$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("b) Cross-validation fold #2 (n=",dim(val2)[1],")"),
color="black",hjust=0,size=textsize)+
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
perfplot_val3 <-
ggplot2::ggplot(data = val3, ggplot2::aes(x = evaluated_score, y = predicted_score)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=3) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val3$predicted_score), y = range(val3$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("c) Cross-validation fold #3 (n=",dim(val3)[1],")"),
color="black",hjust=0,size=textsize)+
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
perfplot_val4 <-
ggplot2::ggplot(data = val4, ggplot2::aes(x = evaluated_score, y = predicted_score)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=4) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val4$predicted_score), y = range(val4$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("d) Cross-validation fold #4 (n=",dim(val4)[1],")"),
color="black",hjust=0,size=textsize)+
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
perfplot_val5 <-
ggplot2::ggplot(data = val5, ggplot2::aes(x = evaluated_score, y = predicted_score)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=5) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val5$predicted_score), y = range(val5$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("e) Cross-validation fold #5 (n=",dim(val5)[1],")"),
color="black",hjust=0,size=textsize)+
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
eval_pred <- all_table[!is.na(all_table$esthe_all),]
perfplot_all <- ggplot2::ggplot(data = eval_pred, ggplot2::aes(x = esthe_all, y = esthe_pred)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=6) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(eval_pred$predicted_score), y = range(eval_pred$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values")+
#ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("f) All training set (n=",dim(eval_pred)[1],")"),
color="black",hjust=0,size=textsize)+
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
plot <- gridExtra::grid.arrange(perfplot_val1,perfplot_val2,perfplot_val3,perfplot_val4,perfplot_val5,perfplot_all)
ggplot2::ggsave(plot = plot, filename = here::here("figures_tables", "FIGURE_M.jpg"),
width = 15, height = 15, units = "cm", dpi = 600)
# ----
# Distribution of the predicted score VS the evaluated ones
# for train+val+test (FIGURE 1b side right) ----
# Plot the densities
ecrtyp <- max(pred_table$esthe_pred)-min(pred_table$esthe_pred)
points_toshow <- tibble::tibble(x = c(min(pred_table$esthe_pred), min(pred_table$esthe_pred)+ecrtyp/11,
min(pred_table$esthe_pred)+2*ecrtyp/11, min(pred_table$esthe_pred)+3*ecrtyp/11,
min(pred_table$esthe_pred)+4*ecrtyp/11, min(pred_table$esthe_pred)+5*ecrtyp/11,
min(pred_table$esthe_pred)+6*ecrtyp/11, min(pred_table$esthe_pred)+7*ecrtyp/11,
min(pred_table$esthe_pred)+8*ecrtyp/11, min(pred_table$esthe_pred)+9*ecrtyp/11,
min(pred_table$esthe_pred)+10*ecrtyp/11, max(pred_table$esthe_pred)), y = 0)
par(mar = c(0,0,0,0), family = "serif")
densplot <-
ggplot2::ggplot(all_table, ggplot2::aes(x = esthe_pred, color = method, fill = method)) +
ggplot2::geom_density(alpha = 0.5) +
ggplot2::theme_light() +
ggplot2::xlab("Aesthetic values") +
ggplot2::ylab("Density") +
ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000)) +
ggplot2::scale_fill_manual(values = c(colors[10], colors[8])) +
ggplot2::scale_color_manual(values = c(colors[10], colors[8])) +
ggplot2::theme(axis.title = ggplot2::element_text(size = 14, family = "serif"),
axis.text = ggplot2::element_text(size = 9, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none") +
# Add the 12 points to show were the photos come from
ggplot2::geom_point(data = points_toshow, ggplot2::aes(x, y),
inherit.aes = FALSE,
size = 3,
color = colors[9]) +
ggplot2::annotate(geom = "text", label = paste0("Predicted (n=",length(which(all_table$method == "predicted")),")"),
x = 1550 , y = 0.00272, family = "serif",
color = colors[10], size = 4.5, hjust = 0) +
ggplot2::annotate(geom = "text", label = paste0("Evaluated (n=",length(which(all_table$method == "evaluated")),")"),
x = 1550 , y = 0.00248, family = "serif",
color = colors[8], size = 4.5, hjust = 0)
# ----
# Gather the two plots (FIGURE 1b) ----
plot <- ggpubr::ggarrange(perfplot, densplot,
ncol = 2, nrow = 1)
# Save
ggplot2::ggsave(plot = plot, filename = here::here("figures_tables", "FIGURE_1b.jpg"),
width = 20, height = 10, units = "cm", dpi = 600, family = "serif")
rm(densplot, all_table, eval, points_toshow, ecrtyp, learn_table, mod1, mod2, mod3, mod4, mod5,
perfplot, plot, pred_table, run1, run2, run3, run4, run5, scores_table, sum1, sum2, sum3,
sum4, sum5, val_all, val1, val2, val3, val4, val5, i, Rsq1, Rsq2, Rsq3, Rsq4, Rsq5, eval_table)
# ----
nmouquet/RLS_AESTHE documentation built on May 9, 2023, 5:45 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
###################################################################################################
#' Vizualize the performances of the deep learning model
#'
#'Produces Figure 1 panel b
#'
#' @author Juliette Langlois, \email{juliette.a.langlois@@gmail.com},
#' Nicolas Mouquet, \email{nicolas.mouquet@@cnrs.fr},
#' Valentine Fleure \email{valentine.fleure@@gmail.com},
#' Julien Renoult \email{jurenoult@@gmail.com}
#'
#'
#' @date 2021/06/29 first created
##################################################################################################
# Load data and functions ----
source(here::here("R", "functions_elo.R"))
all_table <- read.csv(here::here(res_dir_elo, "02_esthe_focus.csv"))
learn_table <- read.csv(here::here(res_dir_elo, "02_deep_train.csv"))
pred_table <- read.csv(here::here(res_dir_deep, "04_esthe_pred50.csv"))
# variation of the MSE across the epochs
run1 <- read.csv(here::here(res_dir_deep, "training", "run1.csv"), header = FALSE)
run2 <- read.csv(here::here(res_dir_deep, "training", "run2.csv"), header = FALSE)
run3 <- read.csv(here::here(res_dir_deep, "training", "run3.csv"), header = FALSE)
run4 <- read.csv(here::here(res_dir_deep, "training", "run4.csv"), header = FALSE)
run5 <- read.csv(here::here(res_dir_deep, "training", "run5.csv"), header = FALSE)
# performances on the validation sets
val1 <- read.csv(here::here(res_dir_deep, "validation_score", "validation_score1.txt"))
val2 <- read.csv(here::here(res_dir_deep, "validation_score", "validation_score2.txt"))
val3 <- read.csv(here::here(res_dir_deep, "validation_score", "validation_score3.txt"))
val4 <- read.csv(here::here(res_dir_deep, "validation_score", "validation_score4.txt"))
val5 <- read.csv(here::here(res_dir_deep, "validation_score", "validation_score5.txt"))
# ----
# R squared on the validation sets ----
colnames(val1)
val1$image_name <- gsub(".png", "", val1$image_name)
for(i in 1:nrow(val1)){
val1$evaluated_score[i] <- learn_table$esthe_all[which(learn_table$name_worms %in% val1$image_name[i])]
}
mod1 <- lm(val1$predicted_score~val1$evaluated_score)
sum1 <- summary(mod1)
Rsq1 <- sum1$r.squared
val2$image_name <- gsub(".png", "", val2$image_name)
for(i in 1:nrow(val2)){
val2$evaluated_score[i] <- learn_table$esthe_all[which(learn_table$name_worms %in% val2$image_name[i])]
}
mod2 <- lm(predicted_score~evaluated_score, data = val2)
sum2 <- summary(mod2)
Rsq2 <- sum2$r.squared
val3$image_name <- gsub(".png", "", val3$image_name)
for(i in 1:nrow(val3)){
val3$evaluated_score[i] <- learn_table$esthe_all[which(learn_table$name_worms %in% val3$image_name[i])]
}
mod3 <- lm(predicted_score~evaluated_score, data = val3)
sum3 <- summary(mod3)
Rsq3 <- sum3$r.squared
val4$image_name <- gsub(".png", "", val4$image_name)
for(i in 1:nrow(val4)){
val4$evaluated_score[i] <- learn_table$esthe_all[which(learn_table$name_worms %in% val4$image_name[i])]
}
mod4 <- lm(predicted_score~evaluated_score, data = val4)
sum4 <- summary(mod4)
Rsq4 <- sum4$r.squared
val5$image_name <- gsub(".png", "", val5$image_name)
for(i in 1:nrow(val5)){
val5$evaluated_score[i] <- learn_table$esthe_all[which(learn_table$name_worms %in% val5$image_name[i])]
}
mod5 <- lm(predicted_score~evaluated_score, data = val5)
sum5 <- summary(mod5)
Rsq5 <- sum5$r.squared
mean(Rsq1, Rsq2, Rsq3, Rsq4, Rsq5)
sd(x = c(Rsq1, Rsq2, Rsq3, Rsq4, Rsq5))
# ----
# FIGURE 1b side left ----
val1$set <- "1"
val2$set <- "2"
val3$set <- "3"
val4$set <- "4"
val5$set <- "5"
val_all <- rbind(val1, val2, val3, val4, val5)
perfplot <-
ggplot2::ggplot(data = val_all, ggplot2::aes(x = evaluated_score, y = predicted_score, col = set)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8) +
ggplot2::theme_light() +
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val_all$predicted_score), y = range(val_all$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
labels = c("1250", "1500", "1750", "2000")) +
ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = 14, family = "serif"),
axis.text = ggplot2::element_text(size = 9, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
# ----
# Clean data ----
# remove path from image name for predicted values
pred_table$image_name <- gsub(pattern = "../../../../data/images/png/", replacement = "",
x = pred_table$image_name)
pred_table$image_name <- gsub(pattern = ".png", replacement = "", x = pred_table$image_name)
colnames(pred_table) <- c("name_worms", "esthe_pred")
# change column names of table of evaluated values
eval_table <- learn_table[, c("name_worms", "esthe_all")]
colnames(eval_table) <- c("name_worms", "esthe_pred")
eval_table$method <- "evaluated"
# remove the images evaluated from the predicted table
pred_table <- pred_table[-which(pred_table$name_worms %in% eval_table$name_worms),]
range(pred_table$esthe_pred)
pred_table$method <- "predicted"
# combine the evaluated and predicted tables
scores_table <- rbind(pred_table, eval_table)
# Add predicted score to the table with everything and save
all_table <- merge(all_table, scores_table, by = "name_worms")
write.csv(all_table, here::here(res_dir_deep, "06_esthe_focus.csv"), row.names = FALSE)
# ----
# Predicted score VS the evaluated (FIGURE S1 M) -----
#(for five folds cross validation and the 481 images used in the questionaires)
textsize <- 2
axissize1 <- 8
axissize2 <- 4
perfplot_val1 <-
ggplot2::ggplot(data = val1, ggplot2::aes(x = evaluated_score, y = predicted_score)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=1) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val1$predicted_score), y = range(val1$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("a) Cross-validation fold #1 (n=",dim(val1)[1],")"),
color="black",hjust=0,size=textsize)+
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
perfplot_val2 <-
ggplot2::ggplot(data = val2, ggplot2::aes(x = evaluated_score, y = predicted_score)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=2) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val2$predicted_score), y = range(val2$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("b) Cross-validation fold #2 (n=",dim(val2)[1],")"),
color="black",hjust=0,size=textsize)+
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
perfplot_val3 <-
ggplot2::ggplot(data = val3, ggplot2::aes(x = evaluated_score, y = predicted_score)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=3) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val3$predicted_score), y = range(val3$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("c) Cross-validation fold #3 (n=",dim(val3)[1],")"),
color="black",hjust=0,size=textsize)+
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
perfplot_val4 <-
ggplot2::ggplot(data = val4, ggplot2::aes(x = evaluated_score, y = predicted_score)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=4) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val4$predicted_score), y = range(val4$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("d) Cross-validation fold #4 (n=",dim(val4)[1],")"),
color="black",hjust=0,size=textsize)+
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
perfplot_val5 <-
ggplot2::ggplot(data = val5, ggplot2::aes(x = evaluated_score, y = predicted_score)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=5) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(val5$predicted_score), y = range(val5$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values") +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("e) Cross-validation fold #5 (n=",dim(val5)[1],")"),
color="black",hjust=0,size=textsize)+
ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
eval_pred <- all_table[!is.na(all_table$esthe_all),]
perfplot_all <- ggplot2::ggplot(data = eval_pred, ggplot2::aes(x = esthe_all, y = esthe_pred)) +
ggplot2::geom_point(size = 2, shape = 20, alpha = 0.8,col=6) +
ggplot2::theme_light() +
ylim(1150, 1750)+xlim(1150, 1800)+
ggplot2::geom_smooth(method = "lm", se = TRUE, col = "black", lty = "solid", alpha = 0.80) +
ggplot2::expand_limits(x = range(eval_pred$predicted_score), y = range(eval_pred$predicted_score)) +
ggplot2::xlab("Evaluated aesthetic values") +
ggplot2::ylab("Predicted aesthetic values")+
#ggplot2::scale_color_manual(values = wes_palette("Darjeeling1")) +
ggplot2::annotate(geom="text", x=1150, y=1750, label=paste0("f) All training set (n=",dim(eval_pred)[1],")"),
color="black",hjust=0,size=textsize)+
#ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
#ggplot2::scale_y_continuous(breaks = c(1250, 1500, 1750, 2000),
# labels = c("1250", "1500", "1750", "2000")) +
ggplot2::theme(axis.title = ggplot2::element_text(size = axissize1, family = "serif"),
axis.text = ggplot2::element_text(size = axissize2, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none")
plot <- gridExtra::grid.arrange(perfplot_val1,perfplot_val2,perfplot_val3,perfplot_val4,perfplot_val5,perfplot_all)
ggplot2::ggsave(plot = plot, filename = here::here("figures_tables", "FIGURE_M.jpg"),
width = 15, height = 15, units = "cm", dpi = 600)
# ----
# Distribution of the predicted score VS the evaluated ones
# for train+val+test (FIGURE 1b side right) ----
# Plot the densities
ecrtyp <- max(pred_table$esthe_pred)-min(pred_table$esthe_pred)
points_toshow <- tibble::tibble(x = c(min(pred_table$esthe_pred), min(pred_table$esthe_pred)+ecrtyp/11,
min(pred_table$esthe_pred)+2*ecrtyp/11, min(pred_table$esthe_pred)+3*ecrtyp/11,
min(pred_table$esthe_pred)+4*ecrtyp/11, min(pred_table$esthe_pred)+5*ecrtyp/11,
min(pred_table$esthe_pred)+6*ecrtyp/11, min(pred_table$esthe_pred)+7*ecrtyp/11,
min(pred_table$esthe_pred)+8*ecrtyp/11, min(pred_table$esthe_pred)+9*ecrtyp/11,
min(pred_table$esthe_pred)+10*ecrtyp/11, max(pred_table$esthe_pred)), y = 0)
par(mar = c(0,0,0,0), family = "serif")
densplot <-
ggplot2::ggplot(all_table, ggplot2::aes(x = esthe_pred, color = method, fill = method)) +
ggplot2::geom_density(alpha = 0.5) +
ggplot2::theme_light() +
ggplot2::xlab("Aesthetic values") +
ggplot2::ylab("Density") +
ggplot2::scale_x_continuous(breaks = c(1250, 1500, 1750, 2000)) +
ggplot2::scale_fill_manual(values = c(colors[10], colors[8])) +
ggplot2::scale_color_manual(values = c(colors[10], colors[8])) +
ggplot2::theme(axis.title = ggplot2::element_text(size = 14, family = "serif"),
axis.text = ggplot2::element_text(size = 9, family = "serif"),
panel.grid = ggplot2::element_blank(),
legend.position = "none") +
# Add the 12 points to show were the photos come from
ggplot2::geom_point(data = points_toshow, ggplot2::aes(x, y),
inherit.aes = FALSE,
size = 3,
color = colors[9]) +
ggplot2::annotate(geom = "text", label = paste0("Predicted (n=",length(which(all_table$method == "predicted")),")"),
x = 1550 , y = 0.00272, family = "serif",
color = colors[10], size = 4.5, hjust = 0) +
ggplot2::annotate(geom = "text", label = paste0("Evaluated (n=",length(which(all_table$method == "evaluated")),")"),
x = 1550 , y = 0.00248, family = "serif",
color = colors[8], size = 4.5, hjust = 0)
# ----
# Gather the two plots (FIGURE 1b) ----
plot <- ggpubr::ggarrange(perfplot, densplot,
ncol = 2, nrow = 1)
# Save
ggplot2::ggsave(plot = plot, filename = here::here("figures_tables", "FIGURE_1b.jpg"),
width = 20, height = 10, units = "cm", dpi = 600, family = "serif")
rm(densplot, all_table, eval, points_toshow, ecrtyp, learn_table, mod1, mod2, mod3, mod4, mod5,
perfplot, plot, pred_table, run1, run2, run3, run4, run5, scores_table, sum1, sum2, sum3,
sum4, sum5, val_all, val1, val2, val3, val4, val5, i, Rsq1, Rsq2, Rsq3, Rsq4, Rsq5, eval_table)
# ----
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.