💾 Archived View for gemlog.blue › users › cariboudatascience › 1614053800.gmi captured on 2022-06-03 at 23:12:05. Gemini links have been rewritten to link to archived content
⬅️ Previous capture (2021-12-04)
-=-=-=-=-=-=-
library(ggplot2) library(scales) library(dplyr) library(tidyr) library(gganimate) library(gapminder) library(ggthemes) rm(list=ls()) fruit_type <- read.csv("./DATA/3-fruit-consumption-by-fruit-type.csv") colnames(fruit_type) <-c("Country","Code","Year", "Apples","Bananas","Citrus", "Dates","Other","Grapefruit", "Grapes","Lemons","Oranges", "Pineapples","Plantains") fruit_type <- fruit_type %>% select(Country:Dates,Grapefruit:Plantains,Other) summary(fruit_type) ### Dplyr::pivot_lionger fruit_type_long <- fruit_type %>% pivot_longer(cols = Apples:Other) %>% rename("Fruit" = "name") %>% rename("perCapita" = "value") %>% select(-Code) summary(fruit_type_long) ### PLot of Fruit Consumption Wroldwide by Fruit Type fruit_type_long %>% filter(Country =="World" & Fruit !="Other") %>% ggplot() + geom_line(aes(x=Year,y=perCapita,col=Fruit),size=1.5) + labs(title="Wroldwide: Fruit Consumption per Person by Fruit and Year", caption = "(Average fruit consumption per person, differentiated by fruit types, measured in Kilograms per Year)", y="Kilograms per Person") + guides(color = guide_legend(override.aes = list(size = 2))) + scale_y_continuous( labels = scales::number_format(accuracy = 0.01, decimal.mark = '.'))
fruit_type_long %>% filter(Country =="United States" & Fruit !="Other") %>% ggplot() + geom_line(aes(x=Year,y=perCapita,col=Fruit,),size=1.5) + labs(title="United States:Fruit Consumption per Person by Fruit and Year", caption = "(Average fruit consumption per person, differentiated by fruit types, measured in Kilograms per Year)", y="Kilograms per Person") + guides(color = guide_legend(override.aes = list(size = 2))) + scale_y_continuous( labels = scales::number_format(accuracy = 0.01, decimal.mark = '.'))
Worldwide Fruit Consumption By Fruit Type
US Fruit Consumption by Fruit Type
fruit <- read.csv("./DATA/1-fruit-consumption-per-capita.csv") fruit <- fruit %>%select(-Code) colnames(fruit) <- c("Country","Year","Fruits") fruits_top <- fruit %>% filter(Year =="2017") %>% top_n(5,Fruits) fruits_top <- as.data.frame(fruits_top) countries_five <- fruits_top %>% select(Country) %>% left_join(fruit_type_long,by="Country") %>% na.omit() head(countries_five) tail(countries_five) ggplot(countries_five) + geom_line(aes(x=Year,y=perCapita,col=Fruit),size=1.5) + facet_wrap(~Country,scale="free_y",ncol=2) + labs(title="Top 5 Countries by Fruit") countries_five %>%filter(Year >="2007") %>% ggplot() + geom_col(aes(x=reorder(Fruit,perCapita),y=perCapita)) + facet_wrap(~Year,scale="free_y",ncol=2) + coord_flip() + labs(title="Fruit Consumption by Year",y="Kilograms/Person/Year")
Top 5 Countries by Fruit Type.
R Language and Rtools40 (click and "base" and "Rtools")