Skip to content
Projects
Groups
Snippets
Help
This project
Loading...
Sign in / Register
Toggle navigation
tutorials
Overview
Overview
Details
Activity
Cycle Analytics
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Charts
Issues
0
Issues
0
List
Board
Labels
Milestones
Merge Requests
0
Merge Requests
0
CI / CD
CI / CD
Pipelines
Jobs
Schedules
Charts
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Charts
Create a new issue
Jobs
Commits
Issue Boards
Open sidebar
saeid
tutorials
Commits
95aea55c
Unverified
Commit
95aea55c
authored
Mar 13, 2018
by
Arham Akheel
Committed by
GitHub
Mar 13, 2018
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
Update Introduction to dplyr
parent
dce82d50
Show whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
111 additions
and
5 deletions
+111
-5
Introduction to dplyr
Introduction to dplyr
+111
-5
No files found.
Introduction to dplyr
View file @
95aea55c
wine = read.csv('wine.csv', stringsAsFactors = F, encoding = 'UTF-8')
################################################################################
install.packages('dplyr')
## This code is property of Data Science Dojo
install.packages('ggplot2')
## Copyright (C) 2017~2018
##
## Objective: Manipulate and visualize data using R
## Please install "dplyr" package: install.packages("dplyr")
## Please install "ggplot2" package: install.packages("ggplot2")
################################################################################
# Script for following along in Introduction to dplyr
# Copy-paste line by line or use the "Run" button in R Studio
#Set the working directory, example: setwd("directory/dataset folder")
install.packages("dplyr")
install.packages("ggplot2")
library(dplyr)
library(dplyr)
library(ggplot2)
library(ggplot2)
wine = wine[,-1]
wine = wine %>% select(-c(description))
#Reading the dataset from the working directory.
#Setting string values as characters
#loading the greek characters
wine = read.csv("wine.csv", stringsAsFactors = FALSE, encoding = 'UTF-8')
View (wine)
#Removing columns from dataset
wine = wine[,-c(1,3)]
#Creating a dataset by counting all observations grouped by country and then creating a new variable called count
wine %>% group_by(country) %>% summarize(count=n()) %>% arrange(desc(count))
wine %>% group_by(country) %>% summarize(count=n()) %>% arrange(desc(count))
#Creating a new variable which contains the top 10 countries
selected_countries = wine %>% group_by(country) %>% summarize(count=n()) %>% arrange(desc(count)) %>% top_n(10) %>% select(country)
selected_countries = wine %>% group_by(country) %>% summarize(count=n()) %>% arrange(desc(count)) %>% top_n(10) %>% select(country)
selected_countries
#Changing the format from data frame to vector as.character referencing the country column
selected_countries = as.character(selected_countries$country)
selected_countries = as.character(selected_countries$country)
class(selected_countries)
#Subsetting data selecting top ten countries and their points from wine
select_points=wine %>% filter(country %in% selected_countries) %>% select(country, points) %>% arrange(country)
select_points=wine %>% filter(country %in% selected_countries) %>% select(country, points) %>% arrange(country)
#Scatterplot with smooth line
ggplot(wine, aes(points,price)) + geom_point() + geom_smooth()
ggplot(wine, aes(points,price)) + geom_point() + geom_smooth()
#Boxplot between country and points, reordered by median of points. Center aligning the Title of the boxplot
ggplot(select_points, aes(x=reorder(country,points,median),y=points)) + geom_boxplot(aes(fill=country)) + xlab("Country") + ylab("Points") + ggtitle("Distribution of Top 10 Wine Producing Countries") + theme(plot.title = element_text(hjust = 0.5))
ggplot(select_points, aes(x=reorder(country,points,median),y=points)) + geom_boxplot(aes(fill=country)) + xlab("Country") + ylab("Points") + ggtitle("Distribution of Top 10 Wine Producing Countries") + theme(plot.title = element_text(hjust = 0.5))
#Filter by countries that do not appear on the selected_countries dataset
wine %>% filter(!(country %in% selected_countries)) %>% group_by(country) %>% summarize(median=median(points)) %>% arrange(desc(median))
wine %>% filter(!(country %in% selected_countries)) %>% group_by(country) %>% summarize(median=median(points)) %>% arrange(desc(median))
#Creating a new variable called top using country and points to rate them based on points
top=wine %>% group_by(country) %>% summarize(median=median(points)) %>% arrange(desc(median))
top=wine %>% group_by(country) %>% summarize(median=median(points)) %>% arrange(desc(median))
class(top)
#Changing the format from data frame to vector as.character referencing the country column
top=as.character(top$country)
top=as.character(top$country)
top
#Using intersect function to select the common values in both datasets
both=intersect(top,selected_countries)
both=intersect(top,selected_countries)
both
#Using setdiff to select the non-overlapping values in both datasets
not = setdiff(top, selected_countries)
not
#Creating a subset based on variety using group by and summarize
topwine = wine %>% group_by(variety) %>% summarize(number=n()) %>% arrange(desc(number)) %>% top_n(10)
topwine = wine %>% group_by(variety) %>% summarize(number=n()) %>% arrange(desc(number)) %>% top_n(10)
topwine=as.character(topwine$variety)
topwine=as.character(topwine$variety)
topwine
#Plot based on variety and points using group by and summarize
wine %>% filter(variety %in% topwine) %>% group_by(variety)%>% summarize(median=median(points)) %>% ggplot(aes(reorder(variety,median),median)) + geom_col(aes(fill=variety)) + xlab('Variety') + ylab('Median Point') + scale_x_discrete(labels=abbreviate)
wine %>% filter(variety %in% topwine) %>% group_by(variety)%>% summarize(median=median(points)) %>% ggplot(aes(reorder(variety,median),median)) + geom_col(aes(fill=variety)) + xlab('Variety') + ylab('Median Point') + scale_x_discrete(labels=abbreviate)
#Creating top 15 percent cheapest wines with high rating using intersect function
top15percent=wine %>% arrange(desc(points)) %>% filter(points > quantile(points, prob = 0.85))
top15percent=wine %>% arrange(desc(points)) %>% filter(points > quantile(points, prob = 0.85))
cheapest15percent=wine %>% arrange(price) %>% head(nrow(top15percent))
cheapest15percent=wine %>% arrange(price) %>% head(nrow(top15percent))
goodvalue = intersect(top15percent,cheapest15percent)
goodvalue = intersect(top15percent,cheapest15percent)
goodvalue
goodvalue
#Feature Engineering
wine = read.csv('wine.csv', stringsAsFactors = FALSE, encoding = 'UTF-8')
save(wine, file = "wine.rda")
load("wine.rda")
#Omiting one column from the wine dataset
wine = wine[,-c(3)]
View(wine)
#Using transmute and mutate functions to append a new column
wine1 = wine %>% mutate(PPratio = points/price)
wine2 = wine %>% transmute(PPratio = points/price)
#Aggregation by country using group by and summarize
wine %>% group_by(country) %>% summarize(total = n())
#Missing country values
wine[wine$country == "",]
#Adding missing values in the dataset
wine$country = ifelse(wine$designation == "Askitikos", "Greece", wine$country)
wine$country = ifelse(wine$designation == "Piedra Feliz", "Chile", wine$country)
wine$country = ifelse(wine$variety == "Red Blend", "Turkey", wine$country)
#Combining Datasets
#Creating a new subset by total number of rows by country
newwine = wine %>% group_by(country) %>% summarize(total = n()) %>% arrange(desc(total))
#Creating subsets with the head of wine and newwine
subset1=head(wine)
subset2=head(newwine)
#Combining two data frames using full join function
full = full_join(subset1, subset2)
full
#Combining two data frames using inner join function
inner = inner_join(subset1, subset2)
inner
#Combining two data frames using left join function
left = left_join(subset1, subset2)
left
#Combining two data frames using right join function
right = right_join(subset1, subset2)
right
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment