DM Track. Lesson 2.

code.R

# > install.packages("gridExtra")
require(gridExtra)
require(ggplot2)

# read kaggle data
stores <- read.csv("raw/stores.csv", header = T)
features <- read.csv("raw/features.csv", header = T)
train <- read.csv("raw/train.csv", header = T)

# create list of stores by splitting train data by store number
stores <- split(train, train$Store)
# > str(stores)
# List of 45
#  $ 1 :'data.frame':     10244 obs. of  5 variables:
#   ..$ Store       : int [1:10244] 1 1 1 1 1 1 1 1 1 1 ...
#   ..$ Dept        : int [1:10244] 1 1 1 1 1 1 1 1 1 1 ...
#   ..$ Date        : Factor w/ 143 levels "2010-02-05","2010-02-12",..: 1 2 3 4 5 6 7 8 9 10 ...
#   ..$ Weekly_Sales: num [1:10244] 24925 46039 41596 19404 21828 ...
#   ..$ IsHoliday   : logi [1:10244] FALSE TRUE FALSE FALSE FALSE FALSE ...
#  $ 2 :'data.frame':     10238 obs. of  5 variables:
#   ..$ Store       : int [1:10238] 2 2 2 2 2 2 2 2 2 2 ...
#   ..$ Dept        : int [1:10238] 1 1 1 1 1 1 1 1 1 1 ...
#   ..$ Date        : Factor w/ 143 levels "2010-02-05","2010-02-12",..: 1 2 3 4 5 6 7 8 9 10 ...
#   ..$ Weekly_Sales: num [1:10238] 35034 60484 58222 25962 27372 ...
#   ..$ IsHoliday   : logi [1:10238] FALSE TRUE FALSE FALSE FALSE FALSE ...
# ...

plotSalesOfStoreByNumWeek <- function(storeNum) {
	store <- stores[[storeNum]]

	store$Date <- as.Date(store$Date)
	store$WeekNum <- as.factor(format(store$Date,"%U"))
	store$Dept <- as.factor(store$Dept)

	#store.byDept <- ddply(store,~Dept,summarise,sum=sum(Weekly_Sales))
	#store.byDept.ordered <- store.byDept[order(-store.byDept$sum),]
	# store.byDate <- ddply(store,~Date,summarise,sum=sum(Weekly_Sales), IsHoliday=IsHoliday)

	# only holidays
	store.OnlyHolidays <- store[which(store$IsHoliday),]
	# summarize all departaments weekly sales for date
	store.OnlyHolidays.byDate <- ddply(store.OnlyHolidays, ~Date, summarise, Weekly_Sales=sum(Weekly_Sales))
	# add number of week
	store.OnlyHolidays.byDate$WeekNum <- as.factor(format(store.OnlyHolidays.byDate$Date,"%U"))
	#store.OnlyHolidays.byDate.byWeekNum <- ddply(store.OnlyHolidays.byDate, ~WeekNum, summarise, Weekly_Sales=sum(Weekly_Sales))
	# the same for weeks except holiday weeks
	store.WOHolidays <- store[which(!store$IsHoliday),]
	store.WOHolidays.byDate <- ddply(store.WOHolidays, ~Date, summarise, Weekly_Sales=sum(Weekly_Sales))
	store.WOHolidays.byDate$WeekNum <- as.factor(format(store.WOHolidays.byDate$Date,"%U"))
	#store.WOHolidays.byDate.byWeekNum <- ddply(store.WOHolidays.byDate, ~WeekNum, summarise, Weekly_Sales=sum(Weekly_Sales))

	plot <- ggplot(store.WOHolidays.byDate, aes(WeekNum, Weekly_Sales)) +	
	ylim(250000, 3500000) + 
		geom_point() +
		geom_point(data=store.OnlyHolidays.byDate, color='red') +
		ggtitle(paste("Store", storeNum))
	return(list(plot))
}

# draw
plots <- c()
for(i in 1:45) {
	plots <- append(plots, plotSalesOfStoreByNumWeek(i))
}

n <- length(plots)
nCol <- floor(sqrt(n))
png("stores_weekly_sales_by_week_number.png", 2000, 1000)
do.call("grid.arrange", c(plots, ncol=nCol))
dev.off()