1 - Code for reading in the dataset and processing the data
Let's first download and read the data into a data frame
temp <- tempfile()
download.file("https://d396qusza40orc.cloudfront.net/repdata%2Fdata%2Factivity.zip",temp)
activity <- read.csv(unz(temp, "activity.csv"))
unlink(temp)Some fields have undesired data format:
str(activity)## 'data.frame': 17568 obs. of 3 variables:
## $ steps : int NA NA NA NA NA NA NA NA NA NA ...
## $ date : Factor w/ 61 levels "2012-10-01","2012-10-02",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ interval: int 0 5 10 15 20 25 30 35 40 45 ...
Let's convert interval variable into factor and create a variable hour based on intervals:
library(plyr)
activity$hour<-as.factor(round_any((as.numeric(activity$interval)+1)/100, 1, f = ceiling)) # create a variable for each hour
activity$interval<-as.factor(activity$interval)2 - Histogram of the total number of steps taken each day
library(ggplot2)
ggplot(data = activity,
aes(as.Date(as.character(activity$date), "%Y-%m-%d"), steps)) +
stat_summary(fun.y = sum, # adds up all observations
geom = "bar") +
xlab ("Dates") +
ggtitle("Total number of steps taken each day")## Warning: Removed 2304 rows containing non-finite values (stat_summary).
3 - Mean and median number of steps taken each day
Let's create a data frame with means and medians for each day (keeping NAs for now)
means<-ddply(activity, "date", summarise, mean = mean(steps))
medians<-ddply(activity, "date", summarise, median = median(steps))
means_medians<-merge(means,medians)All medians are 0:
## date mean median
## 1 2012-10-01 NA NA
## 2 2012-10-02 0.4375000 0
## 3 2012-10-03 39.4166667 0
## 4 2012-10-04 42.0694444 0
## 5 2012-10-05 46.1597222 0
## 6 2012-10-06 53.5416667 0
## 7 2012-10-07 38.2465278 0
## 8 2012-10-08 NA NA
## 9 2012-10-09 44.4826389 0
## 10 2012-10-10 34.3750000 0
## 11 2012-10-11 35.7777778 0
## 12 2012-10-12 60.3541667 0
## 13 2012-10-13 43.1458333 0
## 14 2012-10-14 52.4236111 0
## 15 2012-10-15 35.2048611 0
## 16 2012-10-16 52.3750000 0
## 17 2012-10-17 46.7083333 0
## 18 2012-10-18 34.9166667 0
## 19 2012-10-19 41.0729167 0
## 20 2012-10-20 36.0937500 0
## 21 2012-10-21 30.6284722 0
## 22 2012-10-22 46.7361111 0
## 23 2012-10-23 30.9652778 0
## 24 2012-10-24 29.0104167 0
## 25 2012-10-25 8.6527778 0
## 26 2012-10-26 23.5347222 0
## 27 2012-10-27 35.1354167 0
## 28 2012-10-28 39.7847222 0
## 29 2012-10-29 17.4236111 0
## 30 2012-10-30 34.0937500 0
## 31 2012-10-31 53.5208333 0
## 32 2012-11-01 NA NA
## 33 2012-11-02 36.8055556 0
## 34 2012-11-03 36.7048611 0
## 35 2012-11-04 NA NA
## 36 2012-11-05 36.2465278 0
## 37 2012-11-06 28.9375000 0
## 38 2012-11-07 44.7326389 0
## 39 2012-11-08 11.1770833 0
## 40 2012-11-09 NA NA
## 41 2012-11-10 NA NA
## 42 2012-11-11 43.7777778 0
## 43 2012-11-12 37.3784722 0
## 44 2012-11-13 25.4722222 0
## 45 2012-11-14 NA NA
## 46 2012-11-15 0.1423611 0
## 47 2012-11-16 18.8923611 0
## 48 2012-11-17 49.7881944 0
## 49 2012-11-18 52.4652778 0
## 50 2012-11-19 30.6979167 0
## 51 2012-11-20 15.5277778 0
## 52 2012-11-21 44.3993056 0
## 53 2012-11-22 70.9270833 0
## 54 2012-11-23 73.5902778 0
## 55 2012-11-24 50.2708333 0
## 56 2012-11-25 41.0902778 0
## 57 2012-11-26 38.7569444 0
## 58 2012-11-27 47.3819444 0
## 59 2012-11-28 35.3576389 0
## 60 2012-11-29 24.4687500 0
## 61 2012-11-30 NA NA
That happens because there are no steps in more than 50% of intervals.
4 - Time series plot of the average number of steps taken
Let's visualize avarage number of steps taken (ignoring NAs for now):
ggplot(means_medians, aes(y=mean, x=as.Date(as.character(means_medians$date), "%Y-%m-%d"))) +
geom_bar(stat='identity') +
ylab ("Average daily steps") +
xlab ("Dates") +
ggtitle("Average number of steps taken")## Warning: Removed 8 rows containing missing values (position_stack).
5 - The 5-minute interval that, on average, contains the maximum number of steps
meansbyintervals<-ddply(activity, "interval", summarise, mean = mean(steps,na.rm = TRUE))
meansbyintervals<-arrange(meansbyintervals, desc(mean))
meansbyintervals[1,]## interval mean
## 1 835 206.1698
The most walkable interval is 8:35am
6 - Code to describe and show a strategy for imputing missing data
Let's see at the missing values
NAs <- is.na(activity$steps)
sum(NAs, na.rm = T)## [1] 2304
mean(NAs, na.rm = T)## [1] 0.1311475
13% is a rather high number. Let's see if there any particular days or intervals where there are missing values:
dates <- activity$date
dates <- as.Date(as.character(dates), "%Y-%m-%d")
hist(dates[NAs], "day", , xlab = "Date"
, main = "Histogram: what dates are missing")
Sounds like that missing values are spread between 8 out of 61 days. Not too bad. Let's remove these observations:
completeactivity <- activity[complete.cases(activity), ]7 - Histogram of the total number of steps taken each day after missing values are imputed
The plot obviously doesn't change much:
ggplot(data = completeactivity,
aes(as.Date(as.character(completeactivity$date), "%Y-%m-%d"), steps)) +
stat_summary(fun.y = sum,
geom = "bar") +
xlab ("Dates") +
ggtitle("total number of steps taken each day after missing values are imputed")
8 - Panel plot comparing the average number of steps taken per 5-minute interval across weekdays and weekends
Let's first do some formatting and define weekdays and weekends
library(timeDate)
completeactivity$date<-as.Date(as.character(completeactivity$date), "%Y-%m-%d")
completeactivity$weekday<-isWeekday(completeactivity$date)
completeactivity$weekday[completeactivity$weekday==TRUE] <- "Weekday"
completeactivity$weekday[completeactivity$weekday==FALSE] <- "Weekend"
completeactivity$weekday<-as.factor(completeactivity$weekday)Create a new data frame with means by weekdays and weekends and by hourly intervals:
completeactivitymeans<-aggregate(completeactivity$steps
, by=list(completeactivity$hour,completeactivity$weekday)
, FUN=mean)
names(completeactivitymeans)<-c("hour","day","steps")Finally, build a plot of means:
ggplot(data = completeactivitymeans,
aes(hour, steps)) +
geom_point(aes(color=hour)) +
facet_grid(day~.) +
ggtitle("Means by hourly intervals")
And other statistics:
ggplot(data = completeactivity,
aes(hour, steps)) +
geom_boxplot(aes(color=hour)) +
facet_grid(weekday~.) +
ggtitle("Statistics by hourly intervals")
Weekday and Weekend patters are different
-
Wakes up later on weekends
-
Average amount of steps per interval is higher during weekends
-
Walking on weekends is more popular
Other data caviats and findings
-
8 days are missing oobeservations (forgot to track?)
-
The excercise showed that exploring daily aggregated or hourly aggregated data would make more sense - interval data is skewed because there are a lot of 0 steps as well as intervals with very high numbers of steps
Thank you for the review!