run.R

library(xts)
library(lubridate)
library(dplyr)
library(tidyr)
library(ggplot2)
library(scales)
# library(devtools)
# install_github("jrnold/ggthemes")
library(ggthemes)
library(pander)

activity <- "data/activity.csv" %>%
    read.csv(stringsAsFactors = FALSE) %>%
    tbl_df %>%
    mutate(
        hour = sprintf("%04d", interval),
        hour.posix = ymd_hm(paste("1900-01-01", hour)),
        time = ymd_hm(paste(date, hour)),
        weekday = wday(time)
    )

activity.daily <- activity %>%
    group_by(date) %>%
    summarise(steps = sum(steps))

activity.hourly <- activity %>%
    group_by(hour.posix) %>%
    summarise(steps = mean(steps, na.rm = TRUE))

activity %>% filter(!complete.cases(activity)) %>% tally
activity %>% complete.cases %>% table
summary(activity$steps)
summary(activity.daily)
table(is.na(activity$steps))
length(complete.cases(activity))

activity.daily %>%
    ggplot +
    geom_histogram(aes(steps), binwidth = 2000) +
    theme_pander() +
    scale_colour_pander()

activity.hourly %>%
    ggplot +
    geom_line(aes(hour.posix, steps)) +
    scale_x_datetime(labels = date_format("%H:%M")) +
    theme_pander() +
    scale_colour_pander()

activity.xts <- with(activity, xts(steps, time))
activity.xts.subset <- activity.xts["2012-10-01/2012-10-07"]
# activity.xts.subset <- activity.xts["T08:00/T16:00"]

activity.xts.daily <- apply.daily(
    activity.xts,
    sum
)
activity.xts.subset.daily <- apply.daily(
    activity.xts.subset,
    sum
)
activity.xts.subset.hourly <- period.apply(
    activity.xts.subset,
    endpoints(activity.xts.subset, "hours", 1),
    sum
)

plot(activity.xts.subset)
plot(activity.xts.subset.hourly)

activity.xts.subset.hourly %>%
#     na.approx %>%
    fortify %>%
    ggplot() +
    geom_line(aes(Index, activity.xts.subset.hourly)) +
    scale_x_datetime(breaks = date_breaks("1 day"))

activity.xts %>%
    na.approx %>%
    as.data.frame %>%
    tbl_df

autoplot(activity.xts.daily)
autoplot(activity.xts.daily %>% na.approx)
autoplot(activity.xts.daily %>% na.aggregate)
autoplot(activity.xts.daily %>% na.locf)

activity.daily.imp <- activity.xts %>%
    na.aggregate %>%
    apply.daily(sum) %>%
    fortify %>%
    select(
        date = as.Date(Index),
        steps = ...
    )
    
    fortify %>%
    ggplot +
    geom_histogram(aes(steps), binwidth = 2000) +
    theme_pander() +
    scale_colour_pander()

# --------------------------------------------------------------------------------------------------

activity.imputed <- activity %>%
    left_join(activity.hourly %>% select(interval, steps.imputed = steps)) %>%
    mutate(steps = ifelse(is.na(steps), round(steps.imputed), steps)) %>%
    select(-steps.imputed)