# we should endeavor to collect as much quantitative information about chris-chan as possible.
# e.g. when did she go to McDonald's, how much did she spend, etc.
# I think if we all worked together, we could build a machine learning simulator of chris chan
# which would allow us to predict what she is doing at any given moment. The model might even
# feel more authentic than the source of the ground truth.
library(lubridate)
cwc.file.data <- "
date, amount
2007-04-01, 25.0
2007-05-01, 35.0
2007-06-01, 10.0
2007-07-01, 25.0
2007-08-01, 65.0
2007-09-01, 35.0
2007-10-01, 55.0
2007-11-01, 30.0
2007-12-01, 25.0
2008-01-01, 75.0
2008-02-01, 20.0
2008-03-01, 20.0
2008-04-01, 16.02
2008-05-01, 90.0
2008-06-01, 30.0
2008-07-01, 120.0
2008-08-01, 85.0
2008-09-01, 90.0
2008-10-01, 90.07
2008-11-01, 110.0
2008-12-01, 150.0
2009-01-01, 90.0
2009-02-01, 125.0
2009-03-01, 210.0
2009-04-01, 225.0
2009-05-01, 135.0
2009-06-01, 170.0
2009-07-01, 80.0
2009-08-01, 50.0
2009-09-01, 132.36
2009-10-01, 268.41
2009-11-01, 215.0
2009-12-01, 234.29
2010-01-01, 90.0
2010-02-01, 130.0
"
cwc <- read.csv( text = cwc.file.data, sep=",", header = TRUE)
cwc$date <- as.Date(cwc$date)
cwc.ts <- as.numeric(as.POSIXct(cwc$date))
cwc.fit <- lm(cwc$amount ~ poly(cwc.ts,1))
cwc.pred <- predict(cwc.fit)
years1 <- year(cwc$date)
years2 <- c(0, head(years1, n = -1))
xlables <- Map(function(x, y) if(x == y) "" else x, years1, years2)
png("cwc.png", width=1024, height=768)
plot(cwc, type = "l", xlab="date", ylab="dollars spent per month", lwd = 4, xaxt = 'n')
lines(cwc$date, cwc.pred, col="red", lwd = 4)
axis(1, at = cwc$date, labels = month(cwc$date))
axis(1, at = cwc$date, labels = xlables, pos=-10, tick = FALSE)
dev.off()
