Chart user defined moving average of a metric

[wooliethai]

I want to plot trailing 30 day, 90 day, and 365 day average of a number of metrics. What's the easiest way to do this?

[Ale Martinez]

El viernes, 21 de abril de 2017, 7:40:26 (UTC-3), wooliethai escribió:

I want to plot trailing 30 day, 90 day, and 365 day average of a number of metrics. What's the easiest way to do this?  

On v3.3 you can add simple formula s.t. mean(Metric[Date-30:Date]) to a Metrics Trends chart, but this is broken in v3.4 so the alternative would be an R chart which is much more flexible but not that easy, something in the lines of http://www.cookbook-r.com/Manipulating_data/Calculating_a_moving_average/ with y <- GC.season.metrics()$Metric should do the trick on v3.4

[Mark Liversedge]

Its only broken because we wanted to add array subscripts, we could re-introduce with another operator, e.g: '[[' and ']]' ?

[Ale Martinez]

The change proposed by Mark will be on v3.5, starting from the next development build, the new formula to average a metric for the last 30 days will be: mean(Metric_Name[[Date-30:Date]])

[wooliethai]

Would this be the correct formula to chart 30-day trailing weekly mileage: mean(Distance[[Date-30:Date]])*0.621371*7?

The general trend appears to be correct but the actual mileage being graphed is overstated significantly (Strava says my average weekly mileage over the last 4 weeks was 29 miles, the formula says its 45.70 miles)

[Robert Chung]

On Tuesday, April 25, 2017 at 2:12:56 PM UTC-7, Ale Martinez wrote:

The change proposed by Mark will be on v3.5, starting from the next development build, the new formula to average a metric for the last 30 days will be: mean(Metric_Name[[Date-30:Date]])

Just a comment: in R, operator precedence means the ":" gets evaluated first. If you wanted something like a 30-day running mean, you'd have to add parentheses like (Date-30):Date

[Mark Liversedge]

Thanks, this is a GC "formula" rather than R, but your point is still noted. 

[wooliethai]

On further thought, my suspicion is that days with no activity are being excluded from the average rather than being counted a 0 miles. How can I include days without activity as being 0?

[Ale Martinez]

El domingo, 7 de mayo de 2017, 14:16:16 (UTC-3), wooliethai escribió:

On further thought, my suspicion is that days with no activity are being excluded from the average rather than being counted a 0 miles. How can I include days without activity as being 0?

You can try sum(Distance[[Date-30:Date]])*0.621371*7/30

[Ale Martinez]

Just for clarity sake: mean on date range is an average of the activities in that range not a daily mean. 

[nev...@gmail.com]

I also hit this problem. The R tools all assume a contigious timeseries and don't give a correct result. I use this script to graph rolling cumulative and moving average distances using the zoo package to create a contigious series:zm. Should be pretty easy to adapt to the metrics you want.

require(zoo)

win <- 7  # no. of days in window

z <- zoo(GC.season.metrics()$Distance,GC.season.metrics()$time)

z0 <- zoo(, seq(start(z), end(z), "day")) # empty vector of all the days

zm <- merge(z,z0, fill=0) 

index(zm) <- as.Date(index(zm)) # round to whole days

zm <- aggregate(zm,time(zm), sum) # no dup indexes allowed in zoo

ms7 <- rollapply(zm,win, sum, na.rm=TRUE,fill=NA,align = "right",partial=TRUE)

ma7 <- rollapply(zm,win, mean, na.rm=TRUE,fill=NA,align = "right",partial=TRUE)

zresult <- merge(zm,ma7,ms7)

plot(zresult,plot.type = "single", col=c("yellow","blue","red"),xlab="Date",ylab="Dist/km",main="Sliding Window Distances")

I havent worked out if it is possible to feed these vectors back into the PMC or Diary views yet.