Using data science to gain insight to jogging performance
Victor Clark
Introduction
This R script imports data generated from a runner wearing a GPS-enabled jogging watch with heartrate monitor. First non-standard packages are installed. Next, the data is manipulated to yield results more efficiently. The result is a presentation of charts communicating key metrics including analysis of speed, elevation climbed, heart rate, and GPS-mapped route.
This script includes sanity checkes throughout that have been commented out, including head(), summary(), and class()
Package management
Here we require packages the script calls.
# install.packages("XML")
# install.packages("ggmap")
# install.packages("reshape")
# install.packages("gpairs")
require(XML)## Loading required package: XMLrequire(ggmap)## Loading required package: ggmap
## Loading required package: ggplot2require(reshape)## Loading required package: reshaperequire(gpairs)## Loading required package: gpairsrequire(ggplot2)Data Extraction, Transform, Loading (ETL)
Import data for map generation and jog statistics, import raw TCX file as value:
mapData <- xmlParse("fog.tcx")
#Import raw TCX to dataframe. This will be used to generate map data
mapRawData <- xmlToDataFrame(nodes <- getNodeSet(mapData, "//ns:Trackpoint", "ns"))
head(mapRawData)## Time Position
## 1 2014-09-04T23:03:37.000Z 37.78069546446204-122.47062250971794
## 2 2014-09-04T23:03:40.000Z 37.78062807396054-122.47062703594565
## 3 2014-09-04T23:03:41.000Z 37.78058834373951-122.47063893824816
## 4 2014-09-04T23:03:42.000Z 37.78055163100362-122.47064262628555
## 5 2014-09-04T23:03:43.000Z 37.780513912439346-122.4706432968378
## 6 2014-09-04T23:03:44.000Z 37.78048072010279-122.47064765542746
## AltitudeMeters DistanceMeters HeartRateBpm Extensions
## 1 52.0 1.4900000095367432 87 1.315000057220459
## 2 52.0 8.979999542236328 92 1.5609999895095827
## 3 52.0 13.369999885559082 96 1.8930000066757202
## 4 52.20000076293945 17.459999084472656 99 2.1040000915527344
## 5 52.400001525878906 21.6299991607666 103 2.316999912261963
## 6 52.400001525878906 25.34000015258789 106 2.5250000953674316We see that it is a bunch of factors, which can be tricky to graph in a meaningful way, due to the complications from levels. Instead of using functions like as.numeric or type.convert storing as vectors, and then creating a another dataframe, Golden Cheetah is used to efficiently produce a CSV with managable classes from the TCX file
Import jog statistics from CSV
jogRawData <- read.csv("gold.csv")
# Uncomment sanity checks as needed
#class(jogRawData)
#head(jogRawData)
#str(jogRawData)
# Let's take a looks at the content of this dataframe
head(jogRawData)## Minutes Torq..N.m. Km.h Watts Km Cadence Hrate ID Altitude..m.
## 1 0.01667 0 5.029 0 0.003987 0 88.67 1 52.0
## 2 0.03333 0 5.324 0 0.006483 0 90.33 1 52.0
## 3 0.05000 0 5.620 0 0.008980 0 92.00 1 52.0
## 4 0.06667 0 6.815 0 0.013370 0 96.00 1 52.0
## 5 0.08333 0 7.574 0 0.017460 0 99.00 1 52.2
## 6 0.10000 0 8.341 0 0.021630 0 103.00 1 52.4gpairs(jogRawData)## Warning: 3 columns with less than two distinct values eliminated
We can remove columns for Toruqe since this was not a bike ride and ID since it appears to operate as a mile marker
So let’s make a three subsets
- Useful data to graph from jogRawData
- Metric pace, distance, and altitude data to be converted to standard units and added to 1
- Position data to later be divided into longitude and latitude from mapRawData
jogChartsTemp <- data.frame(jogRawData[c("Minutes",
"Hrate")])
jogMetricKmh <- jogRawData$Km.h
jogMetricKm <- jogRawData$Km
jogMetricAlt <- jogRawData$Altitude..m.
# Check jogChartsTemp
head(jogChartsTemp)## Minutes Hrate
## 1 0.01667 88.67
## 2 0.03333 90.33
## 3 0.05000 92.00
## 4 0.06667 96.00
## 5 0.08333 99.00
## 6 0.10000 103.00# Check jogMetricKmh & jogMetricKm & jogMetricAlt
head(jogMetricKmh)## [1] 5.029 5.324 5.620 6.815 7.574 8.341head(jogMetricKm)## [1] 0.003987 0.006483 0.008980 0.013370 0.017460 0.021630head(jogMetricAlt)## [1] 52.0 52.0 52.0 52.0 52.2 52.4I tend to think in terms of standard units, so let’s convert units from metric to standard
jogStdMph <- jogMetricKmh/1.609
jogStdMi <- jogMetricKm/1.609
jogStdAlt <- jogMetricAlt*3.28
# Check conversion
head(jogStdMph)## [1] 3.126 3.309 3.493 4.235 4.708 5.184head(jogStdMi)## [1] 0.002478 0.004029 0.005581 0.008310 0.010851 0.013443head(jogStdAlt)## [1] 170.6 170.6 170.6 170.6 171.2 171.9Finally, I prefer pace to speed, so we need to convert the data accordingly:
jogMpm <- jogStdMph/60
jogPace <- 1/jogMpm
head(jogPace)## [1] 19.20 18.13 17.18 14.17 12.75 11.57Now we can create a final dataframe for data manipulation:
jogCharts <- data.frame(jogChartsTemp, jogPace, jogStdMi, jogStdAlt)
names(jogCharts) <- c("Time", "Bpm", "Pace", "Dist", "Alt")
head(jogCharts)## Time Bpm Pace Dist Alt
## 1 0.01667 88.67 19.20 0.002478 170.6
## 2 0.03333 90.33 18.13 0.004029 170.6
## 3 0.05000 92.00 17.18 0.005581 170.6
## 4 0.06667 96.00 14.17 0.008310 170.6
## 5 0.08333 99.00 12.75 0.010851 171.2
## 6 0.10000 103.00 11.57 0.013443 171.9Now we want to create a subset of GPS data using reshape package
jogPos <- colsplit(mapRawData$Position, split = "-", names = c("lat",
"lon"))
head(jogPos)## lat lon
## 1 37.78 122.5
## 2 37.78 122.5
## 3 37.78 122.5
## 4 37.78 122.5
## 5 37.78 122.5
## 6 37.78 122.5Since the data capatured separates longitude and latitude by “-”, the column was split by the “-” character. We must return the negative value to the longitudinal coordinates to produce an accurate GPS route.
jogPos$lon <- as.numeric(lapply(jogPos$lon, function (x) 0-x))
head(jogPos)## lat lon
## 1 37.78 -122.5
## 2 37.78 -122.5
## 3 37.78 -122.5
## 4 37.78 -122.5
## 5 37.78 -122.5
## 6 37.78 -122.5Statistical Analysis
Let’s first get an idea of what the data looks like:
gpairs(jogCharts)
summary(jogCharts)## Time Bpm Pace Dist
## Min. : 0.02 Min. : 88.7 Min. : 6 Min. :0.002
## 1st Qu.: 8.91 1st Qu.:174.9 1st Qu.: 9 1st Qu.:0.997
## Median :17.80 Median :177.7 Median : 10 Median :1.901
## Mean :17.80 Mean :176.3 Mean :Inf Mean :1.871
## 3rd Qu.:26.69 3rd Qu.:180.0 3rd Qu.: 11 3rd Qu.:2.753
## Max. :35.58 Max. :195.0 Max. :Inf Max. :3.641
## Alt
## Min. :171
## 1st Qu.:195
## Median :207
## Mean :209
## 3rd Qu.:224
## Max. :236For a more detailed analysis, let’s find min, median, and max values for plotting later
Time analysis
Time.max.long <- max(jogCharts$Time)
Time.max <- round(Time.max.long, 2)Speed analysis
Pace.min <- min(jogCharts$Pace)
Pace.med.long <- median(jogCharts$Pace)
Pace.med <- round(Pace.med.long, 2)
Pace.max <- max(jogCharts$Pace)
Pace.plot <- qplot(Dist, Pace,
data = jogCharts,
geom = "line")
pace <- Pace.plot +
geom_hline(aes(yintercept=Pace.med),
color="darkgreen",
linetype="dashed") +
labs(title="Pace [min/mi] vs Dist [mi]") +
ylim(5,12) +
annotate("text",
x = 0.2,
y = Pace.med+0.25,
color = "darkgreen",
label = "Median pace") +
annotate("text",
x = 0.2,
y = Pace.med-0.25,
color = "darkgreen",
label = Pace.med)
print(pace)## Warning: Removed 5 rows containing missing values (geom_path).
Distance analysis
Dist.max1 <- max(jogCharts$Dist)
Dist.max <- round(Dist.max1, 2)Elevation analysis
Alt.min <- min(jogCharts$Alt)
Alt.med <- median(jogCharts$Alt)
Alt.max <- max(jogCharts$Alt)
Alt.range <- (Alt.max - Alt.min)
Alt.plot <- qplot(Dist,
Alt,
data = jogCharts,
geom = "point")
alt <- Alt.plot +
geom_hline(aes(yintercept=Alt.max),
color="darkblue",
linetype="dashed") +
labs(title="Alt [ft] vs Dist [mi]") +
geom_area() +
ylim (0, 300) +
xlim (0, Dist.max) +
annotate("text",
x = 0.5,
y = Alt.max+5,
color = "darkblue",
label = "Maximum Altitude") +
annotate("text",
x = 0.5,
y = Alt.max-5,
color = "darkblue",
label = Alt.max)
print(alt)## Warning: Removed 2 rows containing missing values (position_stack).
## Warning: Removed 2 rows containing missing values (geom_point).
Heartrate analysis
bpm.Min <- min(jogCharts$Bpm)
bpm.Med1 <- median(jogCharts$Bpm)
bpm.Med <- round(bpm.Med1,0)
bpm.Max <- max(jogCharts$Bpm)
bpm.plot <- qplot(Dist,
Bpm,
data = jogCharts,
geom = "point")
bpm <- bpm.plot +
geom_hline(aes(yintercept=bpm.Med),
color="red",
linetype="dashed") +
labs(title="BPM vs Dist [mi]") +
annotate("text",
x = 0.25,
y = bpm.Med+2,
label = "Median BPM",
color = "red") +
annotate("text",
x = 0.25,
y = bpm.Med-2,
label = bpm.Med,
color = "red")
print(bpm)
Performance anaylsis
This is not typically provided by most jogging software packages, but it is interesting to see how pace affects and heart rate, excluding other factors.
perf <- qplot(Pace,
Bpm,
data = jogCharts,
geom = "point") +
xlim(0,20) +
labs(title = "Heartrate [BPM] as function of Pace [min/mi]") +
geom_hline(aes(yintercept=bpm.Med),
color="red",
linetype="dashed") +
geom_vline(aes(xintercept=Pace.med),
color="blue",
linetype="dashed") +
geom_density2d(color = "white") +
annotate("text",
x = Pace.med+1,
y = 10,
label = "Median pace",
color = "blue") +
annotate("text",
x = Pace.med+1,
y = 1,
label = Pace.med,
color = "blue") +
annotate("text",
x = 1.5,
y = bpm.Med+5,
label = "Median BPM",
color = "red") +
annotate("text",
x = 1.5,
y = bpm.Med-5,
label = bpm.Med,
color = "red")
print(perf)## Warning: Removed 6 rows containing non-finite values (stat_density2d).
## Warning: Removed 5 rows containing missing values (geom_point).
The clustering indicates the body fatiguing during the run. This indicates that while the runner maintained some variablity in their pace, heart rate tended to stay within +/-25bpm. As you could expect, the clustering centers nicely on the median values for each variable.
GPS Map Generation
Now we check if the GPS data maps the way we expect
qplot(lon, lat, data = jogPos)
It does, so let’s overlay it on a googlemap using the ggmap package. First, we want to grab a map from google centered on median longitudinal and latitudinal values, set to an appropriate zoom scale:
mapImageData <- get_googlemap(center = c(lon = median(jogPos$lon), lat = median(jogPos$lat)),
zoom = 15, maptype = c("roadmap"))## Map from URL : http://maps.googleapis.com/maps/api/staticmap?center=37.777859,-122.465516&zoom=15&size=%20640x640&maptype=roadmap&sensor=false
## Google Maps API Terms of Service : http://developers.google.com/maps/termsNow we overlay the longitude and latitude coordinates from the jobPos dataframe
map <- ggmap(mapImageData,
extent = "device") + # takes out axes, etc.
geom_point(aes(x = lon, y = lat), data = jogPos, colour = "darkblue", size = 2, pch = 16)
print(map)
SUCCESS!!
Final stats
print ("Total time [min]")## [1] "Total time [min]"print (Time.max)## [1] 35.58print ("Total distance of run [mi]")## [1] "Total distance of run [mi]"print (Dist.max)## [1] 3.64print ("Median pace [mi/min]")## [1] "Median pace [mi/min]"print (Pace.med)## [1] 9.9print ("Max heart rate [BPM]")## [1] "Max heart rate [BPM]"print (bpm.Max)## [1] 195print ("Total climb [ft]")## [1] "Total climb [ft]"print (Alt.range)## [1] 65.6