Wednesday, February 27, 2013

More on Iditarod analytics

So, the other day I wrote a post on IonEarth's new "analytics" that they'll be using in this Iditarod for the first time.  In rereading it I thought it was too negative and not very helpful, so although I left it up I did not publicize it.

Instead, I thought it might be worthwhile to spend a little time talking about tracker analytics and how they're useful.  The first thing that needs to be addressed is the question of why to have them in the first place.  Frankly, a lot of people are just not interested - they're happy with seeing the location of teams on a map and that tells them what they want to know.

However, sometimes people want to know more.  They want to get a handle on a given team's run/rest schedule, which is so critical to distance dog racing, or they want to know if the gap between a given pair of teams is opening up or closing.  Static maps, in which you can only see the current location of a team, won't help you much unless they also allow you to step backwards and forwards.  And even then you probably still won't get a very good feel for the overall trajectory, and it can be extremely tedious to step through an entire race just to find the couple of hours you're interested in.

There are some obvious tools that seem to be helpful at first glance, like average speed.  Unfortunately conventional averages treat all sample points identically and you can't tell if the speed has been constant, increasing, or decreasing.  Given two teams with two averages, all you can tell is which team has been faster so far, not whether or not a team is accelerating or slowing down.  Two teams can have the same average traveling speed when one goes crazy-fast but rests a lot and the other plods steadily with few breaks.

Analytics can help you get a sense of the dynamics of a race, how things are changing, where the averages come from, and whether or not there's something surprising in the underlying data.  That is, it can help you with those things if they're good analytics.  While I admire Trackleaders' clever use of low-cost, commodity hardware what I really, really love about them is that they understand that a race is dynamic and constantly changing, and they work pretty hard at figuring out ways to tell the story behind those points on a map.

Some of the tools Trackleaders provides are conceptually simple, but they're elegant and very expressive.  For example, their race flow chart just plots at what mile teams were at at a given time.  They put the top 10 teams on it (and they *really* need to allow us to choose which teams we'd like on the plot), but I digress).  It's a very simple idea but you can look at the plot and immediately see who's traveling faster, who's catching up with whom, etc.  You can see the relationships between the teams change over time and space and get a very good handle on what's been happening and what you can expect to happen in the near future.  It's a fantastic tool.  Here's an example, with the end of the 2013 Yukon Quest.  That race was a nailbiter almost right up until the end, but this plot shows Allen Moore opening a clear lead on Hugh Neff, as their lines get further apart over time (Allen's line is the yellow one on top, Hugh's is the light blue line very close to it):

Another of their tools that works well for distance dogsled racing is their plot of speed against time.  Again, you can take a look at one and get an immediate sense of run/rest schedules, whether a team is getting generally faster or generally slower, etc.  Here's a look at Brent Sass's speed/time plot from the 2013 Quest:

As you can see, he took very regular breaks on the trail (fantastic to see this kind of discipline from Brent; he could win this thing in the not-too-distant future and it's discipline like this that will help make it happen), along with his long, 40-hour layover in Dawson City.

Okay, so what about Iditarod and IonEarth?  Well, I'm trying to find something nice to say and it pretty much comes down to that their pictures are pretty.  I don't think they'll find themselves copying Trackleaders, unfortunately, but I think there are a couple of things they could do to improve their analytics on their own.

For one thing, if they can overlay temperature on top of the speed plot (seriously, were drugs involved with that decision?), they can certainly start thinking of ways to combine plots from different teams in a way that expresses their relationship (distance, speed) over time.  Another thing they could do that would help a lot would be to provide moving averages (REAL moving averages, not their ferkakte average speed while moving) over, say, two different sets of terms, one longer and one shorter, to get a sense of how speed is changing and how one team's speed is changing relative to another.

The main thing here is that when we're following a race, those of us who are interested in the dynamics of the race and how things are changing over time would find some analytical tools that can answer our questions or represent graphically what's happening on the trail really, really helpful.  IonEarth has incredible hardware in the units they mount on Iditarod sleds, and if they provided better analytical tools during races I think they'd be nearly unstoppable.

No comments:

Post a Comment