Here’s the Beatpaths graph for Week 4. Horrible record this week of 6-7, lots of upsets – but quite an interesting graph.
One new beatloop this week: KC->DEN->OAK->KC.
-
Beatpaths is an automatic graphical power-ranking system based off of only wins, losses, and who beat who.
- How does it work?
- How are beatloops resolved?
- How do the power rankings work?
- What's a beatfluke?
- Reviews and Quotes
- Contact me (through my music site) -
Pages
Categories
- 2008 NFL Picks (40)
- Beatpath Tech (28)
- MLB (7)
- NBA (5)
- NCAA Football (7)
- News and Notices (18)
- NFL (384)
- Uncategorized (18)
Calender
Recent Comments
Question:
How did you decide which beatloop to remove when KC beat Denver? You removed the KC/DEN/OAK/KC loop, but the KC/DEN/NO/TB/ATL/KC loop is still in place.
Funny how NFC East, ACF North, AFC South, and NFC West all have vertical pecking orders this week.
Eagles loss was a fluke due to Westbrook being out (Andy Reid’s playcalling was just its usual criminal self).
Eagles loss was just that–a loss. Bears were better.
Besides, if you want to play the injury card, Bears were without Pro Bowl DT Tommie Harris.
Hrm…there was supposed to be a joke-end-homerism HTML tag at the end of that comment…I guess the reply software filtered it out.
My graphs are up: http://www.twomuffin.com/BeatPaths.htm
Note that it’s a different link than last week. Also, I’ve built a site that houses explanations of the different methods as well as historical records that are easy to find. If you have any ideas to make the search back better, please let me know now before I start loading the 1800+ weeks of history. There is a link for emailing me there as well.
The AFC South does not have a vertical pecking order; the Texans are not actually below the Colts.
Not trying to be a complete noob, but how is it that the Bills and Titans are in the middle of the graph at 4-0, and there’s a pile of teams above them that have lost one or two games? Is this something that just gets sorted as more data is available? Or, is the algorithm smart enough to register that neither team has beat anyone ‘good’ yet?
Smallest beatloops always get removed first, and then the graph is recalculated. For instance, if two teams split victories, then both victories are removed. Since these are often part of larger beatloops, then it means we get to retain the remaining information.
And Mattpie, yes, the vertical placement is all controlled by the graphing algorithm written by AT&T’s graphviz package.
To help clarify things for Mattpie, the graphing package tries to make the shortest lines, so that seems to determine where teams fall on the graph. Keeping that in mind, teams that haven’t developed beatwins against higher quality opponents tend to show up lower in the graph. It helps to remember that any team with a beatloss (no arrows coming in) could be as high as the top of the graph, but no lower than where the are shown. On the other end, teams with no beatlosses (no arrows going out) could be as low as the bottom of the graph. As more games are played the graph will get better, though it still will try to keep the arrows as short as possible.
I remember talk last season (?) about a suggestion to let people vote or rank teams within the confines of the beatpaths graph. Thus, any team with no beatlosses could be ranked as high as first, and no lower than one above any team they have a beatwin over. It was an interesting idea, but kinda gets away from the theme of just ranking based on wins/losses.
Just a question, what would happen if you removed the longest betloops first? I think I understand the reasoning behind removing the shortest ones, you want to remove minimal information to make the graph work out, and removing the longest loops first would likely end up with more data overall removed from the set. But would it change things much?
This week removing the longest paths wouldn’t change anything since all 3 BeatLoops are independent of each other. However, towards the end of the season when there are tons of loops, removing the longest ones first can have a large effect as every team is involved in loops unless they’re winless or undefeated. Large groups of games would be removed. Additionally, it may be possible to not only make a beatloop containing all 32 teams, but there may be multiple ways to do it. In the end we’d have a lot of disjointed data.
Still no differences between the iterative and the standard method.
It’s perhaps worth noting that the KC->Denver game is a classic “beatfluke”, since Denver retains a beatpath to KC despite the loss.
Good point, doktarr. That’s true. In that case, DEN->OAK->KC would still be part of the graph.