IRI Statistics/Predictions Live Blog

Caleb_Sykes · July 12, 2019, 10:19pm

Here are the Elo changes for every team at the competition from the start of the event up until the present:

Elo changes

team Number	nickname	elo change
111	WildStang	63
6443	AEMBOT	59
2481	Roboteers	56
48	Team E.L.I.T.E.	53
930	Mukwonago BEARs	52
5406	Celt-X	50
1684	The Chimeras	49
1023	Bedford Express	38
3847	Spectrum -△◅	37
2614	Mountaineer Area RoboticS (MARS)	35
3478	PrepaTec - LamBot	35
2767	Stryke Force	33
319	Big Bad Bob	32
1114	Simbotics	32
1807	Redbird Robotics	30
4607	C.I.S.	29
3604	Goon Squad	25
548	Robostangs	25
2403	Plasma Robotics	24
225	TechFire	24
2056	OP Robotics	23
4776	S.C.O.T.S. Bots	22
4028	The Beak Squad	18
1073	The Force Team	15
4265	Secret City Wildbots	15
195	CyberKnights	12
3357	COMETS	10
910	The Foley Freeze	9
5511	Cortechs Robotics	8
118	Robonauts	8
1720	PhyXTGears	4
1923	The MidKnight Inventors	3
7498	Wingus & Dingus	3
1747	Harrison Boiler Robotics	2
5205	FullMetal-Jackets	2
2910	Jack in the Bot	0
1241	THEORY6	-2
2075	Enigma Robotics	-2
1690	Orbit	-6
2168	Aluminum Falcons	-8
330	The Beach Bots	-8
5190	Green Hope Falcons	-9
107	Team R.O.B.O.T.I.C.S.	-10
1024	Kil-A-Bytes	-12
3538	RoboJackets	-15
1718	The Fighting Pi	-16
364	Team Fusion	-17
1730	Team Driven	-20
340	G.R.R. (Greater Rochester Robotics)	-23
2337	EngiNERDs	-25
5010	Tiger Dynasty	-26
3940	CyberTooth	-26
1676	The Pascack PI-oneers	-28
461	Westside Boiler Invasion	-32
4362	Gems	-34
51	The Wings of Fire	-35
2468	Team Appreciate	-36
7457	suPURDUEper Robotics	-36
868	TechHOUNDS	-36
3707	Brighton TechnoDogs	-36
1410	The Kraken	-46
217	ThunderChickens	-46
5801	CTC Inspire	-47
88	TJ(Squared)	-49
33	Killer Bees	-51
5460	Strike Zone	-55
234	Cyber Blue	-66
3641	The Flying Toasters	-79

111, 6443, 2481, 48, and 930 are here to play

Caleb_Sykes · July 12, 2019, 10:44pm

Q61 had some big implications for the top seed race:

Nominally 930 was the winner, but missing an RP means their overall chances didn’t change much. The real winner was 2056 who saw their chances go up 8% because no one got 4RPs. The other teams down the list saw their chances improve as well at 1114’s expense.

Caleb_Sykes · July 12, 2019, 11:24pm

From Q66, 319 goes from 16% to <1% first seed. 1023 goes from 7% to 22%. My model actually believes in 1023. They still have 2 very rough matches ahead of them in q101 and especially q91, but they might not even need to win both to seed first. Losing 1 would put them at 33RPs if they get all the rest, which would be tied with 2056 if 2056 gets all remaining RPs.

Evan_S · July 12, 2019, 11:52pm

Caleb_Sykes · July 12, 2019, 11:55pm

Alright, here are rankings at the end of Friday:

Projections after q70

Team	5th Percentile	Average Rank	95th Percentile	1	Top 4	Top 8	Top 12	Top 15
2056	1	2.2	5	44%	91%	>99%	>99%	>99%
1023	1	3.0	6	22%	84%	>99%	>99%	>99%
2481	1	4.1	9	14%	65%	92%	99%	>99%
930	1	4.2	10	16%	65%	91%	98%	>99%
225	2	5.3	10	3%	38%	90%	98%	>99%
1114	2	6.9	14	1%	30%	73%	91%	97%
319	4	10.5	21	<1%	8%	42%	70%	84%
2910	5	10.8	20	<1%	4%	38%	70%	83%
3538	4	11.3	21	<1%	7%	32%	64%	80%
2767	6	12.6	22	<1%	2%	25%	54%	75%
1684	6	13.4	24	<1%	<1%	18%	51%	67%
2403	6	13.6	24	<1%	2%	19%	47%	67%
330	7	14.1	22	<1%	<1%	14%	37%	63%
4607	6	14.5	26	<1%	1%	15%	44%	61%
3604	6	14.6	26	<1%	<1%	18%	42%	61%
195	7	14.8	24	<1%	1%	14%	39%	55%
48	7	15.3	25	<1%	<1%	11%	33%	56%
111	9	17.7	27	<1%	<1%	4%	17%	35%
4265	12	19.4	27	<1%	<1%	1%	7%	21%
868	10	21.0	32	<1%	<1%	<1%	12%	22%
2168	10	21.7	34	<1%	<1%	2%	10%	20%
910	11	22.1	34	<1%	<1%	<1%	8%	14%
1807	13	23.2	35	<1%	<1%	1%	5%	12%
340	15	24.7	35	<1%	<1%	<1%	<1%	6%
3357	16	26.9	40	<1%	<1%	<1%	1%	5%
6443	14	27.4	41	<1%	<1%	<1%	4%	7%
3847	18	28.2	40	<1%	<1%	<1%	<1%	2%
118	18	29.7	43	<1%	<1%	<1%	<1%	1%
5460	22	31.5	44	<1%	<1%	<1%	<1%	<1%
2075	19	31.6	45	<1%	<1%	<1%	<1%	2%
5406	20	32.3	45	<1%	<1%	<1%	<1%	2%
1747	21	32.9	47	<1%	<1%	<1%	<1%	<1%
364	21	33.2	46	<1%	<1%	<1%	<1%	<1%
1720	22	33.2	45	<1%	<1%	<1%	<1%	<1%
4028	23	34.3	48	<1%	<1%	<1%	<1%	<1%
5511	23	36.5	49	<1%	<1%	<1%	<1%	<1%
1690	25	36.8	49	<1%	<1%	<1%	<1%	<1%
7498	26	38.6	51	<1%	<1%	<1%	<1%	<1%
2468	28	39.0	50	-	<1%	<1%	<1%	<1%
107	27	39.3	50	<1%	<1%	<1%	<1%	<1%
1241	30	41.5	52	-	<1%	<1%	<1%	<1%
33	28	42.0	55	<1%	<1%	<1%	<1%	<1%
3641	30	42.1	52	<1%	<1%	<1%	<1%	<1%
7457	29	42.3	54	<1%	<1%	<1%	<1%	<1%
3478	30	43.6	56	-	<1%	<1%	<1%	<1%
4362	33	44.8	55	-	<1%	<1%	<1%	<1%
5205	33	45.2	56	-	<1%	<1%	<1%	<1%
548	33	45.2	55	-	<1%	<1%	<1%	<1%
3940	35	46.6	56	<1%	<1%	<1%	<1%	<1%
88	36	47.7	56	-	<1%	<1%	<1%	<1%
1923	36	47.7	58	-	<1%	<1%	<1%	<1%
4776	37	49.7	59	-	<1%	<1%	<1%	<1%
5010	42	51.3	58	-	<1%	<1%	<1%	<1%
1676	44	53.7	61	-	<1%	<1%	<1%	<1%
51	45	55.2	63	-	<1%	<1%	<1%	<1%
5190	48	56.4	64	-	<1%	<1%	<1%	<1%
1718	49	58.1	65	-	-	-	<1%	<1%
2614	52	58.9	65	-	-	-	<1%	<1%
1410	51	58.9	65	-	<1%	<1%	<1%	<1%
461	51	59.3	66	-	-	-	<1%	<1%
1073	51	59.6	66	-	-	-	<1%	<1%
5801	53	60.1	66	-	-	-	<1%	<1%
2337	55	61.9	67	-	-	-	<1%	<1%
1730	56	62.7	68	-	-	-	<1%	<1%
1024	56	62.7	68	-	-	-	-	<1%
234	59	64.6	68	-	-	-	-	<1%
3707	60	65.5	68	-	-	-	-	-
217	62	66.4	68	-	-	-	-	-

2056 and 1023 will probably be in the top 4, which will likely be rounded out with 2481 and/or 930. Also reasonable chance one or both of 225 or 1114 gets in the top 4 as well. For the one seed we have 2056 as a comfortable favorite (not yet >50% though) with 1023, 2481, and 930 all in solid contention.

Lidor51 · July 13, 2019, 12:31am

Seems it’s going to be something like 31-32 RP.

Ryan_Swanson · July 13, 2019, 12:35am

@Caleb_Sykes this live blog is really awesome by the way. Thanks for doing it! It’s made it really easy to track how we’re doing (and so much more painful when we miss RP’s for stupid reasons).

Strategos · July 13, 2019, 12:51am

Yes this has been very informative. @Caleb_Sykes thanks for doing this!

Richard_Wallace · July 13, 2019, 1:10am

Yeah, so this is IRI. Two current World Champions, dead last.

Caleb_Sykes · July 13, 2019, 5:22am

Here’s something that maybe is interesting, I’ve taken every team and some of their metrics and compared their pre-IRI seed values against their values at IRI so far.
IRI metrics comparison.xlsx (154.9 KB)

Things I found interesting:

The average change across all metrics is negative. Not particularly surprising as teams are performing at below peak in the offseason, but more importantly, all metrics use the team’s peak in-season rating, not most recent. So of course the average will be lower than everyone’s max.
For the OPR match predictions this is fine, as everyone being overrated by some amount shouldn’t really change things much, but for the RP predictions this can cause some trouble. Almost every match I saw today in my simulator had the HAB RP success likelihood at 100%, which is partially due to this effect. Working on an alternative for that.
2910’s Complete Rocket RP rate is .28 higher than expected. I think that this is largely due to the prevalence of teams that can complete the rocket. Other teams would also benefit from this change though, what I bet makes 2910 special is that they are so good at filling other roles in the match that their rocket-capable teammates don’t have to worry about those things and can better focus on completing the rocket.
4028 has the highest change in endgame points. This is probably due to their unique 3rd level climb that often allows another team to climb along with them.
4028 also has the highest change in auto points. I don’t know exactly the cause for this, but it is strangely coincidental along with the climbing. My suspicion is that they have added some auto modes for IRI so that they can always complement their partners’ routines, but that’s just speculation. Maybe someone from the team can chime in.

Nick_Donahue_40 · July 13, 2019, 11:50am

Hey this is Nick from team 4028. Our auto modes have significantly changed since worlds as we have removed them. We added a 6 NEO drive train in order to more effectively play defense. We switched to teleop for sandstorm, and we have become very consistent due to a well practiced and experienced drive team. In regards to the climb, our neos help to give our climb situation by giving us more pushing power to make the extra room on level 3 if all we need is an inch or two more. In all, the climb only takes up 17-18 inches of the platform.

Caleb_Sykes · July 13, 2019, 3:06pm

Almost everyone has just one match left.

Rankings as of q90

Team	5th Percentile	Average Rank	95th Percentile	1	Top 4	Top 8	Top 12	Top 15
930	1	1.2	2	84%	>99%	>99%	LOCK	LOCK
2056	1	2.0	3	12%	>99%	>99%	LOCK	LOCK
1023	2	3.6	6	4%	83%	>99%	>99%	>99%
2481	3	4.3	7	<1%	62%	>99%	>99%	>99%
1114	3	5.4	9	<1%	31%	94%	>99%	LOCK
225	4	5.6	9	<1%	21%	91%	>99%	>99%
3538	5	9.0	12	<1%	2%	31%	99%	>99%
330	7	9.1	13	-	<1%	53%	93%	>99%
2403	6	9.1	13	<1%	<1%	41%	94%	>99%
2168	6	9.2	14	<1%	<1%	48%	90%	97%
2910	7	11.0	16	-	<1%	22%	68%	91%
319	7	11.2	16	-	<1%	21%	67%	94%
111	11	16.0	23	-	<1%	<1%	18%	53%
3604	10	16.7	24	-	<1%	<1%	27%	40%
364	13	17.3	23	-	-	<1%	<1%	42%
4265	11	17.9	26	-	<1%	<1%	13%	39%
5406	13	18.4	27	-	-	<1%	1%	39%
868	12	18.5	28	-	-	<1%	14%	40%
4607	12	19.1	25	-	<1%	<1%	7%	26%
3357	12	20.0	27	-	<1%	<1%	7%	21%
2767	15	20.0	27	-	<1%	<1%	1%	11%
1747	15	21.2	29	-	-	<1%	<1%	5%
1684	17	21.3	28	-	-	-	<1%	<1%
1807	18	23.5	31	-	-	<1%	<1%	<1%
5460	21	25.5	33	-	-	-	-	<1%
340	18	25.9	32	-	-	-	<1%	<1%
48	19	26.1	30	-	-	<1%	<1%	<1%
195	18	26.7	33	-	-	-	<1%	<1%
1720	23	28.7	36	-	-	-	<1%	<1%
2075	21	29.0	35	-	-	-	<1%	<1%
1241	25	30.6	39	-	-	-	-	<1%
6443	21	31.2	40	-	-	-	<1%	<1%
1690	28	35.3	42	-	-	-	-	<1%
5511	28	35.6	41	-	-	-	-	<1%
3847	31	36.3	45	-	-	-	-	<1%
118	32	36.9	43	-	-	-	-	<1%
4028	33	38.0	45	-	-	-	-	-
4362	30	38.8	46	-	-	-	-	-
3940	31	39.1	46	-	-	-	-	<1%
910	35	39.8	46	-	-	-	-	-
1923	32	40.1	46	-	-	-	-	<1%
548	34	41.6	51	-	-	-	-	-
4776	35	41.8	49	-	-	-	-	-
2468	32	41.9	49	-	-	-	-	-
33	37	43.4	52	-	-	-	-	-
7498	38	44.3	51	-	-	-	-	-
107	38	45.2	48	-	-	-	-	-
7457	42	48.8	52	-	-	-	-	-
5205	46	50.3	56	-	-	-	-	-
1676	45	50.8	56	-	-	-	-	-
5801	46	50.8	55	-	-	-	-	-
88	47	51.7	58	-	-	-	-	-
3641	47	51.9	55	-	-	-	-	-
2337	47	52.9	58	-	-	-	-	-
5010	48	54.0	58	-	-	-	-	-
5190	51	56.8	61	-	-	-	-	-
51	51	56.8	62	-	-	-	-	-
3478	52	57.2	62	-	-	-	-	-
2614	55	60.0	64	-	-	-	-	-
461	57	61.4	65	-	-	-	-	-
234	57	61.5	67	-	-	-	-	-
1073	57	61.5	66	-	-	-	-	-
1024	57	61.6	66	-	-	-	-	-
1730	59	64.0	68	-	-	-	-	-
1410	59	64.1	67	-	-	-	-	-
1718	60	64.7	68	-	-	-	-	-
217	62	66.0	68	-	-	-	-	-
3707	63	66.8	68	-	-	-	-	-

Caleb_Sykes · July 13, 2019, 4:57pm

These were surprisingly accurate

Caleb_Sykes · July 13, 2019, 4:59pm

Here’s a quick reference of team Elos so you can get a feel for alliance selection:

end of quals Elo ratings

team Number	nickname	winning Margin Elo
2056	OP Robotics	1974
2910	Jack in the Bot	1946
3538	RoboJackets	1911
2481	Roboteers	1886
1114	Simbotics	1869
5406	Celt-X	1858
364	Team Fusion	1852
225	TechFire	1847
1690	Orbit	1844
330	The Beach Bots	1837
1023	Bedford Express	1826
930	Mukwonago BEARs	1822
195	CyberKnights	1820
2767	Stryke Force	1814
2168	Aluminum Falcons	1813
6443	AEMBOT	1805
3707	Brighton TechnoDogs	1793
1807	Redbird Robotics	1778
5460	Strike Zone	1777
111	WildStang	1767
340	G.R.R. (Greater Rochester Robotics)	1766
319	Big Bad Bob	1765
5190	Green Hope Falcons	1758
910	The Foley Freeze	1757
33	Killer Bees	1757
2075	Enigma Robotics	1749
3604	Goon Squad	1749
2403	Plasma Robotics	1747
1241	THEORY6	1742
2468	Team Appreciate	1738
4776	S.C.O.T.S. Bots	1734
1720	PhyXTGears	1730
3847	Spectrum -△◅	1730
2337	EngiNERDs	1730
1676	The Pascack PI-oneers	1720
1923	The MidKnight Inventors	1718
118	Robonauts	1717
4028	The Beak Squad	1714
4362	Gems	1714
3641	The Flying Toasters	1714
3357	COMETS	1713
107	Team R.O.B.O.T.I.C.S.	1704
1747	Harrison Boiler Robotics	1697
1684	The Chimeras	1695
4607	C.I.S.	1695
1730	Team Driven	1688
1718	The Fighting Pi	1685
5801	CTC Inspire	1682
48	Team E.L.I.T.E.	1675
88	TJ(Squared)	1674
5511	Cortechs Robotics	1673
51	The Wings of Fire	1670
234	Cyber Blue	1670
3478	PrepaTec - LamBot	1665
7498	Wingus & Dingus	1659
217	ThunderChickens	1657
3940	CyberTooth	1654
7457	suPURDUEper Robotics	1652
548	Robostangs	1651
5205	FullMetal-Jackets	1638
5010	Tiger Dynasty	1633
868	TechHOUNDS	1633
1024	Kil-A-Bytes	1630
1073	The Force Team	1628
2614	Mountaineer Area RoboticS (MARS)	1617
461	Westside Boiler Invasion	1608
4265	Secret City Wildbots	1607
1410	The Kraken	1596

BordomBeThyName · July 13, 2019, 5:22pm

How on earth did 364 get passed over?

Caleb_Sykes · July 13, 2019, 5:35pm

Here’s 2056’s path to the 1 seed

Here are the top 4 teams and their probabilities of being in the top 4:

XaulZan11 · July 13, 2019, 6:23pm

I think IRI is the toughest event to scout. Roughly the same amount of teams in a division, fewer qual matches and just about every team can make an argument to be picked – unlike the championship where you can probably ignore a good percentage of teams before the event begins. On top of all that, teams make robot changes (or bring brand new robots) so the value in watching previous matches is reduced. Finally, not every team is there to ‘win at all cost’ like the championship as some teams use IRI to train new drivers. So, even if every team brought the same resources to scout IRI as they do the championship, you’d probably see significantly less consensus among picks.

Csherm · July 13, 2019, 7:30pm

319’s scouting data has them 37th in game pieces scored per match. So they weren’t an exceptional offensive threat, plus they come from a region where very few people have heard of or worked with them before. My conclusion: IRI is brutal, if you want to get picked you need to stand out from the crowd.

samthesnake · July 14, 2019, 1:31am

I disagree. It is not the hardest because so many teams have played so many matches already. There is a pit scouting element of finding the new drivers, but for the most part in the stands you are verifying what you already know from pre scouting. And noting what is different throughout. That said, I agree, you won’t have much consensus. A lot of that is teams not all looking for the same thing, though.

Big_frc_fan · July 14, 2019, 5:21am

5460 / Strike Zone seemed to have some problem as an alliance partner with 930 in the very last qual match. Bot wouldn’t turn properly and sounded like there was a real drivetrain issue. The vagaries of competition and the robotics deities decided to mix things up.

Mind you they threw the same vagaries at 2056 shortly before by denying them 4 RP in quals 91 so it seems they are ambivalent with their human playthings