STAT CHECK: ACS Groups

Earlier I predicted the ACS of every player at Champions. I made a post after the first round comparing how right I was as well. Now that the group stages are done, its time to analyze.

TLDR: The results have gotten closer to my predictions since the first round.

The ACS is what I predicted, the Groups is their ACS after groups, Diff is the difference between the two, Rank is their rank based on ACS, and R Diff is the difference in rank.
My predicted ACS has an average of 206.9 while the Champions ACS is 198 so there will be a lot of negative numbers as I predicted too much ACS in general. The bulk of players are more towards the middle than I had originally predicted. I had previously after the first round, been wrong by an average of 29.62 ACS per player, but now it is down to 23.25 ACS per player. The larger sample size took out a lot of the variance.

The Good: Derke, DubsteP, d3ffo, Mazino, Sushiboys, and dispenser all did better than I predicted. However, they all still have tougher matches ahead and some of them will drop.
The Bad: ScreaM, mwzera, PTC, frz, Rb, and ade all performed worse than I expected so far. ScreaM could still bring it back.
The Interesting: ade managed to be one of my biggest overpredictions despite me picking him to be 2nd to last. Yikes.

    Player  ACS Groups  Diff    Rank    R Diff
1   ScreaM  280.0   217.4   -62.6   22  -21
2   TenZ    275.0   225.2   -49.8   19  -17
3   mwzera  271.0   200.4   -70.6   41  -38
4   cNed    270.0   232.4   -37.6   12  -8
5   heat    259.0   264.8   5.8     2   3
6   Sacy    255.0   253.7   -1.3    3   3
7   PTC     246.0   168.8   -77.2   67  -60
8   Patiph  245.0   242.9   -2.1    7   1
9   yay     245.0   242.7   -2.3    8   1
10  zeek    245.0   242.4   -2.6    9   1
11  keznit  241.0   223.9   -17.1   20  -9
12  nAts    240.0   200.7   -39.3   40  -28
13  SicK    238.0   203.3   -34.7   34  -21
14  sheydos 235.0   226.5   -8.5    17  -3
15  JohnOls 234.0   192.0   -42.0   47  -32
16  leaf    233.0   250.0   17.0    4   12
17  saadhak 231.0   229.0   -2.0    16  1
18  BuZz    230.0   237.4   7.4     10  8
19  crashi  230.0   205.3   -24.7   30  -11
20  Jamppi  230.0   190.8   -39.2   50  -30
21  frz     229.0   169.3   -59.7   66  -45
22  xand    226.0   231.2   5.2     14  8
23  ShahZ   225.0   226.0   1.0     18  5
24  Rb      225.0   174.5   -50.5   63  -39
25  starxo  223.0   192.0   -31.0   48  -23
26  xeta    223.0   176.8   -46.2   62  -36
27  Munchk  221.0   183.3   -37.7   56  -29
28  gtnziN  220.0   190.9   -29.1   49  -21
29  MaKo    220.0   186.0   -34.0   55  -26
30  Nivera  219.0   205.4   -13.6   29  1
31  dapr    219.0   195.9   -23.1   43  -12
32  Witz    219.0   193.1   -25.9   45  -13
33  Medusa  218.0   193.5   -24.5   44  -11
34  foxz    215.0   205.0   -10.0   31  3
35  murizzz 215.0   173.8   -41.2   65  -30
36  Chronic 214.0   234.3   20.3    11  25
37  Xeppaa  214.0   213.6   -0.4    23  14
38  Bazzi   214.0   205.0   -9.0    32  6
39  Nozwerr 213.0   205.8   -7.2    28  11
40  DubsteP 211.0   246.4   35.4    6   34
41  Derke   210.0   283.0   73.0    1   40
42  Lakia   210.0   203.0   -7.0    36  6
43  d3ffo   205.0   249.5   44.5    5   38
44  Victor  205.0   162.3   -42.7   74  -30
45  Khalil  202.0   196.5   -5.5    42  3
46  LAMMYS  201.0   213.0   12.0    24  22
47  stax    196.0   192.6   -3.4    46  1
48  Fisker  195.0   206.5   11.5    26  22
49  Doma    195.0   204.3   9.3     33  16
50  sScary  194.0   174.3   -19.7   64  -14
51  Klaus   192.0   180.8   -11.2   58  -7
52  Marved  190.0   200.9   10.9    39  13
53  JessieV 188.0   203.3   15.3    35  18
54  soulcas 187.0   212.4   25.4    25  29
55  Quick   187.0   188.7   1.7     53  2
56  BORKUM  187.0   167.1   -19.9   69  -13
57  Mazino  186.0   231.6   45.6    13  44
58  BONECO  186.0   177.6   -8.4    60  -2
59  mitch   184.0   189.1   5.1     51  8
60  neth    184.0   164.3   -19.7   73  -13
61  NagZ    183.0   201.6   18.6    37  24
62  Sushibo 181.0   229.1   48.1    15  47
63  sutecas 180.0   182.9   2.9     57  6
64  ChAlala 179.0   139.5   -39.5   80  -16
65  k1Ng    178.0   177.3   -0.7    61  4
66  Redgar  178.0   161.5   -16.5   75  -9
67  dispens 177.0   222.3   45.3    21  46
68  SuperBu 175.0   179.3   4.3     59  9
69  Boaster 175.0   164.7   -10.3   71  -2
70  zombs   175.0   157.4   -17.6   78  -8
71  Crws    172.0   164.6   -7.4    72  -1
72  vanity  171.0   159.9   -11.1   77  -5
73  Magnum  170.0   201.3   31.3    38  35
74  Kiles   170.0   187.6   17.6    54  20
75  L1NK    169.0   206.2   37.2    27  48
76  FNS     165.0   188.9   23.9    52  24
77  delz1k  165.0   167.0   2.0     70  7
78  Mazin   165.0   161.0   -4.0    76  2
79  v1xen   165.0   149.2   -15.8   79  0
80  Mistic  165.0   135.0   -30.0   81  -1
81  ade     160.0   101.8   -58.2   82  -1
82  Jhow    145.0   168.2   23.2    68  14

ok

nice

Also after the first round I had the players rank wrong by an average of 18.7 per player and now it is 16.7.

That 0.4 on Xeppaa though.

Less go ScreaM on top #1taparmy

pretty amazing work but this forum is mainly children who like to make fun of regions. Trust me, more people will appreciate your post if you post it on the valorantcompetitive subreddit.

edit: double post bug

Getting more accurate on all fronts, probably because more players are getting a better true sampling of their usual ACS. The champions ACS distribution is getting closer in shape to the prior mean ACS and approaching its mean.

Round 1 plots:

• Kernel density estimates
• Mean error over time

Post-groups plots:

• Kernel density estimates
• Mean error over time

Round 1 scoring:

Total prior error = 2582 Total pred error = 2436 Total rand error = 4183.0
Average prior error = 32 Average pred error = 30 Average rand error = 52

Correct predictions(predicted acs): 30/81 = 37%
Correct predictions(prior acs): 24/81 = 29%
Correct predictions(random acs): 17/81 = 20%

Post-groups scoring:

Total prior error = 2107 Total pred error = 1919 Total rand error = 3465
Average prior error = 25 Average pred error = 23 Average rand error = 42

Correct predictions(predicted acs): 42/82 = 51%
Correct predictions(prior acs): 35/82 = 43%
Correct predictions(random acs): 26/82 = 32%

Total prior error is the absolute difference between the predictor and true ACS. The average error is just the mean of those differences. A "correct" prediction is categorized by whether it falls within +/- 20 ACS of the current champions ACS.

Note: The computed random predictions are different datasets(because I overwrote the other one by accident)