ABCDEFGHIJKLMNOPQRSTUVWXYZ
1
Guide for Interpreting Data
DiscordTwitchPatreonTwitterYoutube
2
JoINrbs, 3/13/2018
3
4
This is data collected from 95 Ironclad and 105 Silent Ascension 15 runs from Slay the Spire played by JoINrbs between February 23rd, 2018 and April 10th, 2018.
5
6
The data is extracted from the /steamapps/common/SlayTheSpire/runs folder.
7
The specific data analyzed is a record of card and relic choices. This record stores which card/relic was chosen and which it was chosen over.
8
Unfortunately it DOES NOT record choices where the player chose to "Skip" entirely, and it DOES NOT record whether cards were upgraded or not.
9
10
Times Seen records how many times a card/relic was seen at a choice where the player ended up picking one of the available cards/relics.
11
Pick% records what percent of the time it was chosen under those circumstances.
12
Win% records what percent of the time the run was won after it was picked.
13
14
Elo Rating is complex. The Elo rating system is used in games like Chess to track the strength of players. After each game against another player it adjusts
15
the ratings of the players based on how likely they were to win the game based on its own previous assessment of their strength.
16
17
For this Elo system the "game" is getting picked by me. A Card/Relic wins if it gets picked. This is reported as two or three wins (one against each of the
18
opposing Cards/Relics), and each of the opposing Cards/Relics receives a loss.
19
20
This Elo system begins Cards/Relics at 1600 rating. It rates the first few "games" they play higher than others (exponential decay on the increased weight).
21
It also rates the first 10 picks of each run more highly than other picks on the run.
22
23
The end result is that the Elo Rating of a card represents how much I personally value picking it over other cards, with a little increased importance being placed
24
on the beginning of runs. Also, Elo Rating is sensitive to recent occurrences, so it somewhat reflects my current opinions of cards over my previous opinions of cards.
25
26
Because this is exclusively an Ascension 15 sample it is a relatively small sample. It's large enough for Elo to capture the most important elements of each class fairly
27
well in my opinion, and for information about commons and uncommons to be fairly good, but winrates are fairly unreliable here, and some rares and relics
28
haven't been seen enough to normalize to their appropriate values yet.
29
30
Also keep in mind that this is describing what has happened in my runs, not trying to infer things about them. The data suggest that I value Backflip highly, but do not
31
necessarily suggest that Backflip has a particular abstract winrate when picked. Maybe I don't pick Backflip in decks which I think are poor, or maybe I load up on it
32
particularly often when the deck is already good. Things like this can change the win%'s of different cards such that the win% is less valuable as a predictor of a card's
33
actual strength.
34
35
Regression on the data suggests that only 5-15% of the variance in winning or losing runs can be explained by this level of analysis of the Cards/Relics picked in it.
36
Part of that is that the analysis is somewhat shallow, and part of it is that the game itself is huge with many other factors going on at once. A good lesson to take away
37
for reviewing card tier lists etc. in the future.
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100