You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+15-12Lines changed: 15 additions & 12 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -5,7 +5,7 @@ Knowing how individual abilities change is essential in a wide range of activiti
5
5
Most estimators implemented for the video game industry cannot obtain reliable initial estimates nor guarantee comparability between distant estimates.
6
6
7
7
TrueSkill Through Time solves all these problems by modeling the entire history of activities using a single Bayesian network.
8
-
This algorithm requires only a few iterations to converge, allowing millions of observations to be analyzed using any low-end computer.
8
+
This algorithm requires a few iterations to converge, allowing millions of observations to be analyzed using any low-end computer.
9
9
10
10
To support the use of reliable learning estimators, we provide the first implementations of TrueSkill Through Time for `Julia`, `Python` and `R`.
11
11
A complete scientific documentation allows scientists to make sense of all epistemological and technical aspects of the estimation process.
@@ -14,35 +14,38 @@ A complete scientific documentation allows scientists to make sense of all epist
14
14
15
15
You can find the full scientific documentation of TrueSkill Through Time packages at:
16
16
17
-
0.[English version](https://github.com/glandfried/TrueSkillThroughTime/releases/download/doc.0.0.0/landfried2021.07.26-learning.pdf) (Last version 2021-07-26)
18
-
19
-
0.[Versión en español](https://github.com/glandfried/TrueSkillThroughTime/releases/download/doc.0.0.0/landfried2021.07.26-aprendizaje.pdf) (Última versión 2021-07-26)
0. R package: [https://github.com/glandfried/TrueSkillThroughTime.R](https://github.com/glandfried/TrueSkillThroughTime.R)
26
25
26
+
### Computational details
27
+
28
+
Our Python package solves individual events ten times faster than the original trueskill 0.4.5 (Lee 2012) package.
29
+
In turn, our Julia package converge a history of events ten times faster than our Python package.
30
+
In contrast, our R package is slower than the other packages, including the original trueskill 0.4.5
31
+
package.
32
+
27
33
### The History of the ATP
28
34
29
35
The following figure presents **the learning curves of some famous players in ATP history**.
30
36
A one-point difference between skills is equivalent to 76% probability of winning.
37
+
The top bar indicates which player was at the top of the ATP’s ranking.
31
38
32
39
33
40
34
-
Those who know the history of tennis will be able to recognize the periods of crisis, stability and success of the players.
35
-
There is a relative coincidence between skill and who is at any given moment at the top of the ATP rankings, it is possible to observe the effects of injuries, and even the effect of emotional slumps such as those that Aggasi and Djockovic had.
41
+
It is possible recognize the periods of crisis, stability and success of the players, even the effect of emotional slumps such as those that Aggasi and Djockovic had.
42
+
The skill of tennis players did not increase so much over the years: on the contrary the players of the 1980s were more skilled than those of the 1990s, and reached a skill similar to what Federer, Nadal and Djokovic had in 2020.
36
43
37
-
It is interesting to see that the skill of tennis players did not increase so much over the years: on the contrary the players of the 1980s were more skilled than those of the 1990s, and reached a skill similar to what Federer, Nadal and Djokovic had in 2020.
38
-
39
-
The 10th player in the historical ATP rankings, Hewitt, is only a product of the window of opportunity that was opened in the year 2000; and the 4th most skilled player, Murray, is ranked 14th just above Nastase.
44
+
There is a relative coincidence between skill and who is at any given moment at the top of the ATP rankings.
45
+
However, TrueSkill Through Time allows comparing the relative ability of players over time: the 10th player in the historical ATP's ranking, Hewitt, is a product of the window of opportunity that was opened in the year 2000; and the 4th most skilled player, Murray, is ranked 14th just above Nastase.
40
46
41
47
The **models commonly used in industry and academia** (TrueSkill, Glicko, Item-Response Theory) are based on an ad-hoc solution that prevents them both from having good initial estimates and from guaranteeing comparability between estimates separated in time.
42
48
43
49
44
50
45
51
The advantage of TrueSkill Through Time lies in its temporal causal model, that links all historical activities in the same Bayesian network, which guarantees both good initial estimates and the temporal and spatial comparability of the estimates.
46
-
We have shown that this procedure converges quickly on large databases.
47
-
48
-
With this article we promote their adoption by providing the first packages for Julia, Python and R, together with the complete scientific documentation.
0 commit comments