Download presentation
Presentation is loading. Please wait.
Published byTabitha Hines Modified over 6 years ago
1
Effect of the location on the quality of the curve
ABHINAV KUMAR VIGNESHWARAN VASANTHAKUMAR
2
HYPOTHESIS The goal is to check whether there is any observable difference in the basic s,c fit when data used is from only the games played in a specific country/continent. Based on the fit, we want to make inferences about the quality of players from different countries.
3
METHODOLOGY We collected the data of the games played in 15 different countries spanning across 4 continents. We used the EventCountryFilter available in the IR program to filter out games played in a specific country. Then the obtained data was plotted to check the variation in the curve. We made two sets of plots. One set is for each country and the other set is for a continent in whole. For each country we obtained the data starting from the rating level 1950 to The ELO was increased in steps of We used SF7 engine. We plotted the following graphs: ELO, S ELO, C S, IPR ASD, IPR
4
METHODOLOGY(cont’d) We see the effect on the average error for different IPR values by plotting a bar plot between IPR and the corresponding ASD values.
5
Possible biases in the data
In the tournaments played in a specific country there could be certain foreign players which might affect the outcome. But on closer examination this doesn’t seem to make much of an impact on the curve.
6
RESULTS From the plots obtained on the s and IPR values, we see more linearity in European countries as compared to the rest . The linearity metric for North American countries is pretty low as compared to others. It is hard to come to a conclusion as the data points obtained were only from USA and Canada. The projected IPR values for USA is less than the corresponding ELO ratings for the ELO From the final bar plot ,we see that as the IPR level increases ,the ASD value decreases.
7
RESUlts (cont’d) Elo vs c for Azerbaijan
Example s vs ELO and C vs ELO plots for different countries and variation in data : Elo vs c for Azerbaijan
8
RESUlts (cont’d) Elo vs s for China
9
RESUlts (cont’d) ASD vs IPR for Russia
Example ASD vs IPR plots for different countries and variation in data : ASD vs IPR for Russia
10
ASD vs IPR for Argentina
RESUlts (cont’d) ASD vs IPR for Argentina
11
ASD vs IPR for South America
RESULTS(CONT’D) ASD vs IPR for South America
12
RESUlts (cont’d) ASD Vs IPR plot for Germany IPR ASD
13
RESUlts (cont’d) ASD Vs IPR plot for North America IPR ASD
14
RESUlts (cont’d) ELO Vs C plot for CANADA C ELO
15
RESUlts (computed Average squared error)
Europe North America South America Asia
16
CONCLUSIONS AND FURTHER HYPOTHESIS TESTING
From our observations it is clear that the projected values do not show a significant deviation from the actual ELO values. Even in the cases where the projected value is much higher or lower than the actual values we know that the it is due to certain one off games which have been magnified by the lesser number of datapoints. The same process can be repeated with a dataset where we can obtain games tournaments played in a country without any foreign players. This will help in getting a more accurate understanding of the relation. To sum up, the model follows the natural logistic curve even if tested for different countries/continents.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.