This is a project seeking to figure out whether the 9-3 curse is real. The 9-3 curse is a superstition in the Valorant community that if you get to a score of 9-3 favourable to you in the first half, you will lose. I am using Joshua Broas' dataset to extract match IDs, then using selenium (maybe not the best idea) to scrape vlr.gg for the score timeline.
git clone https://github.com/FrankWhoee/93.git
cd 93
python3 venv venv
(requires you to have venv installed)source venv/bin/activate
pip install requirements.txt
To begin collecting data:
python3 main.py
- Data will be saved every 50 matches to data.pickle.
To begin analysis:
python3 analysis.py
- If you want to run analysis on a custom file, do:
python3 analysis.py example.pickle
data.pickle
contains lists of game timelines, where for each game, 1 is a CT win, and -1 is a T win.
As of 8:55pm PDT these are the results:
Sample size: 5137
- Chance of losing from 7-5: 0.23533778767631774
- Chance of losing from 8-4: 0.1242344706911636
- Chance of losing from 9-3: 0.058091286307053944
- Chance of losing from 10-2: 0.025185185185185185
- Chance of losing from 11-1: 0.003367003367003367
- Chance of losing from 12-0: 0.0
Data for this particular sample is availabe here. Stored as a pickle of a 2D array (a list of game timelines).
The naive interpretation is that there is no 9-3 curse because the probability of losing at 9-3 is not higher than either 8-4 or 7-5, which our preliminary data clearly proves. A more nuanced interpretation would be finding the expected probability of losing given a score of 9-3 is lower than the empirical data.
This graph shows there is no noticeable difference in loss probability at 9-3, and our data fits the trendline pretty well, so the probability of losing a 9-3 game is the same as expected.
All of this data is pulled from pro/semi-pro matches as Valorant lacks a way to collect data en masse for ranked games, so it's possible that the phenomenon felt in ranked doesn't show up in pro/semi-pro settings.
Thanks to Joshua Broas for the dataset.
Work sponsored by SDN. Join SDN today.