"Solving" Wordle from Shared Scores at 100%* Accuracy
3 days ago
- #Wordle
- #Machine Learning
- #Data Analysis
- The author developed a system to solve Wordles using shared scores from social media, initially inspired by a Kaggle project.
- The approach involves analyzing patterns in shared Wordle scores to deduce the correct answer, using word frequency data and penalty terms for impossible patterns.
- Transitioning from Twitter to Bluesky for data collection resulted in fewer shared scores, impacting the system's accuracy.
- The system was updated with additional features like fraction of found patterns and Kolmogorov-Smirnov test statistics to improve accuracy.
- Despite improvements, the system's accuracy is not 100%, with occasional failures due to edge cases and insufficient data.
- The author acknowledges the system's limitations and the ongoing need for adjustments to maintain accuracy.