Hello everyone,
I have ultimately decided to not do a GCC recap as to how the model performed. The primary reason being that it was power-lagged and not power-matched like my model expected. This allowed weird anomalies such as two 6-0's and one 5-1 in LD. Also, I want to use my time to focus more on my future NITOC predictions and other things. If anyone wants to do an accuracy test for me I'm willing to go ahead and post it anyway.
I also want to update the model in a couple of significant ways that I believe will prove to be more accurate. First, an adjustment to make the model more confident about those debaters who have attended more tournaments. If you prove to be doing consistently well at many tournaments, the model should be more confident about your performance. To adjust for this I will add one win and one loss to everyone's record. This will more greatly affect those debaters who have debated less and make the model less confident about their chances if they're good, and more confident about their chances if they're not as good. I also don't want debaters to be continually plagued by past performances if they've improved significantly. It's hard to quantify this or even say that anyone has improved significantly in the course of the year. Nonetheless, I'm considering either limiting it to the past six performances or only to performances during the current semester. I haven't decided which if any yet but I would appreciate feedback on it.
Dylan
I have ultimately decided to not do a GCC recap as to how the model performed. The primary reason being that it was power-lagged and not power-matched like my model expected. This allowed weird anomalies such as two 6-0's and one 5-1 in LD. Also, I want to use my time to focus more on my future NITOC predictions and other things. If anyone wants to do an accuracy test for me I'm willing to go ahead and post it anyway.
I also want to update the model in a couple of significant ways that I believe will prove to be more accurate. First, an adjustment to make the model more confident about those debaters who have attended more tournaments. If you prove to be doing consistently well at many tournaments, the model should be more confident about your performance. To adjust for this I will add one win and one loss to everyone's record. This will more greatly affect those debaters who have debated less and make the model less confident about their chances if they're good, and more confident about their chances if they're not as good. I also don't want debaters to be continually plagued by past performances if they've improved significantly. It's hard to quantify this or even say that anyone has improved significantly in the course of the year. Nonetheless, I'm considering either limiting it to the past six performances or only to performances during the current semester. I haven't decided which if any yet but I would appreciate feedback on it.
Dylan
Comments
Post a Comment