Since January 8, the Blender has performed the worst overall season to date on Totals.
Today as I was experimenting with some things I noticed that I had trained the Blender Regression Model on the wrong dataset. I am not saying this change is all of a sudden going to make it amazing, but the Blender has performed the best on spread and money line bets this season, so I expect this change to hopefully get it closer to that.
I noticed the mistake when I was testing a Random Forest model(for spread bets) and despite my best efforts, my simple Blender Linear Regression model continued to beat the Random Forest test on residual standard error each time, no matter what test set I was using. So, the Linear Regression approach is still king of the mountain for spread bets in that regard. If I do still decide to add a Random Forest model going forward, it will probably be in the offseason for next season. I do plan on trying one for MLB season, however.