Hi! I tried working out the formula at 3:45. Shouldn't it be y_hat_t+1 = a*y_t + (1-a)*y_hat_t ? This would give the smoothing formula we looked at earlier on.
Nice. I'm starting to figure this out. Doing some NFL QB stats and want to forecast/predict/guess the next game stats. Looked at XGB, then ARIMA and learned enough to know I want a time series forecast but my stats don't have a good trend, seasonal yes but not seasonal in the model sense and no cycle. Much variation week to week. The only influencing factor I am using is the opposing team passing defense ranking. Will switch to this model. I'm still wondering if I can feed the defensive factor in to train the model and have it predict and adjusted outcome.
@@egorhowell No problem. I think I realized that some problems just don't need a model. If the model is just getting an average for instance I can just do the math myself. I was just working on QB passing stats and predicting the stat for the next game. I just went with the next team on the schedule defensive stats to adjust the prediction using the cumulative average for the stat up to that point. But I enjoyed learning about a few models and played with data that had trend, seasonality and cycle. Thanks for sharing.
Hi! I tried working out the formula at 3:45. Shouldn't it be y_hat_t+1 = a*y_t + (1-a)*y_hat_t ? This would give the smoothing formula we looked at earlier on.
Great video, Egor
Thanks Bonnie!
Nice. I'm starting to figure this out. Doing some NFL QB stats and want to forecast/predict/guess the next game stats. Looked at XGB, then ARIMA and learned enough to know I want a time series forecast but my stats don't have a good trend, seasonal yes but not seasonal in the model sense and no cycle. Much variation week to week. The only influencing factor I am using is the opposing team passing defense ranking. Will switch to this model. I'm still wondering if I can feed the defensive factor in to train the model and have it predict and adjusted outcome.
Hey, sorry but I don’t have enough context without the code, data etc. to help on this :)
@@egorhowell No problem. I think I realized that some problems just don't need a model. If the model is just getting an average for instance I can just do the math myself. I was just working on QB passing stats and predicting the stat for the next game. I just went with the next team on the schedule defensive stats to adjust the prediction using the cumulative average for the stat up to that point. But I enjoyed learning about a few models and played with data that had trend, seasonality and cycle. Thanks for sharing.
no problem, glad you found it useful :)