You can find materials supporting this vid (and others) at github.com/equitable-equations/youtube. Here's the link to my vid on the {broom} package: ruclips.net/video/Oy1_A_ZhCY0/видео.html
Excellent explanation and well presented, appreciated! Would be nice to make a video about using formula in smoothing function in ggplot2, e.g. sometimes square term is used as a predictor in poisson regression and then just glm method in smoothing function will not be sufficient.
Andrew, Your point about the missings is interesting but it only amounts to a percent (113 hrs) and may thus not be worth trying to impute. Besides that there could be simple explaination for the missing hours. Perhaps the bike company was closed for a holiday and/or buisness-related activities. Imputation might therefore not be warranted.
Yep, totally! I actually have another vid set for release next week that talks about all of this before jumping in to add zeros. Those missing values are pretty much all at odd hours in the middle of the night. The effect on the model is on the order .7 bike per hour, which is surprisingly high.
You can find materials supporting this vid (and others) at github.com/equitable-equations/youtube. Here's the link to my vid on the {broom} package: ruclips.net/video/Oy1_A_ZhCY0/видео.html
Good discussion - especially of why you chose to use Poisson.
There is nothing better than good old statistics in the era of ml/nn
At the very end of the video I think you were hinting to a zero inflated model to account for the zeros...
Thank you for this Prof.
Good stuff!
Excellent explanation and well presented, appreciated!
Would be nice to make a video about using formula in smoothing function in ggplot2, e.g. sometimes square term is used as a predictor in poisson regression and then just glm method in smoothing function will not be sufficient.
I totally agree! On my list for sure.
I have a vid about the ggplot loess smoother coming out next month, so maybe after that would be the right time.
Andrew, Your point about the missings is interesting but it only amounts to a percent (113 hrs) and may thus not be worth trying to impute. Besides that there could be simple explaination for the missing hours. Perhaps the bike company was closed for a holiday and/or buisness-related activities. Imputation might therefore not be warranted.
Yep, totally! I actually have another vid set for release next week that talks about all of this before jumping in to add zeros. Those missing values are pretty much all at odd hours in the middle of the night. The effect on the model is on the order .7 bike per hour, which is surprisingly high.
@@EquitableEquations By the way I see you are in the Chicago region teaching. I attended uni up there and have family in the area.
There is only ONE value with heavy rain/ snow so why not just skip it??? (run Bikeshare |> count(weathersit) for a table).