If someone knows any time-series stats I could run on it, let me know.
There’s a few different things you could be interested in here for ‘time series segmentation’, which slightly shift what sort of method you want to approach.
Identifying the structural breaks. Basically, you can view pom count as drawn from some distribution, but which distribution changes over time.
Identifying local factors. For example, maybe Mondays are persistently different from Tuesdays, or days when you log more poms are followed by a day when you work fewer poms.
Often, people use ARMA models for time series because they can easily capture lots of different local factors, and HMMs for structural breaks (when you have as many as you do). I’m not aware of standardized methods that are good for this problem because often there’s lots of tweaks inherent to your distribution; take a look at all the detail in this accepted answer, for example. But you also might be able to stick your data into seglearn and get something cool out of it.
There’s a few different things you could be interested in here for ‘time series segmentation’, which slightly shift what sort of method you want to approach.
Identifying the structural breaks. Basically, you can view pom count as drawn from some distribution, but which distribution changes over time.
Identifying local factors. For example, maybe Mondays are persistently different from Tuesdays, or days when you log more poms are followed by a day when you work fewer poms.
Often, people use ARMA models for time series because they can easily capture lots of different local factors, and HMMs for structural breaks (when you have as many as you do). I’m not aware of standardized methods that are good for this problem because often there’s lots of tweaks inherent to your distribution; take a look at all the detail in this accepted answer, for example. But you also might be able to stick your data into seglearn and get something cool out of it.