Using conversions about robyn

[Tony-009]

Okay! Yes, that makes sense. I was also looking to see how the mean_spend and mean_reponse were calculated? because that will provide a point of reference(initial attributed conversions) which can be compared with the budget allocator optimised conversion/sales for the future weeks. Appreciate the help!

[gufengzhou]

mean_spend is the mean of all non-zero periods within modelling window. for example, if you have 365 days of data and a new channel is only active since 30 days, then the mean_spend for this channel will be the mean of the 30 days. mean_response is bit more complicated. if it's spend representing the channel, then the pseudo-code is mean_response = hill_function(mean_spend)*ridge_beta_coef. If it's exposure representing the channel, then pseudo-code is mean_response = hill_function(reverse_michaelis_menten(mean_spend))*ridge_beta_coef. Basically, the mean_spend has to go through all transformation to get to mean_response. In the current .func script, it's row 1771- 1839. hope it helps.

[Tony-009]

Okay! Thanks for the in-depth response. This is exactly what I was looking for because something wasn't adding when I was looking at the code and trying to understand the mean_response. Anyways, Appreciate the time and help:-)

[gufengzhou]

hey, first of all, there's been a bug introduced by the sorting issue from the data.table library that is fixed just now. please use the latest update.

Regarding "forecast the conversion", it's done in the budget allocator already. Looking at the example plot in readme, the lower left plot from the _reallocated.png tells you how much of the response (conversions in your case) will increase or decrease per time unit (if you use weekly data, it'll be the change of a week). It's basically saying this: according to your input data and the model you choose, if you shift media budget as recommended, you will expect this much of change in response (conversions) within a week. This is the column optmResponseUnit in the _reallocated.csv