Robyn Bugs in Allocator and Response Functions

[mekarim]

Attention: @laresbernardo, @gufengzhou , @michellegrushkometa

In the demo.R, if I run the following code it does not use media spend data to document the (mean, %share) spend in the outputs. Currently using (v3.12.1 - the latest I believe).

Instead, it uses whatever is in the paid_media_selected, which is paid_media_vars for the demo example. We know paid_media_vars can have impressions, spend, clicks everything mixed up. So the spend metrics here (line: 215 to 241) will have mixed units ... thus not correct and makes no sense. Or am I thinking it wrong?

Given such, can I trust any of the numbers that this function outputs? Or its just the spend numbers which are off, and ROAS, or CPA will be correct?

attribution_window_start_date <- '2018-10-01'
attribution_window_end_date <- '2018-12-31'

allocator_growth <- robyn_allocator(
    InputCollect = InputCollect
    , OutputCollect = OutputCollect    
    , scenario = "max_response"
    , select_model = '3_56_25'
    , channel_constr_low = 0.5
    , channel_constr_up = 2    
    , date_range = c(attribution_window_start_date
                    , attribution_window_end_date)
)

print(allocator_growth$plots[2])

Thanks

[mekarim]

Should the following code match the output of the initial unoptimised part of allocator_growth$plots[2]?

select_model <- "3_56_25"
channel_spend = "facebook_S"

window_loc <- InputCollect$rollingWindowStartWhich:InputCollect$rollingWindowEndWhich
dt_optimCost <- slice(InputCollect$dt_input, window_loc)
date_range = tail(dt_optimCost$DATE, 14)

histFiltered <- filter(dt_optimCost, DATE >= min(date_range) & DATE <= max(date_range))

## Get the "next dollar" marginal response on spend1
spend1 = sum(histFiltered[channel_spend])
response1 <- robyn_response(
  InputCollect = InputCollect
  , OutputCollect = OutputCollect
  , select_model = select_model
  , metric_name = channel_spend
  , metric_value = spend1
  , date_range = date_range
  , is_allocator = TRUE
)

response2 <- robyn_response(
  InputCollect = InputCollect
  , OutputCollect = OutputCollect
  , select_model = select_model
  , metric_name = channel_spend
  , metric_value = spend1 * 1.1
  , date_range = date_range
  , is_allocator = TRUE
)

mroas = (response2$sim_mean_response - response1$sim_mean_response)/(response2$sim_mean_spend - response1$sim_mean_spend)
mroas

[mekarim]

Finally does the robyn_response mean response output considers seasonality within the timeline defined by date_range?

Context: geometrics adstrock.