No, kpi_without_optim should behave similarly between geo and national cases.

Some of the inputs provided would had to be different at some point for those two to provide such different outputs.

from lightweight_mmm.

no I tried complete data with extra features and just last 52 weeks and extra features. You can see from below that all other variables are same

Run optimization with the parameters of choice.

solution, kpi_without_optim, previous_budget_allocation = optimize_media.find_optimal_budgets(
n_time_periods=n_time_periods,
media_mix_model=mmm,
extra_features=extra_features,
budget=budget,
prices=prices,
bounds_lower_pct=0.3,
bounds_upper_pct=jnp.array([1,0.3580]),
media_scaler=media_scaler,
target_scaler=target_scaler,
seed=SEED)

from lightweight_mmm.

I would need a reproducible example (can be a colab with mock data) in order to investigate further, its too broad many things play in on this one.

from lightweight_mmm.

please give me an email id so that I can share my colab notebook

from lightweight_mmm.

Im sorry but that is not something we can do. If the error is reproducible, feel free to provide any mock data and code to reproduce and we can look into it.

from lightweight_mmm.