optimize spend lower/upper bounds held constant don't work when spend is not linear to avg media spend * n_time_periods

[ohad-monday]

Im running the demo, and trying to keep one of the channels constant (with lower/upper bounds = 0). It works fine when im using (btw i removed the extra features in the demo): setting lower/upper bounds with 1st channel constant: lower = jnp.repeat(0.2, mmm.n_media_channels) upper = jnp.repeat(0.2, mmm.n_media_channels) lower = lower.at[0].set(0) upper = upper.at[0].set(0) and (same as in the demo): n_time_periods = 10 budget = jnp.sum(jnp.dot(prices, media_data.mean(axis=0)))* n_time_periods

works fine, channel 1 is same (previous=optimal)

BUT, when im using another budget (for example: budget = jnp.sum(jnp.dot(prices, media_data.mean(axis=0)))* 9) channel_1 is not constant (previous!=optimal)

How can i keep channels constraints when the spend is not avg media spend * n_time_periods?

thanks!!

[michevan]

Thanks a lot for flagging this! I just wanted to share a link to our new "support" section in the readme to give you some more information on how we can and cannot support such issues at the moment.

[ohad-monday]

@michevan any idea how to approach this? the readme doesn't help here

[ar-asur]

Hi @ohad-monday - Were you able to fix this? I am also facing similar issue where lower bound/upper bound per channel is not met in the optimization.