Nevergrad about robyn

[gufengzhou]

Hi, regarding Nevergrad setting, I'd really recommend to study Nevergrad documentation and ask questions to the Nevergrad team where you can get the best possible consultation.

Regarding RSSD, or DECOMP.RSSD, it's short for "decomposition root sum of squared distance", a metric we "invented" to account for business logic. The intuition is this: assuming you're spending 90% on TV and 10% on FB. If you get 10% effect for TV and 90% for FB, you'd probably not believe this result, no matter how low the model error (NRMSE) is. If you get 80% TV and 20% FB as effect share, it'll more "realistic". This is where the logic is from: minimising the distance between share of spend and share of effect. It's really about getting rid of the very extreme cases and have a set of results that are more realistic.

[gufengzhou]

Hi, Nevergrad is gradient-free/evolutionary, meaning it repeats mutation of our hyperparameters towards better objective scores (minimising NRMSD & RSSD in our case). To be precise, the function optimizer$ask() will provide a set of candidate hyperparameters that is based on last mutation and optimizer$tell() is where we feed in the scores (NRMSE, RSSD) of the current mutation. For more Nevergrad details please refer to Nevergrad's documentation: https://facebookresearch.github.io/nevergrad/optimization.html#ask-and-tell-interface

[wenyanyy]

Understood~~.Thanks a lot

[wenyanyy]

Hi Gufeng zhou,
Two more questions for Nevergrad. 1. Is it possible to set mutation or step for choosing parameters in nevergrad? 2. you mentioned minimizing RSSD and could you explain the reason that why I need to chose the minimum RSSD and what's the meaning for having minimum RSSD?

Thank you so much

Best,