Show the learning improvement plots for each iterations

[OmkarEllicium]

Hi Team,

Is it possible to show how much the model is improving in each iteration?

[OmkarEllicium]

@gufengzhou any idea?

[gufengzhou]

There're many already. Among the exported plots:

• pareto_front.png shows convergence of NRMSE & RSSD over time (same as iterations)
• ROAS_convergence1.png shows convergence of ROAS over iteration
• ts_validation.png shows NRMSE convergence over iteration

Or what do you mean by improving?

[OmkarEllicium]

Any other way than saving PNGs? My data is large so while saving PNG it basically crashes.
Which files are used to plot graphs shown in these PNGs?

[gufengzhou]

Robyn provides very comprehensive raw outputs and I'd recommend you to explore them. For example you can

• use the table pareto_hyperparameters.csv, (schema of exported tables here),
• or after robyn_run, check OutputModels$convergence$moo_distrb_plot$data, or OutputModels$convergence$moo_cloud_plot$data
• or after robyn_outputs, check OutputCollect$resultHypParam

[OmkarEllicium]

Thanks, Will look into this

[OmkarEllicium]

@gufengzhou this is how NRMSE looks over all iterations:

It this expected? Shouldn't it be a smoother line with larger value at the start and slowly coming down?

[gufengzhou]

No it's not expected. Yes it should improve/ minimise. The R version looks like this below. Would you please try running more iterations? 500 is not enough usually. Are you using own data or our sample data? Our sample data usually converges after 1-2k.

[OmkarEllicium]

@gufengzhou I am using my own data

[gufengzhou]

please try more iterations and hope it looks better. btw the exported ts_validation.png is showing this already, in case you haven't found it