How to interpret bgf.summary

Hi, I have 2 questions, please help me with some guidance:

1. How to read bgf.summary? what does each column and row mean?

   coef	se(coef)	lower 95% bound	upper 95% bound
   0.271124	0.003597	0.264073	0.278175
   156.815975	3.151111	150.639797	162.992154
   0.331748	0.014789	0.302761	0.360735
   1.176144	0.061768	1.055079	1.297209

2. I’ve tried to change the penalizer_coef, below is set at zero, the more I increase it, the worse it is.

   

I need some guidance on closing the model’s prediction gap. Which part do most beginners get wrong?

Thanks

[psygo]

1. How to read bgf.summary? what does each column and row mean?

The most important part is the first column, the coefficients themselves. The other columns refer to the errors around their estimation. Imagine each coefficient is a Gaussian distribution: se(coef) is the standard error, the sigma, it gives you an estimate as to how good your model conversion was. The lower and upper bounds are the range of values the coefficients are assuming under 95% confidence. If the range is too high, you might want to check for better convergence.

1. I’ve tried to change the penalizer_coef, below is set at zero, the more I increase it, the worse it is.

Usually, the penalizer_coef can vary somewhat comfortably from 0.001 to 0.5 — though it would be nicer to have it at max around 0.1. How high did you go? Too high and the model migh basically become useless.

Also, I would say that you should focus on making your model fit well with the 0 and 1 groups of the holdout vs calibration purchases graph, as they represent 70-90% of the customers usually. There are other graphs that will indicate your model’s performance also, try checking them out in conjunction with this graph.

[GrowthJeff]

Also, I would say that you should focus on making your model fit well with the 0 and 1 groups of the holdout vs calibration purchases graph, as they represent 70-90% of the customers usually. There are other graphs that will indicate your model’s performance also, try checking them out in conjunction with this graph.

I found this very insightful. Thank you!