Beautiful plot_calibration_purchases_vs_holdout_purchases plot! But disappointing plot_cumulative_transactions

[invictus2010]

I trained a BGF fitter against 30 months of customer data. I held back 1 month of data. Look how amazing this fit appears:

Then, I ran plot_cumulative_transactions and see this

Much bigger disparity in actual v forecasted than I hoped to see. Is this an example of overfitting my data? Or is there something else going on?

[ryanbonifacino]

@invictus2010 if you plot from cohorts on both sides of your first month cohort, do you see the similar disparity?