Interpreting CLV Frequency-Recency heatmap

[AMeynen]

I need some help in the interpretation of this Frequency/Recency graphs of the same beta-geometric model.

The left plot is a plot with max_recency = 365, the right plot is a plot with max_recency = 730. The documentation states the following about the max_recency parameter: max_recency: The maximum recency to plot. This also determines the age of the customer. Default to max observed age.

I find it striking that the expected number of future purchases for the same recency-frequency differs between the two graphs. E.g. Recency = 350, Frequency = 50 -> 4 on first graph, 0 on second graph.

My current interpretation is that this plot shows the expect number of purchases if the maximum observed recency was equal to max_recency. Could someone confirm whether my interpretation is correct?

The reason why I have to limit the max_recency and max_frequency is due to memory issues. Solutions for this might be helpful as well.

KR
Alex

[ColtAllen]

Hey @AMeynen,

Was this model fit with the default MCMC? If so, then chain X draw computations are probably being made for every value in the plotted matrix, which could explain the memory issues. Try plotting with model.fit(fit_method = 'map') and let us know what happens.

My current interpretation is that this plot shows the expect number of purchases if the maximum observed recency was equal to max_recency. Could someone confirm whether my interpretation is correct?

Your interpretation is correct. This plotting function also sets T=max_recency, which is why the results appear so different between the plots. T is an important parameter, and assuming every customer is the same age exposes a fundamental limitation in this plot, so thanks for bringing it to our attention. It may be better if this plot were 3-dimensional.