[BUG] (null) Column in markov matrix

[sergiotj]

Describe the bug
Markov matrix is wrong in the (null) column. All the values are 0.

To Reproduce
Steps to reproduce the behavior:

1. Create markov attribution
2. Print markov matrix
3. Look at values for (null) column
4. See error

Expected behavior
That column should have the frequency of times that the other channels end in (null)

Screenshots

[Nathandsn]

Can you send us a full notebook file with this error or at least a print of the full matrix? I wasn't able to reproduce the issue...

What do you have at the "(null)" row? I'm asking because this seems to be the expected behavior of the '(null)' row, your matrix may just be the transpose.

[MoritzMohrStade]

I discovered the same behavior as @sergiotj for all different datasets I tried, even when using a random dataframe with:
attributions = MAM(random_df=True)
The column (null) shows 0s for every row except the (null) to (null) transition. My expectation would be to have the non-null values for probabilities where people did not buy.
This is a huge bug because it would imply that there are no non-converting transitions that end in (null).

My colab for the random dataframe:
https://colab.research.google.com/drive/1_FEcffn-pdh4r9xo5BPfaTQ3q2T9b7k-?usp=sharing

To reproduce use:

• GA4 Public BigQuery Dataset for the Google Merchandise Store
• random dataframe

[yanMHG]

The root of the issue is in the conversion_value_as_frequency parameter of the attribution_markov method.

When the value of the parameter is True, which is the default, the method does not count the transitions ending in (null), since their conversion value is 0. To circumvent the issue, the value must be set to False, which forces standard counting.

Technically not a bug, but this choice of default behavior does feel a bit awkward/unexpected to me, specially because it does not seem to apply to the examples in the README.