river: Adaptive Random Forest Regressor/Hoeffding Tree Regressor splitting gives AttributeError

Hi @JPalm1, this question is out of the scope of the decision tree module. Could you please open another issue to directly tackle this problem?

Meanwhile, @MaxHalford do you have any idea of what might be happening?

Well, that was quite some reading. Thanks to all involved in the interesting discussion.

Missing data is unavoidable in real-world applications. My comment is in two main aspects:

1. regarding this particular case. When the data dictionaries are created some features are skipped, which results in missing data as the tree has “seen” those features in the past, in other words, they are not “emerging features”. In this case, the missing data is “generated” and it means something (a given condition was met). My suggestion is that the data should be properly set to some value to indicate that the condition was met.

2. In a general sense, missing data imputation is a relevant topic that has been partially addressed in streaming data. Options 1 and 2 above seems reasonable. For simplicity, option 1 does the job and provides protection against missing data, this is the strategy used by XGBoost for example. In practice is like replacing the missing data with a constant value. We can argue that this is not the most elegant imputation method but it does the job. On the other hand, option 2 implies an overhead when processing complex trees and very sparse data.

The HATR is now running smoothly for me with no errors - thanks for taking the time to look into this @smastelini !

It’s not giving me the best predictions, however, but that’s on me to investigate… 😃

(As a side note: is there an existing feature, or planned feature, that would allow you to assign feature types in the input dictionary, which would then allow the different data types to be processed separately but in the same pipeline? IE some get sent to StandardScaler, some get sent to a different encoder? Apologies if this is a feature already, I may have missed it!)

You can check compose.SelectType for that! 😄

Please, also check the examples section of iSOUPTreeRegressor to see that in action.

My bad! I missed one additional place to protect the trees. The fix is in place.

Anyhow, my point is that the error appears because only valid labels should be passed to the learning models. I am assuming that None is being passed as target to the trees in your custom implementation. Am I wrong?

Hey, @smastelini - thanks for looking into it!

That explanation makes sense. Just to note: I’m not set up to use evaluate.progressive_val_score - I haven’t set everything up in a pipeline yet, I am just doing the ‘standard’ approach of iterating through my data points, predicting for each and then evaluating if the delay time has been exceeded. Therefore I’m calling learn_one(x, y) on the ARFR model every time there is an item to be popped out of the delayed evaluation queue.

The changes you have made in #394 seem to have solved the initial exception but pushed the exception to a different location: this time line 223 in numeric_attribute_regression_observer.py

Let me know if you need any more info - I’m having a dig around too to try further my own understanding as well. 😃

@smastelini I guess this is for you 😃