the use of statistics like range,mode,max and min creates discontinous boundaries in compositional phase spaces between binaries are ternaries. Is it possible to limit the featurization to smooth / continous features to avoid this problem?
I hope this was clear I will clarify if needed!
Hmm, could you elaborate a bit more on what you have in mind?
You could just remove those stats after the fact; for example, retaining only the mean features, before passing into the next stage of what you’re doing, whether that be DataCleaner or your own code…
You can use matminer alone to also define your own ElementProperty using different statistics, such as only Holder means and arithmetic means (see matminer PropertyStats for more available stats), and just apply that featurizer alone. If you have multiple of these kinds of featurizers, you can put them in an AutoFeaturizer to apply them all at once. The Autofeaturizer code documentation should have specifics on how to do this.
Let me know if you use either of the above two options and need help, I’m glad to provide more detail if needed.