Number of features in matminer vs. literature magpie

Hello all,

I’m looking at using matminer to featurize compositions. The ElementProperty.from_preset(preset_name='magpie') matminer method returns 132 descriptors. However, the original Magpie paper (A general-purpose machine learning framework for predicting properties of inorganic materials | npj Computational Materials) says that it generates 145 descriptors. Is there a list somewhere of which descriptors matminer is not calculating?


from matminer.featurizers.composition import ElementProperty

magpie_properties = ElementProperty.from_preset(preset_name='magpie')
len(magpie_properties.feature_labels())  # returns 132

Thank you!

Hmm, I’m not sure. You might check out the readme here which has a bunch of the features matminer/matminer/utils/data_files/magpie_elementdata at main · hackingmaterials/matminer · GitHub