Automatminer: Turn off composition and structure and only use user-supplied features

Is it possible to do the above? (Assuming it even makes sense to do so). A colleague of mine has some custom features vectors, one per material, based on structure and would like to assess their usefulness for matbench tasks. I figured Automatminer would be a reasonable tool to use to “drop in” the new feature vectors.

This would be in addition to running everything together (composition + structure + custom vectors), in which case the custom features may or may not be used.

Open to other suggestions of how to test the utility of a custom set of feature vectors.

Yeah, you can just keep them in the initial dataframe and automatminer should just keep them in until the cleaning stage is reached. At that stage they’re treated like any other feature generated by automatminer. They can be kept, removed, etc. etc. depending on the automatic procedure.

For more info on running your own pipelines (including maybe customizing each of the DFTransformers) to your liking, see this help page:

1 Like