In our last video, we talked about why features are so important for machine learning and materials informatics. In this video, we dive into the composition-based feature vector. Composition is a useful thing to construct a feature vector with since we don't always know the structure- particularly for new materials. This video describes several different tools for creating a composition-based feature vector including JARVIS, MAGPIE, mat2vec, atom2vec, oliynyk, and more. We show where these approaches originate from and how to actually implement them using our slick CBFV package in python which can be pip installed! See more here
[ Ссылка ]

Check out the whole materials informatics series at [ Ссылка ] with workbooks and course notes available at [ Ссылка ]

0:00 what is a CBFV?
1:30 CBFV from elemental properties
5:00 using CBFV to predict properties
8:00 non-obvious correlations
10:15 jarvis vs oliynyk vs magpie vs atom2vec vs mat2vec vs elemnet
19:00 does domain knowledge CBFV improve predictions?
25:21 example of using CBFV library in python

свернуть