Materials genomics approaches to materials design emphasize data production, data organization, and new types of data analysis based on large databases of information. In this talk, I will present several new open-source codebases being developed at LBNL to aid researchers in all aspects of this process (all codebases are currently available at https://www.github.com/hackingmaterials). The FireWorks and MatMethods codebases allow one to automatically generate workflows that calculate multiple materials properties given only a crystal structure. Furthermore, users can scale those computations to millions of jobs at supercomputing clusters. The FireWorks codebase is a completely general platform for automating any type of calculation, whereas MatMethods contains materials science workflows implemented using FireWorks (currently interfacing with the VASP and LAMMPS software packages). Next, I will talk about MatMiner, which is a code that is specifically aimed at data extraction, data analysis, and machine learning. MatMiner allows researchers to quickly obtain data from a variety of online resources (the Citrine database, the Materials Project database, and general MongoDB databases are currently supported). MatMiner formats results from these databases into a "pandas" DataFrame object, which is a standard tool in the Python data analytics community for selecting, grouping, and concatenating data. Next, researchers can automatically decorate the entries with descriptors that we have developed for the purposes of data mining. Finally, researchers can plot results using a plug-in we have developed to the Plot.ly library for interactive HTML plots or use the sci-kit learn package to develop multiple types of machine learning models. I will present a demo showing how these tools can be used to perform simple studies within a day that once may have taken several weeks to complete.

Open-source tools for generating and analyzing large materials data sets: FireWorks, MatMethods, and MatMiner

253rd National Meeting (2017)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.