MaterialsImages

This repository contains code for handling the materials images and spectra dataset that is the largest expermental materials science dataset containign composition, images, UV-Vis spectra and processing information for about 180.000 compositions. The stand alone dataset can be found here and is published under the creative commons cc-by license. The data was aquired at the Joint Center for Artificial Photosynthesis at Caltech.

Installation

You'll need to install h5py and tqdm in a python >=3.6 environment via

pip install h5py tqdm

Usage

During the first run the script will try to download the dataset to the current work directory. To use this script simply run

from MaterialsImages import MaterialsImages

depending on your internet connection this might take a while. If you want to know which attributes to this dataset are available run this code that will also tell you the shape of each attribute

m = MaterialsImages()
m.show_datasets()

There are come convenience functions build in like get spectrum that will return a spectrum (here number 1337)

spectrum = m.get_spectrum(1337)

The same functions exist for image and loading. A special function is get_composition_string which will return a materialsproject compatible composition spring. Please note that we do not know the actual Oxygen content so that the string does not contain Oxygen.

composition_string = m.get_composition_string(1337)

If you have sufficient RAM (ca. 16GB should suffice) you can also load al spectra,images etc. into memory by calling

m.load_all_spectra_to_memory()
m.load_all_compositions_to_memory()
m.load_all_images_to_memory() #this will take a while

The images and spectra are then accessibile (faster) by calling m.spectra[1337] etc.

Citation

When using this data for a scientific publication please consider citing the data descriptor published in Scientific Data:

Stein, H.S., Soedarmadji, E., Newhouse, P.F., Guevarra, D., Gregoire, J.M., Synthesis, optical imaging, and absorption spectroscopy data for 179072 metal oxides

This dataset was first used in our publication on variational and generative autoencoders to use materials image embeddings for spectrum prediction and vice versa. Please consider citing:

Stein, H.S., Guevarra, D., Newhouse, P.F., Soedarmadji, E., Gregoire, J.M., Machine learning of optical properties of materials
– predicting spectra from images and images from spectra Chemical Science, 2019, 10, 47 DOI: 10.1039/c8sc03077d