porous materials AI gym

open data sets for machine learning pertaining to porous materials.

crystal structures

MOFs: CoRE MOFs (Paper, Database), CSD MOF Subset (Paper, Database), CSD MOF Collection (Paper, Database)
COFs: CURATED COFs (Paper, Database), CoRE COFs (Paper, Database)
zeolites: IZA database (Database)

MOFs: B&W (Paper, Database), ToBaCCo (Paper, Database, Code), hMOFs (Paper, Database), Anderson et al. (Paper, Database), MOF-5 analogues (Paper, Database), PORMAKE (Paper, Code)
COFs: Mercado et al. (Paper, Database), Haranczyk's 3D COF database (Paper, Database)

material class	target y	features x provided?	Reference	size of data set
MOFs (hypothetical)	CO2, N2 adsorption (sim)	yes	Paper, Database	ca. 325,000
MOFs (experimental and hypothetical)	Band gaps, density of states, charge densities (sim)	yes	Paper, Database	ca. 18,000
MOFs (experimental)	Color (exp)	yes	Paper, Database	?
COFs (hypothetical)	CH4 deliverable capacity (sim)	yes, hand-crafted features provided.	Paper, Database	ca. 70,000
COFs (experimental)	CH4, H2, O2, Xe, Kr, H2S adsorption (sim)	?	Paper	ca. 500
MOFs (hypothetical)	H2 adsorption (sim)	yes	Paper / Database	ca. 100K
MOFs (experimental)	thermal stability, solvent removal stability	yes (RAC & geometric features)	Paper / Database	ca. 2-3K (extracted from experimental lit)
MOFs (experimental)	CO2, H2O DFT-calculated adsorption energy	no	Paper, DataBase	ca. 8400 MOFs, but 38M DFT calcs

material class	target y	Reference	size of data set (# materials)
MOFs (experimental)	DDEC6 charges on atoms (sim)	Paper, Database	ca. 3,000
MOFs (experimental and hypothetical)	DDEC6/CM5/Bader charges on atoms (sim)	Paper, Database	ca. 18,000 (DDEC6/CM5), ca. 5,000 (Bader)
MOFs (experimental and hypothetical)	Effective bond orders on atoms (sim)	Paper, Database	ca. 18,000
MOFs (experimental)	Formal oxidation states on atoms (exp)	Paper, Database	ca. 49,000

see matminer here. pointed out by Jack Evans.

here is a list of open-source codes for building your own crystal structure models.