The data contained inside SPAHM+.tar are organized in 2 folder-types corresponding to 2 datasets:

A) Folder containing the structures and SPAHM representations
[Azophotoswitches (APS) - S. Vela, A. Fabrizio, K. R. Briling, and C. Corminboeuf, J. Phys. Chem. Lett. 12, 5957–5962 (2021)]
1) azoswitch_full/
	geometries/		: directory with the structures (xyz format)
	repr-neutral/		: directory with the QML representations (NumPy binary)
	repr-cations/		: directory with the script to generate the SPAHM representations of the radical cations
	repr-anions/		: directory with the script to generate the SPAHM representations of the radical anions
[QM7 dataset - M. Rupp, R. Ramakrishnan and O. A. von Lilienfeld, J. Phys. Chem. Lett., 2015, 6, 3309–3313]]
2) qm7/
	geometries/		: directory with the structures (xyz format)
	repr-neutral/		: directory with the QML representations (NumPy binary)
	repr-cations/		: directory with the script to generate the SPAHM representations of the radical cations

IMPORTANT NOTE: 	Due to the size-limit of the archive it is not possible to provide all representations.
			We provide the SPAHM(a,b) representations for the neutral copounds of both databases (QM7 and azoswitch).
			The representations for the radical cations and anions of the same structures can be generated using the dedicated script.

B) Folders containing the computed properties, learning curves, hyperparameters for a specific dataset (the same folder architecture are gathered together for simplicity)
./azoswitch_full__excited		: result for the excited states of the APS dataset
./azoswitch_full__neutral		: result for the APS dataset
./azoswitch_full__cations/		: result for the radical cations of the APS dataset
						./1_hyper 	: hyperparameters used for each regression (plain text/json)
						./2_lc		: learning curves obtained for each regression (plain text/json)
						./y		: directory with the lists of target molecular properties (plain text)
						./dft		: directory with the raw dft calculations output (plain text)
						./oos		: directory the prediction for the out-of-sample (OOS) structure (structured Dir)
qm7-cations/cations_slatm		: result for the QM7-RC dataset using the SLATM representation
qm7-cations/cations_atom		: result for the QM7-RC dataset using SPAHM(a) representations
qm7-cations/cations_bond		: result for the QM7-RC dataset using SPAHM(b) representations
qm7-cations/neutral+cations		: result for the QM7/2+QM-RC/2 dataset using SPAHM(a) representations
qm7-cations/cations_soap		: result for the QM7-RC dataset using the SOAP representation
qm7-neutral/atom_representation/	: result for the QM7 dataset using the SPAHM(a) representation
qm7-neutral/bond_representation/	: result for the QM7 dataset using the SPAHM(b) representation
						./1_hyper 	: hyperparameters used for each regression (plain text/json)
						./2_lc		: learning curves obtained for each regression (plain text/json)
						./y		: directory with the lists of target molecular properties (plain text)