def get_qm7_energies():
"""
Loads the energies of the molecules of the QM7 dataset.
Returns
-------
energies: numpy array
array containing the energies of the molecules
"""
qm7 = fetch_qm7()
return qm7['energies']
def get_qm7_positions_and_charges(sigma, overlapping_precision=1e-1):
"""
Loads the positions and charges of the molecules of the QM7 dataset.
QM7 is a dataset of 7165 organic molecules with up to 7 non-hydrogen
atoms, whose energies were computed with a quantun chemistry
computational method named Density Functional Theory.
This dataset has been made available to train machine learning models
to predict these energies.
Parameters
----------
sigma : float
width parameter of the Gaussian that represents a particle
overlapping_precision : float, optional
affects the scaling of the positions. The positions are re-scaled
such that two Gaussian functions of width sigma centerd at the qm7
positions overlapp with amplitude <= the overlapping_precision
Returns
-------
positions, charges, valence_charges: torch arrays
array containing the positions, charges and valence charges
of the QM7 database molecules
"""
qm7 = fetch_qm7(align=True)
positions = qm7['positions']
charges = qm7['charges'].astype('float32')
valence_charges = get_valence(charges)
# normalize positions
min_dist = np.inf
for i in range(positions.shape[0]):
n_atoms = np.sum(charges[i] != 0)
pos = positions[i, :n_atoms, :]
min_dist = min(min_dist, pdist(pos).min())
delta = sigma * np.sqrt(-8 * np.log(overlapping_precision))
positions = positions * delta / min_dist
return (torch.from_numpy(positions), torch.from_numpy(charges),
torch.from_numpy(valence_charges))
# Finally, we import the utility functions that let us access the QM7 dataset # and the cache directories to store our results. from kymatio.datasets import fetch_qm7 from kymatio.caching import get_cache_dir ############################################################################### # Data preparation # ---------------- # # Fetch the QM7 database and extract the atomic positions and nuclear charges # of each molecule. This dataset contains 7165 organic molecules with up to # seven non-hydrogen atoms, whose energies were computed using density # functional theory. qm7 = fetch_qm7(align=True) pos = qm7['positions'] full_charges = qm7['charges'] n_molecules = pos.shape[0] ############################################################################### # From the nuclear charges, we compute the number of valence electrons, which # we store as the valence charge of that atom. mask = full_charges <= 2 valence_charges = full_charges * mask mask = np.logical_and(full_charges > 2, full_charges <= 10) valence_charges += (full_charges - 2) * mask