QM9 数据集性质预测的单位

           QM 9dataset unit:   property_unit_dict = {
            QM9.A: "GHz",
            QM9.B: "GHz",
            QM9.C: "GHz",
            QM9.mu: "Debye",
            QM9.alpha: "a0 a0 a0",
            QM9.homo: "Ha",
            QM9.lumo: "Ha",
            QM9.gap: "Ha",
            QM9.r2: "a0 a0",
            QM9.zpve: "Ha",
            QM9.U0: "Ha",
            QM9.U: "Ha",
            QM9.H: "Ha",
            QM9.G: "Ha",
            QM9.Cv: "cal/mol/K",
        }

如果做3D分子性质预测的时候，需要用HAR2EV=27.211386246进行转换，然后1000meV = EV, 因此最后MAE的loss 应该*1000.

     HAR2EV = 27.211386246
     KCALMOL2EV = 0.04336414
     
     conversion = torch.tensor([
         1., 1., HAR2EV, HAR2EV, HAR2EV, 1., HAR2EV, HAR2EV, HAR2EV, HAR2EV, HAR2EV,
         1., KCALMOL2EV, KCALMOL2EV, KCALMOL2EV, KCALMOL2EV, 1., 1., 1.
     ])

     https://github.com/atomistic-machine-learning/schnetpack/blob/master/src/schnetpack/datasets/qm9.py