Basic unit conversion for experiments #11
Conversation
| "pbe-ts", | ||
| "ccsd/cc-pvdz", | ||
| "ccsd(t)/cc-pvdz", | ||
| # "pbe+mbd/light", #MD22 |
There was a problem hiding this comment.
We are not using those methods should we?
There was a problem hiding this comment.
It depends if by GDML we are talking about MD17, MD22 or both. Both of them are calculated from AIMD with FHI-aims with All Electron DFT. MD17 uses the pbe functional with vdw-ts, while MD22 uses pbe+mbd with different settings per system.
I'm going to calculate the isolated atoms energy for ccsd/cc-pvdz, ccsd(t)/cc-pvdz and I'm creating as we speak the pseudopotential for the pbe functional for the pbe+vdw-ts level of theory on DFT PAW. We would have issue on the many body dispersion as I can't use the correct convergence light/tight so I think just using the pbe+vdw-ts should be good enough
| @@ -59,6 +59,10 @@ class GEOM(BaseDataset): | |||
| __name__ = "geom" | |||
| __energy_methods__ = ["gfn2_xtb"] | |||
There was a problem hiding this comment.
@FNTwin Do still need a basis set for semi-empirical methods
There was a problem hiding this comment.
We don't to specify a basis set for semi empirical method as they are fully defined by their name. You don't get to choose the basis set as it is "builded inside" the technique itself.
| @@ -46,15 +46,19 @@ def _read_sdf(sdf_path, properties_path): | |||
|
|
|||
| class Molecule3D(BaseDataset): | |||
| __name__ = "molecule3d" | |||
| __energy_methods__ = ["b3lyp_6-31g*"] | |||
| __energy_methods__ = ["b3lyp/6-31g*"] | |||
| # UNITS MOST LIKELY WRONG, MUST CHECK THEM MANUALLY | |||
There was a problem hiding this comment.
@FNTwin is this already checked?
There was a problem hiding this comment.
No, I need to download the dataset and run some calculation and manually see what unit they are using as they don't provide any info. This PR should be just to have a quick method to interface openMLIP and convert the datasets between units for experimenting. A more heavy PR with every unit checked and validated will be with the calculated isolated atom energy
src/openqdc/datasets/qm7x.py
Outdated
|
|
||
| energy_target_names = ["ePBE0", "eMBD"] | ||
|
|
||
| __force_methods__ = ["pbe-ts", "vdw"] | ||
| __force_methods__ = ["pbe+vdw-ts", "mbd"] |
There was a problem hiding this comment.
@FNTwin Are you sure the second force is mbd or just vdw forces?
There was a problem hiding this comment.
My mistake, corrected. It is a semi empirical tight binding with a many body dispersion correlation and a pbe0 for the dft calculation
| @@ -51,13 +53,13 @@ class WaterClusters(BaseDataset): | |||
|
|
|||
| # Energy in hartree, all zeros by default | |||
| atomic_energies = np.zeros((MAX_ATOMIC_NUMBER,), dtype=np.float32) | |||
|
|
|||
| # need to know where to find the data | |||
There was a problem hiding this comment.
There was a problem hiding this comment.
Ok so you downloaded the entire dataset made from the flexible potential. WIth the isolated atom energy PR I'll have a manual check of the data as with the other unsure dataset
Unit conversion for handling the experiments. Opening the PR while I clean the API and add some tests
It will take care of #9
TODO
Implement a registry of the most common unit conversions (units.py) and modify the constructor for the datasets to keep track of the current units and provide a way to change them.