dump.py

import torch
import mysql.connector
import numpy as np
import tqdm
from tqdm.contrib.concurrent import process_map
from ase.data import atomic_numbers

from materials_toolkit.data import HDF5Dataset, StructureData, batching, Batching

import os


@batching(
    entry_id=Batching(dtype=torch.long),
    structure_id=Batching(dtype=torch.long),
    calculation_id=Batching(dtype=torch.long),
    spacegroup=Batching(dtype=torch.long),
)
class OQMDData(StructureData):
    pass


class OQMD(HDF5Dataset):
    data_class = OQMDData


def get_all_structures(host, user, password, database="qmdb"):

    with mysql.connector.connect(
        host=host, user=user, password=password, database=database
    ) as db:
        structures = []
        with db.cursor() as c:
            # calculate atoms count for all structures
            c.execute(
                'select TABLE_NAME from information_schema.tables where table_name="structures_natoms";'
            )
            if len(c.fetchall()) == 0:
                c.execute(
                    "create table structures_natoms select structure_id,count(*) as natoms from atoms group by structure_id;"
                )

            # filter valid structures (natoms match between structures table and structures_natoms)
            c.execute(
                'select TABLE_NAME from information_schema.tables where table_name="valid_structure";'
            )
            if len(c.fetchall()) == 0:
                c.execute(
                    """
                    create table valid_structure 
                    select structures.id as structure_id,structures.natoms as natoms from structures
                    inner join structures_natoms
                    on
                        structures.id=structures_natoms.structure_id
                        and structures.natoms=structures_natoms.natoms;
                    """
                )

            # get minimal energy of an entry, select lower energy from converged calculations and reject lda setup
            c.execute(
                'select TABLE_NAME from information_schema.tables where table_name="entries_minimal_energy";'
            )
            if len(c.fetchall()) == 0:  # TODO filtrer standard et static
                c.execute(
                    f"""
                    create table entries_minimal_energy
                    select calculations.entry_id as entry_id, min(calculations.energy_pa) as energy_pa
                    from calculations
                    inner join valid_structure
                    on
                        calculations.converged is not null
                        and calculations.converged=true
                        and calculations.label NOT LIKE '%lda'
                        and (calculations.label LIKE 'static%'
                        or calculations.label LIKE 'standard%')
                        and valid_structure.structure_id=calculations.output_id
                    group by calculations.entry_id;
                    """
                )

            c.execute(
                'select TABLE_NAME from information_schema.tables where table_name="entries_with_structure";'
            )
            if len(c.fetchall()) == 0:
                c.execute(
                    """
                    create table entries_with_structure
                    select calculations.entry_id as entry_id, calculations.output_id as structure_id, calculations.id as calculation_id, calculations.label as calculation_label, calculations.energy_pa as energy_pa
                    from entries_minimal_energy
                    left join calculations
                    on
                        calculations.entry_id=entries_minimal_energy.entry_id
                        and calculations.energy_pa=entries_minimal_energy.energy_pa
                    inner join valid_structure
                    on
                        valid_structure.structure_id=calculations.output_id
                    group by calculations.entry_id;
                    """
                )

            c.execute("select * from entries_with_structure;")
            entries_calc = c.fetchall()

            for entry_id, struct_id, calc_id, calc_label, energy_pa in tqdm.tqdm(
                entries_calc
            ):
                c.execute(
                    f"""
                    select struct.natoms, spacegroups.hm, struct.x1, struct.x2, struct.x3, struct.y1, struct.y2, struct.y3, struct.z1, struct.z2, struct.z3, struct.volume
                    from (
                            select natoms, x1, x2, x3, y1, y2, y3, z1, z2, z3, volume, coalesce(spacegroup_id, 1) as spacegroup_id
                            from structures 
                            where structures.id={struct_id}
                        ) as struct
                    inner join spacegroups on
                        struct.spacegroup_id=spacegroups.number;
                    """
                )
                results = c.fetchall()
                assert len(results) == 1, f"error structure {struct_id}"
                (
                    natoms,
                    space_group,
                    x1,
                    x2,
                    x3,
                    y1,
                    y2,
                    y3,
                    z1,
                    z2,
                    z3,
                    volume,
                ) = results[0]

                c.execute(
                    f"""
                    select element_id,x,y,z
                    from atoms
                    where structure_id={struct_id}
                    """
                )
                atoms = c.fetchall()

                if len(atoms) != natoms:
                    print(entry_id, struct_id, calc_label, energy_pa)
                    for element_id, x, y, z in atoms:
                        print(f"{element_id:>4s} {x:>.3f} {y:>.3f} {z:>.3f}")

                assert len(atoms) == natoms, f"error {len(atoms)} {natoms}"

                z_lst = []
                coords = []
                for e_id, x, y, z in atoms:
                    z_lst.append(atomic_numbers[e_id])
                    coords.append([x, y, z])
                coords = np.array(coords)
                cell = np.array([[x1, x2, x3], [y1, y2, y3], [z1, z2, z3]])

                structures.append(
                    OQMDData(
                        z=torch.tensor(z_lst, dtype=torch.long).view(-1),
                        pos=torch.from_numpy(coords).float().view(-1, 3),
                        cell=torch.from_numpy(cell).float().view(1, 3, 3),
                        entry_id=torch.tensor([entry_id]).long().view(1),
                        structure_id=torch.tensor([struct_id]).long().view(1),
                        calculation_id=torch.tensor([calc_id]).long().view(1),
                        spacegroup=torch.tensor([space_group]).long().view(1),
                        energy_pa=torch.tensor([energy_pa]).float().view(1),
                    )
                )

    return structures


def main():
    import argparse

    parser = argparse.ArgumentParser()

    parser.add_argument("--host", default="localhost", help="mysql host serveur")
    parser.add_argument("--database", default="qmdb", help="name of the database")
    parser.add_argument("--user", help="user of the mysql server")
    parser.add_argument("--password", help="password of the user")

    args = parser.parse_args()

    structures = get_all_structures(
        host=args.host, user=args.user, password=args.password, database="qmdb"
    )

    oqmd_dir = os.path.join(".", "open-quantum-materials-database")
    processed_dir = os.path.join(oqmd_dir, "processed")

    HDF5Dataset.create_dataset(processed_dir, structures)

    dataset = OQMD(oqmd_dir)
    dataset.compute_convex_hulls()


if __name__ == "__main__":
    main()  # check entry 351317