Dataset basic information

[matsuken92]

There are 4,658,147 rows in train data. There are 2,505,542 rows in test data. There are 85,003 distinct molecules in train data. There are 45,772 distinct molecules in test data. There are 29 unique atoms. There are 8 unique types.

type count

	train	test
1JHC	709416	380609
1JHN	43363	24195
2JHC	1140674	613138
2JHH	378036	203126
2JHN	119253	64424
3JHC	1510379	811999
3JHH	590611	317435
3JHN	166415	90616

atom count

index	atom
H	1208387
C	831726
O	183187
N	132361
F	2996

type label encorder

idx	type
0	1JHC
1	1JHN
2	2JHC
3	2JHH
4	2JHN
5	3JHC
6	3JHH
7	3JHN

[matsuken92]

dipole_moments = pd.read_csv(../input//dipole_moments.csv) nb of cols = 85003

index	molecule_name	X	Y	Z
0	dsgdb9nsd_000001	0.00	0	0.000
1	dsgdb9nsd_000002	-0.00	0	1.626
2	dsgdb9nsd_000003	0.00	0	-1.851
3	dsgdb9nsd_000005	0.00	0	-2.894
4	dsgdb9nsd_000007	0.00	0	0.000

============================================================

structures = pd.read_csv(../input//structures.csv) nb of cols = 2358657

index	molecule_name	atom_index	atom	x	y	z
0	dsgdb9nsd_000001	0	C	-0.012698	1.08580	0.008001
1	dsgdb9nsd_000001	1	H	0.002150	-0.00603	0.001976
2	dsgdb9nsd_000001	2	H	1.011731	1.46375	0.000277
3	dsgdb9nsd_000001	3	H	-0.540815	1.44753	-0.876644
4	dsgdb9nsd_000001	4	H	-0.523814	1.43793	0.906397

============================================================

mulliken_charges = pd.read_csv(../input//mulliken_charges.csv) nb of cols = 1533537

index	molecule_name	atom_index	mulliken_charge
0	dsgdb9nsd_000001	0	-0.5357
1	dsgdb9nsd_000001	1	0.1339
2	dsgdb9nsd_000001	2	0.1339
3	dsgdb9nsd_000001	3	0.1339
4	dsgdb9nsd_000001	4	0.1339

============================================================

magnetic_shielding_tensors = pd.read_csv(../input//magnetic_shielding_tensors.csv) nb of cols = 1533537

index	molecule_name	atom_index	XX	YX	ZX	XY	YY	ZY	XZ	YZ	ZZ
0	dsgdb9nsd_000001	0	195.31	0.000	-0.0001	0.000	195.32	0.0007	-0.0001	0.0007	195.32
1	dsgdb9nsd_000001	1	31.34	-1.232	4.0544	-1.232	28.95	-1.7173	4.0546	-1.7173	34.09
2	dsgdb9nsd_000001	2	31.58	1.217	-4.1474	1.217	28.90	-1.6036	-4.1476	-1.6036	33.90
3	dsgdb9nsd_000001	3	31.52	4.109	1.2723	4.109	33.91	1.6950	1.2724	1.6951	28.96
4	dsgdb9nsd_000001	4	31.40	-4.094	-1.1793	-4.094	34.08	1.6259	-1.1795	1.6260	28.90

============================================================

potential_energy = pd.read_csv(../input//potential_energy.csv) nb of cols = 85003

index	molecule_name	potential_energy
0	dsgdb9nsd_000001	-40.52
1	dsgdb9nsd_000002	-56.56
2	dsgdb9nsd_000003	-76.43
3	dsgdb9nsd_000005	-93.43
4	dsgdb9nsd_000007	-79.84

============================================================

scalar_coupling_contributions = pd.read_csv(../input//scalar_coupling_contributions.csv) nb of cols = 4658147

index	molecule_name	atom_index_0	atom_index_1	type	fc	sd	pso	dso
0	dsgdb9nsd_000001	1	0	1JHC	83.02	0.2546	1.259	0.2720
1	dsgdb9nsd_000001	1	2	2JHH	-11.03	0.3530	2.858	-3.4336
2	dsgdb9nsd_000001	1	3	2JHH	-11.03	0.3529	2.859	-3.4339
3	dsgdb9nsd_000001	1	4	2JHH	-11.03	0.3529	2.859	-3.4339
4	dsgdb9nsd_000001	2	0	1JHC	83.02	0.2546	1.259	0.2720

============================================================

train = pd.read_csv(../input//train.csv) nb of cols = 4658147

index	id	molecule_name	atom_index_0	atom_index_1	type	scalar_coupling_constant
0	0	dsgdb9nsd_000001	1	0	1JHC	84.81
1	1	dsgdb9nsd_000001	1	2	2JHH	-11.26
2	2	dsgdb9nsd_000001	1	3	2JHH	-11.25
3	3	dsgdb9nsd_000001	1	4	2JHH	-11.25
4	4	dsgdb9nsd_000001	2	0	1JHC	84.81

============================================================

[matsuken92]

[matsuken92]

atom_index_0は全て H のはず

train_atom1 = train[["molecule_name", "atom_index_0"]].drop_duplicates()
train_joint = train_atom1.sample(n=100000).merge(structures, 
                      left_on=["molecule_name", "atom_index_0"],
                      right_on=["molecule_name", "atom_index"])
train_joint["atom"].value_counts()

H    100000
Name: atom, dtype: int64