File size: 6,193 Bytes
3ea26d1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
from astartes import train_val_test_split
from astartes.utils.warnings import NormalizationWarning
import numpy as np
import pytest
from rdkit import Chem

from chemprop.data.splitting import _unpack_astartes_result, make_split_indices


@pytest.fixture(params=[["C", "CC", "CCC", "CN", "CCN", "CCCN", "CCCCN", "CO", "CCO", "CCCO"]])
def mol_data(request):
    """A dataset with single molecules"""
    return [Chem.MolFromSmiles(smi) for smi in request.param]


@pytest.fixture(params=[["C", "CC", "CN", "CN", "CO", "C"]])
def mol_data_with_repeated_mols(request):
    """A dataset with repeated single molecules"""
    return [Chem.MolFromSmiles(smi) for smi in request.param]


@pytest.fixture(params=[["C", "CC", "CCC", "C1CC1", "C1CCC1"]])
def molecule_dataset_with_rings(request):
    """A dataset with rings (for scaffold splitting)"""
    return [Chem.MolFromSmiles(smi) for smi in request.param]


def test_splits_sum1_warning(mol_data):
    """Testing that the splits are normalized to 1, for overspecified case."""
    with pytest.warns(NormalizationWarning):
        make_split_indices(mols=mol_data, sizes=(0.4, 0.6, 0.2))


def test_splits_sum2_warning(mol_data):
    """Testing that the splits are normalized to 1, for underspecified case."""
    with pytest.warns(NormalizationWarning):
        make_split_indices(mols=mol_data, sizes=(0.1, 0.1, 0.1))


def test_three_splits_provided(mol_data):
    """Testing that three splits are provided"""
    with pytest.raises(ValueError):
        make_split_indices(mols=mol_data, sizes=(0.8, 0.2))


def test_seed0(mol_data):
    """

    Testing that make_split_indices can get expected output using astartes as backend for random split with seed 0.

    Note: the behaviour of randomness for data splitting is not controlled by chemprop but by the chosen backend.

    """
    train, val, test = make_split_indices(mols=mol_data, seed=0)
    train_astartes, val_astartes, test_astartes = _unpack_astartes_result(
        train_val_test_split(np.arange(len(mol_data)), sampler="random", random_state=0), True
    )
    assert set(train[0]) == set(train_astartes)
    assert set(val[0]) == set(val_astartes)
    assert set(test[0]) == set(test_astartes)


def test_seed100(mol_data):
    """

    Testing that make_split_indices can get expected output using astartes as backend for random split with seed 100.

    Note: the behaviour of randomness for data splitting is not controlled by chemprop but by the chosen backend.

    """
    train, val, test = make_split_indices(mols=mol_data, seed=100)
    train_astartes, val_astartes, test_astartes = _unpack_astartes_result(
        train_val_test_split(np.arange(len(mol_data)), sampler="random", random_state=100), True
    )
    assert set(train[0]) == set(train_astartes)
    assert set(val[0]) == set(val_astartes)
    assert set(test[0]) == set(test_astartes)


def test_split_4_4_2(mol_data):
    """Testing the random split with changed sizes"""
    train, val, test = make_split_indices(mols=mol_data, sizes=(0.4, 0.4, 0.2))
    train_astartes, val_astartes, test_astartes = _unpack_astartes_result(
        train_val_test_split(
            np.arange(len(mol_data)),
            sampler="random",
            train_size=0.4,
            val_size=0.4,
            test_size=0.2,
            random_state=0,
        ),
        True,
    )
    assert set(train[0]) == set(train_astartes)
    assert set(val[0]) == set(val_astartes)
    assert set(test[0]) == set(test_astartes)


def test_split_empty_validation_set(mol_data):
    """Testing the random split with an empty validation set"""
    train, val, test = make_split_indices(mols=mol_data, sizes=(0.4, 0, 0.6))
    assert set(val[0]) == set([])


def test_random_split(mol_data_with_repeated_mols):
    """

    Testing if random split yield expected results.

    Note: This test mainly serves as a red flag. Test failure strongly indicates unexpected change of data splitting backend that needs attention.

    """
    split_type = "random"
    train, val, test = make_split_indices(
        mols=mol_data_with_repeated_mols, sizes=(0.4, 0.4, 0.2), split=split_type
    )

    assert train[0] == [2, 1]


def test_repeated_smiles(mol_data_with_repeated_mols):
    """

    Testing if random split with repeated smiles yield expected results.

    Note: This test mainly serves as a red flag. Test failure strongly indicates unexpected change of data splitting backend that needs attention.

    """
    split_type = "random_with_repeated_smiles"
    train, val, test = make_split_indices(
        mols=mol_data_with_repeated_mols, sizes=(0.8, 0.0, 0.2), split=split_type
    )

    assert train[0] == [4, 1, 0, 5]
    assert test[0] == [2, 3]


def test_kennard_stone(mol_data):
    """

    Testing if Kennard-Stone split yield expected results.

    Note: This test mainly serves as a red flag. Test failure strongly indicates unexpected change of data splitting backend that needs attention.

    """
    split_type = "kennard_stone"
    train, val, test = make_split_indices(mols=mol_data, sizes=(0.4, 0.4, 0.2), split=split_type)

    assert set(test[0]) == set([9, 5])


def test_kmeans(mol_data):
    """

    Testing if Kmeans split yield expected results.

    Note: This test mainly serves as a red flag. Test failure strongly indicates unexpected change of data splitting backend that needs attention.

    """
    split_type = "kmeans"
    train, val, test = make_split_indices(mols=mol_data, sizes=(0.5, 0.0, 0.5), split=split_type)

    assert train[0] == [0, 1, 2, 3, 7, 8, 9]


def test_scaffold(molecule_dataset_with_rings):
    """

    Testing if Bemis-Murcko Scaffolds split yield expected results.

    Note: This test mainly serves as a red flag. Test failure strongly indicates unexpected change of data splitting backend that needs attention.

    """
    split_type = "scaffold_balanced"
    train, val, test = make_split_indices(
        mols=molecule_dataset_with_rings, sizes=(0.3, 0.3, 0.3), split=split_type
    )

    assert train[0] == [0, 1, 2]