|
1
|
|
|
""" |
|
2
|
|
|
Calculations of concordance between annotations. |
|
3
|
|
|
""" |
|
4
|
|
|
import abc |
|
5
|
|
|
import enum |
|
6
|
|
|
import math |
|
7
|
|
|
from typing import Collection, Dict, Generator, Sequence, Set, Tuple, Union, Type |
|
|
|
|
|
|
8
|
|
|
|
|
9
|
|
|
import numpy as np |
|
|
|
|
|
|
10
|
|
|
import pandas as pd |
|
|
|
|
|
|
11
|
|
|
from typeddfs import TypedDfs |
|
|
|
|
|
|
12
|
|
|
|
|
13
|
|
|
from mandos.analysis import AnalysisUtils |
|
|
|
|
|
|
14
|
|
|
from mandos.analysis import SimilarityDfLongForm, SimilarityDfShortForm |
|
15
|
|
|
from mandos.model import CleverEnum |
|
16
|
|
|
|
|
17
|
|
|
ConcordanceDf = ( |
|
18
|
|
|
TypedDfs.typed("ConcordanceDf") |
|
19
|
|
|
.require("phi", "psi", dtype=str) |
|
20
|
|
|
.require("sample", dtype=int) |
|
21
|
|
|
.require("tau", dtype=float) |
|
22
|
|
|
).build() |
|
23
|
|
|
|
|
24
|
|
|
|
|
25
|
|
|
class ConcordanceCalculator(metaclass=abc.ABCMeta): |
|
|
|
|
|
|
26
|
|
|
def __init__(self, n_samples: int, seed: int): |
|
27
|
|
|
self.n_samples = n_samples |
|
28
|
|
|
self.seed = seed |
|
29
|
|
|
self.rand = np.random.RandomState(seed) |
|
30
|
|
|
|
|
31
|
|
|
def calc_all(self, phis: SimilarityDfLongForm, psis: SimilarityDfLongForm) -> ConcordanceDf: |
|
|
|
|
|
|
32
|
|
|
for phi in phis["phi"].unique(): |
|
33
|
|
|
for psi in psis["psi"].unique(): |
|
34
|
|
|
self.calc(None, None, phi, psi) # TODO |
|
|
|
|
|
|
35
|
|
|
|
|
36
|
|
|
def calc( |
|
|
|
|
|
|
37
|
|
|
self, phi: SimilarityDfShortForm, psi: SimilarityDfShortForm, phi_name: str, psi_name: str |
|
|
|
|
|
|
38
|
|
|
) -> ConcordanceDf: |
|
39
|
|
|
phi_cols, psi_cols = phi.columns.tolist(), psi.columns.tolist() |
|
40
|
|
|
if phi_cols != psi_cols: |
|
41
|
|
|
raise ValueError(f"Mismatched compounds: {phi_cols} != {psi_cols}") |
|
42
|
|
|
df = pd.DataFrame(data=self.generate(phi, psi), columns=["score"]) |
|
|
|
|
|
|
43
|
|
|
df = df.reset_index() |
|
|
|
|
|
|
44
|
|
|
df["phi"] = phi_name |
|
45
|
|
|
df["psi"] = psi_name |
|
46
|
|
|
df.columns = ["sample", "tau", "phi", "psi"] |
|
47
|
|
|
return ConcordanceDf.convert(df) |
|
48
|
|
|
|
|
49
|
|
|
def generate( |
|
|
|
|
|
|
50
|
|
|
self, phi: SimilarityDfShortForm, psi: SimilarityDfShortForm |
|
|
|
|
|
|
51
|
|
|
) -> Generator[float, None, None]: |
|
52
|
|
|
if self.n_samples == 1: |
|
53
|
|
|
yield self._calc(phi, psi) |
|
54
|
|
|
else: |
|
55
|
|
|
for b in range(self.n_samples): |
|
|
|
|
|
|
56
|
|
|
phi_b = self.rand.choice(phi, replace=True) |
|
57
|
|
|
psi_b = self.rand.choice(psi, replace=True) |
|
58
|
|
|
yield self._calc(phi_b, psi_b) |
|
59
|
|
|
|
|
60
|
|
|
def _calc(self, phi: SimilarityDfShortForm, psi: SimilarityDfShortForm) -> float: |
|
|
|
|
|
|
61
|
|
|
raise NotImplemented() |
|
|
|
|
|
|
62
|
|
|
|
|
63
|
|
|
|
|
64
|
|
|
class TauConcordanceCalculator(ConcordanceCalculator): |
|
|
|
|
|
|
65
|
|
|
def _calc(self, phi: SimilarityDfShortForm, psi: SimilarityDfShortForm) -> float: |
|
66
|
|
|
n = len(phi) |
|
|
|
|
|
|
67
|
|
|
numerator = self._n_z(phi, psi, 1) - self._n_z(phi, psi, -1) |
|
68
|
|
|
denominator = math.factorial(n) / (2 * math.factorial(n - 2)) |
|
69
|
|
|
return numerator / denominator |
|
70
|
|
|
|
|
71
|
|
|
def _n_z(self, a: Sequence[float], b: Sequence[float], z: int) -> int: |
|
|
|
|
|
|
72
|
|
|
values = [self._i_sum(a, b, i, z) for i in range(len(a))] |
|
73
|
|
|
return int(np.sum(values)) |
|
74
|
|
|
|
|
75
|
|
|
def _i_sum(self, a: np.array, b: np.array, i: int, z: int): |
|
|
|
|
|
|
76
|
|
|
values = [int(np.sign(a[i] - a[j]) == z * np.sign(b[i] - b[j]) != 0) for j in range(i)] |
|
77
|
|
|
return int(np.sum(values)) |
|
78
|
|
|
|
|
79
|
|
|
|
|
80
|
|
|
class ConcordanceAlg(CleverEnum): |
|
|
|
|
|
|
81
|
|
|
tau = enum.auto() |
|
82
|
|
|
|
|
83
|
|
|
@property |
|
84
|
|
|
def clazz(self) -> Type[ConcordanceCalculator]: |
|
|
|
|
|
|
85
|
|
|
return {ConcordanceAlg.tau: TauConcordanceCalculator}[self] |
|
86
|
|
|
|
|
87
|
|
|
|
|
88
|
|
|
class ConcordanceCalculation: |
|
|
|
|
|
|
89
|
|
|
@classmethod |
|
90
|
|
|
def create( |
|
|
|
|
|
|
91
|
|
|
cls, |
|
|
|
|
|
|
92
|
|
|
algorithm: Union[str, ConcordanceAlg], |
|
|
|
|
|
|
93
|
|
|
phi_name: str, |
|
|
|
|
|
|
94
|
|
|
psi_name: str, |
|
|
|
|
|
|
95
|
|
|
n_samples: int, |
|
|
|
|
|
|
96
|
|
|
seed: int, |
|
|
|
|
|
|
97
|
|
|
) -> ConcordanceCalculator: |
|
98
|
|
|
algorithm = ConcordanceAlg.of(algorithm).clazz |
|
99
|
|
|
return algorithm(n_samples=n_samples, seed=seed, phi_name=phi_name, psi_name=psi_name) |
|
100
|
|
|
|
|
101
|
|
|
|
|
102
|
|
|
__all__ = [ |
|
103
|
|
|
"ConcordanceCalculator", |
|
104
|
|
|
"TauConcordanceCalculator", |
|
105
|
|
|
"ConcordanceDf", |
|
106
|
|
|
"ConcordanceCalculation", |
|
107
|
|
|
"ConcordanceAlg", |
|
108
|
|
|
] |
|
109
|
|
|
|
dmyersturnbull /
mandos
Loading history...