foreblocks/examples/adaptive_mrmr_demo.py at main · lseman/foreblocks · GitHub

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
from __future__ import annotations

import numpy as np
import pandas as pd

from foretools import AdaptiveMI, AdaptiveMRMR
from foretools.fengineer import FeatureConfig
from foretools.fengineer.selectors.feature_selector import FeatureSelector


class ToyAdaptiveMI:
    """
    Tiny deterministic scorer used only to illustrate why MID and MIQ can
    select different features from the same relevance/redundancy structure.
    """

    def __init__(self, feature_names):
        self.feature_names = list(feature_names)
        self._relevance = {
            "lead": 0.95,
            "candidate_mid": 0.94,
            "candidate_miq": 0.75,
        }
        self._redundancy = {
            frozenset(("lead", "candidate_mid")): 0.30,
            frozenset(("lead", "candidate_miq")): 0.12,
            frozenset(("candidate_mid", "candidate_miq")): 0.18,
        }
        self._values_to_name = {}

    def bind_columns(self, X_values):
        self._values_to_name = {
            tuple(np.asarray(X_values[:, idx], dtype=float).tolist()): name
            for idx, name in enumerate(self.feature_names)
        }

    def score_pairwise(self, X_values, y_values, return_raw_mi=False):
        self.bind_columns(X_values)
        return np.array(
            [self._relevance[name] for name in self.feature_names], dtype=float
        )

    def score(self, xa, xb, return_raw_mi=False):
        a = self._values_to_name[tuple(np.asarray(xa, dtype=float).tolist())]
        b = self._values_to_name[tuple(np.asarray(xb, dtype=float).tolist())]
        return float(self._redundancy[frozenset((a, b))])


def _print_title(title: str) -> None:
    print()
    print(title)
    print("-" * len(title))


def demo_criterion_difference() -> None:
    _print_title("Toy scorer: MID and MIQ can diverge")

    X = pd.DataFrame(
        {
            "lead": [0.0, 1.0, 2.0, 3.0],
            "candidate_mid": [10.0, 11.0, 12.0, 13.0],
            "candidate_miq": [20.0, 21.0, 22.0, 23.0],
        }
    )
    y = pd.Series([0.0, 1.0, 0.0, 1.0])

    for criterion in ("mid", "miq"):
        selector = AdaptiveMRMR(
            scorer=ToyAdaptiveMI(X.columns),
            criterion=criterion,
            candidate_pool=3,
            stable_relevance=False,
            random_state=7,
        )
        selector.fit(
            X,
            y,
            min_features=2,
            max_features=2,
            mi_threshold=0.0,
            min_samples=1,
        )
        print(f"{criterion.upper()} selected: {selector.selected_features_}")
        print(selector.selection_scores_)


def demo_real_usage() -> None:
    _print_title("Real AdaptiveMRMR usage on synthetic data")

    rng = np.random.default_rng(42)
    n = 320

    lead = rng.normal(size=n)
    duplicate = lead + 0.03 * rng.normal(size=n)
    complementary = rng.normal(size=n)
    weak = rng.normal(size=n)
    y = pd.Series(1.3 * lead + 0.75 * complementary + 0.05 * rng.normal(size=n))
    X = pd.DataFrame(
        {
            "lead": lead,
            "duplicate": duplicate,
            "complementary": complementary,
            "weak": weak,
        }
    )

    for criterion in ("mid", "miq"):
        selector = AdaptiveMRMR(
            scorer=AdaptiveMI(random_state=42, rho_threshold=0.3),
            criterion=criterion,
            candidate_pool=4,
            stable_relevance=False,
            random_state=42,
        )
        selector.fit(
            X,
            y,
            min_features=2,
            max_features=2,
            mi_threshold=0.0,
        )
        print(f"{criterion.upper()} selected: {selector.selected_features_}")
        print(
            "note: on this synthetic dataset both criteria agree on the kept features"
        )
        print("relevance:")
        print(selector.relevance_scores_)
        print("selection scores:")
        print(selector.selection_scores_)

    cfg = FeatureConfig(
        selector_method="mrmr",
        selector_stable_mi=False,
        min_features=2,
        max_features=2,
        mi_threshold=0.0,
        mrmr_candidate_pool=4,
        mrmr_criterion="miq",
    )
    fs = FeatureSelector(cfg)
    fs.ami_scorer = AdaptiveMI(random_state=42, rho_threshold=0.3)
    fs.fit(X, y)
    print("FeatureSelector with mRMR/MIQ selected:", fs.selected_features_)


if __name__ == "__main__":
    demo_criterion_difference()
    demo_real_usage()