The #1 adoption lever from the competitive analysis: a sklearn-compatible package (python/lawsynth-sklearn) so the scikit-learn ecosystem adopts LawSynth with near-zero switching cost (why gplearn/PySINDy win mindshare). - LawSynthDynamics: fit(X, t=)/predict/simulate/score/equations over the real Study discovery; full get_params/set_params/__sklearn_tags__/NotFittedError. - LawSynthRegressor (RegressorMixin), LawSynthTransformer (feature library + deterministic decorrelation prune), parsimony='auto' (lambda = Cov/Var). - Inherits sklearn mixins when present; faithful standalone fallback otherwise. Demo recovers an oscillator R2=0.999996; 50 tests pass (7 sklearn-integration skip cleanly when sklearn absent). Deterministic, offline. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
83 lines
2.7 KiB
Python
83 lines
2.7 KiB
Python
"""LawSynthTransformer: feature-library engineering and correlation prune."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import math
|
|
|
|
import pytest
|
|
|
|
from lawsynth_sklearn import LawSynthTransformer
|
|
|
|
|
|
@pytest.fixture
|
|
def simple():
|
|
return [[1.0, 2.0], [2.0, 1.0], [3.0, 0.5], [0.5, 3.0], [4.0, 2.0]]
|
|
|
|
|
|
def _rows(Z):
|
|
return Z.tolist() if hasattr(Z, "tolist") else Z
|
|
|
|
|
|
def test_polynomial_names_and_count(simple):
|
|
tr = LawSynthTransformer(degree=2, include_bias=True)
|
|
Z = tr.fit_transform(simple)
|
|
names = list(tr.get_feature_names_out())
|
|
# bias + 2 linear + 3 quadratic = 6
|
|
assert names == ["1", "x0", "x1", "x0^2", "x0 x1", "x1^2"]
|
|
assert len(_rows(Z)[0]) == 6
|
|
|
|
|
|
def test_no_bias(simple):
|
|
tr = LawSynthTransformer(degree=2, include_bias=False).fit(simple)
|
|
assert "1" not in tr.feature_names_out_
|
|
|
|
|
|
def test_trigonometric_augmentation(simple):
|
|
tr = LawSynthTransformer(degree=1, include_bias=False, include_trigonometric=True)
|
|
Z = _rows(tr.fit_transform(simple))
|
|
names = list(tr.get_feature_names_out())
|
|
assert "sin(x0)" in names and "cos(x1)" in names
|
|
j = names.index("sin(x0)")
|
|
assert abs(Z[0][j] - math.sin(1.0)) < 1e-12
|
|
|
|
|
|
def test_values_are_correct(simple):
|
|
tr = LawSynthTransformer(degree=2, include_bias=True)
|
|
Z = _rows(tr.fit_transform(simple))
|
|
# row [1,2] -> [1, 1, 2, 1, 2, 4]
|
|
assert Z[0] == [1.0, 1.0, 2.0, 1.0, 2.0, 4.0]
|
|
|
|
|
|
def test_transform_matches_fit_transform(simple):
|
|
tr = LawSynthTransformer(degree=2, include_bias=True)
|
|
a = _rows(tr.fit_transform(simple))
|
|
b = _rows(tr.transform(simple))
|
|
assert a == b
|
|
|
|
|
|
def test_correlation_prune_reduces_columns(simple):
|
|
full = LawSynthTransformer(degree=2, include_bias=True).fit(simple)
|
|
pruned = LawSynthTransformer(
|
|
degree=2, include_bias=True, prune_correlation=True, correlation_threshold=0.9
|
|
).fit(simple)
|
|
assert len(pruned.feature_names_out_) <= len(full.feature_names_out_)
|
|
# kept indices are a subset of the full library, in ascending order.
|
|
assert list(pruned.kept_indices_) == sorted(pruned.kept_indices_)
|
|
|
|
|
|
def test_prune_is_deterministic(simple):
|
|
a = LawSynthTransformer(degree=2, prune_correlation=True, correlation_threshold=0.95).fit(simple)
|
|
b = LawSynthTransformer(degree=2, prune_correlation=True, correlation_threshold=0.95).fit(simple)
|
|
assert a.kept_indices_ == b.kept_indices_
|
|
|
|
|
|
def test_invalid_correlation_threshold(simple):
|
|
with pytest.raises(ValueError):
|
|
LawSynthTransformer(prune_correlation=True, correlation_threshold=1.5).fit(simple)
|
|
|
|
|
|
def test_transform_feature_count_guard(simple):
|
|
tr = LawSynthTransformer(degree=2).fit(simple)
|
|
with pytest.raises(ValueError):
|
|
tr.transform([[1.0, 2.0, 3.0]])
|