Datasets:

PureOne
/

luminous-hypercontinuum-foundation-v3

import math
import numpy as np
import pytest
from lhf.core import *
from lhf.recovery import *
from lhf.memory import prune_dominated

@pytest.mark.parametrize('budget',[0.,.01,.1,.3,math.log(2),1.,2.,5.])
def test_flagship_frontier(budget):
    from lhf.core import flagship_contraction
    family,_,_=flagship_family()
    assert robust_floor(family,flagship_contraction(budget))==pytest.approx(flagship_frontier(budget),abs=1e-12)

@pytest.mark.parametrize('d',[1,2,3,4,8,12])
@pytest.mark.parametrize('budget',[0.,.1,1.,6.])
def test_determinant_and_contraction(d,budget):
    y=np.diag(np.geomspace(.25,8,d))
    c=contraction_from_lower_bound(y,budget)
    assert np.linalg.slogdet(c)[1]==pytest.approx(-budget,abs=1e-9)
    assert np.linalg.norm(c,2)<=1+1e-10
    assert np.linalg.eigvalsh(c.T@y@c)[0]>=frontier_from_lower_bound(y,budget)-1e-10

@pytest.mark.parametrize('a,b',[(np.eye(2),2*np.eye(2)),(np.diag([4.,1]),np.diag([1.,4.])),
                               (np.array([[2.,2.],[2.,8.]]),np.array([[2.,-2.],[-2.,8.]]))])
def test_pair_duality(a,b):
    y,z=envelope_pair(a,b)
    assert logdet_dual_bound([a,b],z)==pytest.approx(np.linalg.slogdet(y)[1],abs=1e-10)

@pytest.mark.parametrize('invalid',[[],[np.zeros((2,2))],[np.array([[1.,2.],[0.,1.]])],
                                    [np.eye(2),np.eye(3)],[np.array([[np.nan]])]])
def test_validation_rejects_bad_family(invalid):
    with pytest.raises(ValueError):validate_family(invalid)

@pytest.mark.parametrize('budget',[-1.,float('inf'),float('nan')])
def test_budget_rejection(budget):
    with pytest.raises(ValueError):contraction_from_lower_bound(np.eye(2),budget)

@pytest.mark.parametrize('d',[1,2,10,30])
@pytest.mark.parametrize('h',[0.,.01,.4,1.,2.])
def test_sobolev_radius(d,h):
    r=sobolev_radius(d,1.,1.,h)
    expected=1. if h>=1 else np.sqrt(h*h+(1-h*h)/(d+1)**2)
    assert r==pytest.approx(expected)

def test_nonassociativity():
    a,b=np.diag([4.,1.]),np.diag([1.,4.]); n=np.array([[4.,3.],[3.,4.]])/5
    ab,_=envelope_pair(a,b); left,_=envelope_pair(ab,n)
    bn,_=envelope_pair(b,n);right,_=envelope_pair(a,bn)
    assert not np.allclose(left,right)
    assert np.allclose(right,n)

def test_pruning_witnesses():
    ms=[3*np.eye(2),np.diag([1.,2.]),np.eye(2),2*np.eye(2)]
    keep,witnesses=prune_dominated(ms)
    assert keep==[2]
    for j,i in witnesses.items():
        assert np.linalg.eigvalsh(ms[j]-ms[i])[0]>=-1e-10

def test_bad_dilation():
    with pytest.raises(ValueError):unitary_dilation(2*np.eye(2))

def test_bad_requested_floor():
    with pytest.raises(ValueError):contraction_from_lower_bound(np.eye(2),1.,floor=1.)

    with pytest.raises(ValueError):contraction_from_lower_bound(np.eye(2),0.,floor=2.)
    with pytest.raises(ValueError):contraction_from_lower_bound(np.eye(2),0.,floor=float("nan"))

	import math
	import numpy as np
	import pytest
	from lhf.core import *
	from lhf.recovery import *
	from lhf.memory import prune_dominated

	@pytest.mark.parametrize('budget',[0.,.01,.1,.3,math.log(2),1.,2.,5.])
	def test_flagship_frontier(budget):
	from lhf.core import flagship_contraction
	family,_,_=flagship_family()
	assert robust_floor(family,flagship_contraction(budget))==pytest.approx(flagship_frontier(budget),abs=1e-12)

	@pytest.mark.parametrize('d',[1,2,3,4,8,12])
	@pytest.mark.parametrize('budget',[0.,.1,1.,6.])
	def test_determinant_and_contraction(d,budget):
	y=np.diag(np.geomspace(.25,8,d))
	c=contraction_from_lower_bound(y,budget)
	assert np.linalg.slogdet(c)[1]==pytest.approx(-budget,abs=1e-9)
	assert np.linalg.norm(c,2)<=1+1e-10
	assert np.linalg.eigvalsh(c.T@y@c)[0]>=frontier_from_lower_bound(y,budget)-1e-10

	@pytest.mark.parametrize('a,b',[(np.eye(2),2*np.eye(2)),(np.diag([4.,1]),np.diag([1.,4.])),
	(np.array([[2.,2.],[2.,8.]]),np.array([[2.,-2.],[-2.,8.]]))])
	def test_pair_duality(a,b):
	y,z=envelope_pair(a,b)
	assert logdet_dual_bound([a,b],z)==pytest.approx(np.linalg.slogdet(y)[1],abs=1e-10)

	@pytest.mark.parametrize('invalid',[[],[np.zeros((2,2))],[np.array([[1.,2.],[0.,1.]])],
	[np.eye(2),np.eye(3)],[np.array([[np.nan]])]])
	def test_validation_rejects_bad_family(invalid):
	with pytest.raises(ValueError):validate_family(invalid)

	@pytest.mark.parametrize('budget',[-1.,float('inf'),float('nan')])
	def test_budget_rejection(budget):
	with pytest.raises(ValueError):contraction_from_lower_bound(np.eye(2),budget)

	@pytest.mark.parametrize('d',[1,2,10,30])
	@pytest.mark.parametrize('h',[0.,.01,.4,1.,2.])
	def test_sobolev_radius(d,h):
	r=sobolev_radius(d,1.,1.,h)
	expected=1. if h>=1 else np.sqrt(hh+(1-hh)/(d+1)**2)
	assert r==pytest.approx(expected)

	def test_nonassociativity():
	a,b=np.diag([4.,1.]),np.diag([1.,4.]); n=np.array([[4.,3.],[3.,4.]])/5
	ab,_=envelope_pair(a,b); left,_=envelope_pair(ab,n)
	bn,_=envelope_pair(b,n);right,_=envelope_pair(a,bn)
	assert not np.allclose(left,right)
	assert np.allclose(right,n)

	def test_pruning_witnesses():
	ms=[3np.eye(2),np.diag([1.,2.]),np.eye(2),2np.eye(2)]
	keep,witnesses=prune_dominated(ms)
	assert keep==[2]
	for j,i in witnesses.items():
	assert np.linalg.eigvalsh(ms[j]-ms[i])[0]>=-1e-10

	def test_bad_dilation():
	with pytest.raises(ValueError):unitary_dilation(2*np.eye(2))

	def test_bad_requested_floor():
	with pytest.raises(ValueError):contraction_from_lower_bound(np.eye(2),1.,floor=1.)

	with pytest.raises(ValueError):contraction_from_lower_bound(np.eye(2),0.,floor=2.)
	with pytest.raises(ValueError):contraction_from_lower_bound(np.eye(2),0.,floor=float("nan"))