def test_orthogonal():
# Test shape, orthogonality, determinant, python2 error, MathArray
if six.PY2:
with raises(NotImplementedError, match='This feature requires newer versions of '
'numpy and scipy than are available.'):
matrices = OrthogonalMatrices(unitdet=False)
matrices.gen_sample()
with raises(NotImplementedError, match='This feature requires newer versions of '
'numpy and scipy than are available.'):
matrices = OrthogonalMatrices(unitdet=True)
matrices.gen_sample()
else:
# These are the doctests
matrices = OrthogonalMatrices()
assert matrices.gen_sample().shape == (2, 2)
matrices = OrthogonalMatrices(dimension=4)
assert matrices.gen_sample().shape == (4, 4)
matrices = OrthogonalMatrices(unitdet=True)
assert within_tolerance(np.linalg.det(matrices.gen_sample()), 1, 1e-14)
matrices = OrthogonalMatrices(unitdet=False)
m = matrices.gen_sample()
assert (within_tolerance(np.linalg.det(m), 1, 1e-14)
or within_tolerance(np.linalg.det(m), -1, 1e-14))
matrices = OrthogonalMatrices(unitdet=True)
m = matrices.gen_sample()
assert within_tolerance(m * np.conjugate(np.transpose(m)), MathArray(np.eye(2)), 1e-14)
matrices = OrthogonalMatrices(unitdet=False)
m = matrices.gen_sample()
assert within_tolerance(m * np.conjugate(np.transpose(m)), MathArray(np.eye(2)), 1e-14)
shapes = tuple(range(2, 5))
dets = (True, False)
# More general testing
for shape in shapes:
for det in dets:
matrices = matrices = OrthogonalMatrices(dimension=shape, unitdet=det)
m = matrices.gen_sample()
assert m.shape == (shape, shape)
assert np.array_equal(np.conj(m), m)
assert within_tolerance(m * np.transpose(m), MathArray(np.eye(shape)), 1e-14)
assert (approx(np.abs(np.linalg.det(m)), 1)
or approx(np.abs(np.linalg.det(m)), -1))
if det:
assert approx(np.linalg.det(m), 1)
assert isinstance(m, MathArray)
def test_general_matrices():
# Test shape, real/complex, norm, triangular options, MathArray
shapes = product(tuple(range(2, 5)), tuple(range(2, 5)))
norms = ([2, 6], [3, 4], [4, 12], [1 - 1e-12, 1 + 1e-12])
triangles = (None, 'upper', 'lower')
for shape in shapes:
for norm in norms:
for triangle in triangles:
matrices = RealMatrices(shape=shape, norm=norm, triangular=triangle)
m = matrices.gen_sample()
assert m.shape == shape
assert norm[0] <= np.linalg.norm(m) <= norm[1]
assert np.array_equal(np.conj(m), m)
assert isinstance(m, MathArray)
if triangle is None:
assert not within_tolerance(m, MathArray(np.triu(m)), 0)
assert not within_tolerance(m, MathArray(np.tril(m)), 0)
elif triangle == "upper":
assert within_tolerance(m, MathArray(np.triu(m)), 0)
elif triangle == "lower":
assert within_tolerance(m, MathArray(np.tril(m)), 0)
matrices = ComplexMatrices(shape=shape, norm=norm, triangular=triangle)
m = matrices.gen_sample()
assert m.shape == shape
assert norm[0] <= np.linalg.norm(m) <= norm[1]
assert not np.array_equal(np.conj(m), m)
assert isinstance(m, MathArray)
if triangle is None:
assert not within_tolerance(m, MathArray(np.triu(m)), 0)
assert not within_tolerance(m, MathArray(np.tril(m)), 0)
elif triangle == "upper":
assert within_tolerance(m, MathArray(np.triu(m)), 0)
elif triangle == "lower":
assert within_tolerance(m, MathArray(np.tril(m)), 0)
def _within_tolerance(x, y):
return within_tolerance(x, y, self.config['tolerance'])
def raw_check(self, answer, cleaned_input):
"""Perform the numerical check of student_input vs answer"""
var_samples = gen_symbols_samples(self.config['variables'],
self.config['samples'],
self.config['sample_from'],
self.functions, {})
func_samples = gen_symbols_samples(self.random_funcs.keys(),
self.config['samples'],
self.random_funcs, self.functions,
{})
# Make a copy of the functions and variables lists
# We'll add the sampled functions/variables in
funcscope = self.functions.copy()
varscope = self.constants.copy()
num_failures = 0
for i in range(self.config['samples']):
# Update the functions and variables listings with this sample
funcscope.update(func_samples[i])
varscope.update(var_samples[i])
# Evaluate integrals. Error handling here is in two parts because
# 1. custom error messages we've added
# 2. scipy's warnings re-raised as error messages
try:
expected_re, expected_im, _ = self.evaluate_int(
answer['integrand'],
answer['lower'],
answer['upper'],
answer['integration_variable'],
varscope=varscope,
funcscope=funcscope)
except IntegrationError as e:
msg = "Integration Error with author's stored answer: {}"
raise ConfigError(msg.format(e.message))
student_re, student_im, used_funcs = self.evaluate_int(
cleaned_input['integrand'],
cleaned_input['lower'],
cleaned_input['upper'],
cleaned_input['integration_variable'],
varscope=varscope,
funcscope=funcscope)
# scipy raises integration warnings when things go wrong,
# except they aren't really warnings, they're just printed to stdout
# so we use quad's full_output option to catch the messages, and then raise errors.
# The 4th component only exists when its warning message is non-empty
if len(student_re) == 4:
raise IntegrationError(student_re[3])
if len(expected_re) == 4:
raise ConfigError(expected_re[3])
if len(student_im) == 4:
raise IntegrationError(student_im[3])
if len(expected_im) == 4:
raise ConfigError(expected_im[3])
self.log(
self.debug_appendix_template.format(
samplenum=i,
variables=varscope,
student_re_eval=student_re[0],
student_re_error=student_re[1],
student_re_neval=student_re[2]['neval'],
student_im_eval=student_im[0],
student_im_error=student_im[1],
student_im_neval=student_im[2]['neval'],
author_re_eval=expected_re[0],
author_re_error=expected_re[1],
author_re_neval=expected_re[2]['neval'],
author_im_eval=expected_im[0],
author_im_error=expected_im[1],
author_im_neval=expected_im[2]['neval'],
))
# Check if expressions agree
expected = expected_re[0] + (expected_im[0] or 0) * 1j
student = student_re[0] + (student_im[0] or 0) * 1j
if not within_tolerance(expected, student,
self.config['tolerance']):
num_failures += 1
if num_failures > self.config["failable_evals"]:
return {
'ok': False,
'grade_decimal': 0,
'msg': ''
}, used_funcs
# This response appears to agree with the expected answer
return {'ok': True, 'grade_decimal': 1, 'msg': ''}, used_funcs
def test_square_matrices():
# Test shape, real/complex, norm, symmetry, traceless, det, MathArray
shapes = tuple(range(2, 5))
norms = ([2, 6], [1 - 1e-12, 1 + 1e-12])
symmetries = (None, 'diagonal', 'symmetric', 'antisymmetric', 'hermitian', 'antihermitian')
traceless_opts = (True, False)
dets = (None, 0, 1)
complexes = (True, False)
for det, traceless, symmetry in product(dets, traceless_opts, symmetries):
# Handle the cases that don't work
if det == 0 and traceless:
with raises(ConfigError, match='Unable to generate zero determinant traceless matrices'):
SquareMatrices(traceless=traceless,
symmetry=symmetry,
determinant=det)
continue
# Continue with further cases
for shape, norm, comp in product(shapes, norms, complexes):
# Check for matrices that don't exist
args = {'dimension': shape, 'norm': norm, 'traceless': traceless,
'symmetry': symmetry, 'determinant': det, 'complex': comp}
if det == 1:
if traceless and shape == 2:
if symmetry == 'diagonal' and not comp:
with raises(ConfigError, match='No real, traceless, unit-determinant, diagonal 2x2 matrix exists'):
SquareMatrices(**args)
continue
elif symmetry == 'symmetric' and not comp:
with raises(ConfigError, match='No real, traceless, unit-determinant, symmetric 2x2 matrix exists'):
SquareMatrices(**args)
continue
elif symmetry == 'hermitian':
with raises(ConfigError, match='No traceless, unit-determinant, Hermitian 2x2 matrix exists'):
SquareMatrices(**args)
continue
elif shape % 2 == 1: # Odd dimension
if symmetry == 'antisymmetric':
with raises(ConfigError, match='No unit-determinant antisymmetric matrix exists in odd dimensions'):
SquareMatrices(**args)
continue
elif symmetry == 'antihermitian':
with raises(ConfigError, match='No unit-determinant antihermitian matrix exists in odd dimensions'):
SquareMatrices(**args)
continue
if det == 0 and symmetry == 'antisymmetric':
if comp:
with raises(ConfigError, match='Unable to generate complex zero determinant antisymmetric matrices'):
SquareMatrices(**args)
continue
if shape % 2 == 0: # Even dimension
with raises(ConfigError, match='Unable to generate real zero determinant antisymmetric matrices in even dimensions'):
SquareMatrices(**args)
continue
# Matrix exists, so let's test
matrices = SquareMatrices(**args)
if symmetry in ('hermitian', 'antihermitian'):
comp = True
m = matrices.gen_sample()
# MathArray
assert isinstance(m, MathArray)
# Shape
assert m.shape == (shape, shape)
# Norm
if det != 1:
computed_norm = np.linalg.norm(m)
assert norm[0] <= computed_norm or abs(computed_norm - norm[0]) < 1e-12
assert computed_norm <= norm[1] or abs(computed_norm - norm[1]) < 1e-12
# Complex
if not comp:
assert np.array_equal(np.conj(m), m)
# Trace
if traceless:
assert within_tolerance(m.trace(), 0, 5e-13)
# Determinant
if det == 0:
assert within_tolerance(np.abs(np.linalg.det(m)), 0, 1e-12)
elif det == 1:
assert within_tolerance(np.abs(np.linalg.det(m)), 1, 1e-12)
# Symmetries
if symmetry == 'diagonal':
assert np.array_equal(np.diag(np.diag(m)), m)
elif symmetry == 'symmetric':
assert np.array_equal(m, m.T)
elif symmetry == 'antisymmetric':
assert np.array_equal(m, -m.T)
elif symmetry == 'hermitian':
assert np.array_equal(m, np.conj(m.T))
elif symmetry == 'antihermitian':
assert np.array_equal(m, -np.conj(m.T))
