def fill_params_dict_list_sample_sizes(base_path, do_fast_mgc=False):
mcorr = DCorr(which_test='unbiased')
dcorr = DCorr(which_test='biased')
mantel = DCorr(which_test='mantel')
mgc = MGC()
hhg = HHG()
pearson = RVCorr(which_test='pearson')
mdmr = MDMR()
independence_tests = [mcorr, dcorr, mantel, mgc, hhg, pearson, mdmr]
params_dict_list = []
for sim_name, sim_func in simulations.items():
for test in independence_tests:
params_dict = {
'independence_test': test,
'simulation_type': sim_func[1],
'base_path': base_path,
'additional_params': {}
}
params_dict_list.append(params_dict)
if do_fast_mgc:
fast_mgc = MGC()
additional_params = {"is_fast": True}
params_dict = {
'independence_test': fast_mgc,
'simulation_type': sim_func[1],
'base_path': base_path,
'additional_params': additional_params
}
params_dict_list.append(params_dict)
return params_dict_list
def fill_params_dict_list_dimensions(do_fast_mgc=False):
mcorr = DCorr(which_test='unbiased')
dcorr = DCorr(which_test='biased')
mantel = DCorr(which_test='mantel')
mgc = MGC()
rv_corr = RVCorr(which_test='rv')
hhg = HHG()
cca = RVCorr(which_test='cca')
mdmr = MDMR()
independence_tests = [] # [mgc, mcorr, dcorr, mantel, rv_corr, cca]
params_dict_list = []
for sim_name, sim_func in simulations.items():
for test in independence_tests:
params_dict = {
'independence_test': test,
'simulation_type': sim_func[1],
'dim': find_dim(sim_name),
'additional_params': {}
}
params_dict_list.append(params_dict)
if do_fast_mgc:
fast_mgc = MGC()
additional_params = {"is_fast": True}
params_dict = {
'independence_test': fast_mgc,
'simulation_type': sim_func[1],
'dim': find_dim(sim_name),
'additional_params': additional_params
}
params_dict_list.append(params_dict)
return params_dict_list
def paired_two_sample_test_dcorr(x,
y,
which_test="biased",
compute_distance_matrix=None,
is_fast=False):
'''
Compute paired two sample test's DCorr test_statistic
:param X: is interpreted as either:
- a ``[n*n]`` distance matrix, a square matrix with zeros on diagonal for n samples OR
- a ``[n*p]`` data matrix, a matrix with n samples in p dimensions
:type X: 2D numpy.array
:param Y: is interpreted as either:
- a ``[n*n]`` distance matrix, a square matrix with zeros on diagonal for n samples OR
- a ``[n*p]`` data matrix, a matrix with n samples in p dimensions
:type Y: 2D numpy.array
:return: paired two sample DCorr test_statistic
:rtype: float
'''
assert x.shape == y.shape, "Matrices X and Y need to be of same dimensions [n, p]"
dcorr = DCorr(is_paired=True,
which_test=which_test,
compute_distance_matrix=compute_distance_matrix)
return dcorr.p_value(x, y, is_fast=is_fast)
def fill_params_dict_list_epsilons(base_path, do_fast_mgc=False):
mcorr = DCorr(which_test='unbiased')
mgc = MGC()
manova = Manova()
independence_tests = [manova, mcorr, mgc]
three_sample_simulation_types = [1, 2, 3]
params_dict_list = []
for sim_type in three_sample_simulation_types:
for test in independence_tests:
params_dict = {
'independence_test': test,
'simulation_type': sim_type,
'base_path': base_path,
'additional_params': {}
}
params_dict_list.append(params_dict)
if do_fast_mgc:
fast_mgc = MGC()
additional_params = {"is_fast": True}
params_dict = {
'independence_test': fast_mgc,
'simulation_type': sim_type,
'base_path': base_path,
'additional_params': additional_params
}
params_dict_list.append(params_dict)
return params_dict_list
def fill_params_dict_list_sample_sizes():
mcorr = DCorr(which_test='unbiased')
dcorr = DCorr(which_test='biased')
mantel = DCorr(which_test='mantel')
mgc = MGC()
hhg = HHG()
pearson = RVCorr(which_test='pearson')
independence_tests = [] # [mgc, mcorr, dcorr, mantel, pearson]
params_dict_list = []
for sim_name, sim_func in simulations.items():
for test in independence_tests:
params_dict = {
'independence_test': test,
'simulation_type': sim_func[1]
}
params_dict_list.append(params_dict)
return params_dict_list
def fill_params_dict_list_dimensions():
mcorr = DCorr(which_test='unbiased')
dcorr = DCorr(which_test='biased')
mantel = DCorr(which_test='mantel')
mgc = MGC()
rv_corr = RVCorr(which_test='rv')
hhg = HHG()
cca = RVCorr(which_test='cca')
mdmr = MDMR()
# initialize additional test
# add the test that you wish to run in the `independence_tests` list
independence_tests = [] # [mgc, mcorr, dcorr, mantel, rv_corr, cca]
params_dict_list = []
for sim_name, sim_func in simulations.items():
for test in independence_tests:
params_dict = {
'independence_test': test,
'simulation_type': sim_func[1],
'dim': find_dim(sim_name)
}
params_dict_list.append(params_dict)
return params_dict_list
def __init__(self,
compute_distance_matrix=None,
which_test='unbiased',
max_lag=0):
'''
:param compute_distance_matrix: a function to compute the pairwise distance matrix, given a data matrix
:type compute_distance_matrix: FunctionType or callable()
:param which_test: the type of distance covariance estimate to use, can be 'unbiased', 'biased' 'mantel'
:type which_test: string
:param max_lag: Maximum lead/lag to check for dependence between X_t and Y_t+j (M parameter)
:type max_lag: int
'''
IndependenceTest.__init__(self)
if which_test not in ['unbiased', 'biased']:
raise ValueError('which_test must be unbiased or biased.')
self.which_test = which_test
self.dcorr = DCorr(which_test=self.which_test)
self.max_lag = max_lag
def test_power():
test = DCorr(which_test='unbiased')
simulation_type = 4
sim = joint_sim
sample_sizes = [i for i in range(5, 101, 5)]
matlab_file_name = './mgcpy/benchmarks/matlab_power_results/sample_size/CorrIndTestType{}N100Dim1.mat'.format(
simulation_type)
with h5py.File(matlab_file_name, 'r') as f:
matlab_results = {}
for k, v in f.items():
matlab_results[k] = np.transpose(np.array(v))
matlab_power = matlab_results['powerM'][0, :]
estimated_power = np.zeros(len(sample_sizes))
for i in range(len(sample_sizes)):
estimated_power[i] = power(test,
sim,
num_samples=sample_sizes[i],
num_dimensions=1)
assert np.allclose(estimated_power, matlab_power, atol=0.2)
def test_dcorr_p_value():
'''
test p value
analytical p value for unbiased dcorr is compared with R package energy
other p values are compared with the permutation tests in mgc-paper
the value is the mean and atol is set to 4 times standard deviation
'''
dir_name = './mgcpy/independence_tests/unit_tests/dcorr/data/'
unbiased = DCorr(which_test='unbiased')
biased = DCorr(which_test='biased')
mantel = DCorr(which_test='mantel')
X = np.genfromtxt(dir_name + 'pvalue_X_mtx.csv', delimiter=',')
Y = np.genfromtxt(dir_name + 'pvalue_Y_mtx.csv', delimiter=',')
Y = Y[:, np.newaxis]
# p value
assert np.allclose(unbiased.p_value(X, Y)[0], 0.0640, atol=0.03)
assert np.allclose(biased.p_value(X, Y)[0], 0.0510, atol=0.03)
assert np.allclose(mantel.p_value(X, Y)[0], 0.1020, atol=0.03)
# p value (faster versions)
assert np.allclose(unbiased.p_value(X, Y, is_fast=True)[0],
0.7429,
atol=0.03)
assert np.allclose(biased.p_value(X, Y, is_fast=True)[0],
1 / 1000,
atol=0.03)
assert np.allclose(mantel.p_value(X, Y, is_fast=True)[0],
1 / 1000,
atol=0.03)
def test_dcorr_stat():
# test the special case when one of the dataset has zero variance
X = np.array([1, 1, 1])[:, np.newaxis]
Y = np.array([1, 2, 3])[:, np.newaxis]
unbiased = DCorr(which_test='unbiased')
assert np.allclose(unbiased.test_statistic(X, Y)[0], 0)
dir_name = './mgcpy/independence_tests/unit_tests/dcorr/data/'
X = np.genfromtxt(dir_name + 'test_stat_X_mtx.csv', delimiter=',')
Y = np.genfromtxt(dir_name + 'test_stat_Y_mtx.csv', delimiter=',')
Y = Y[:, np.newaxis]
unbiased = DCorr(which_test='unbiased')
biased = DCorr(which_test='biased')
mantel = DCorr(which_test='mantel')
# test get_name
assert unbiased.get_name() == 'unbiased'
assert biased.get_name() == 'biased'
assert mantel.get_name() == 'mantel'
# test statistic
assert np.allclose(unbiased.test_statistic(X, Y)[0], 0.1174, atol=1e-4)
assert np.allclose(biased.test_statistic(X, Y)[0], 0.1548, atol=1e-4)
assert np.allclose(mantel.test_statistic(X, Y)[0], 0.2421, atol=1e-4)
# test statistic (fast versions)
assert np.allclose(unbiased.test_statistic(X, Y, is_fast=True)[0],
0.1562,
atol=1e-4)
assert np.allclose(biased.test_statistic(X, Y, is_fast=True)[0],
0.3974,
atol=1e-4)
assert np.allclose(mantel.test_statistic(X, Y, is_fast=True)[0],
0.3392,
atol=1e-4)
# additional test for mantel
X = np.genfromtxt(dir_name + 'mantel_test_stat_X_mtx.csv', delimiter=',')
Y = np.genfromtxt(dir_name + 'mantel_test_stat_Y_mtx.csv', delimiter=',')
X = X[:, np.newaxis]
Y = Y[:, np.newaxis]
assert np.allclose(mantel.test_statistic(X, Y)[0], 0.7115, atol=1e-4)
assert np.allclose(mantel.test_statistic(X, Y, is_fast=True)[0],
0.4575,
atol=1e-4) # faster version
sns.set_palette("deep")
n_sims = 100
n_samples = 100
n_components = 2
n_permutations = 1000
size = (n_samples, n_components)
#%% mgcpy package
p_vals = np.zeros(n_sims)
for i in tqdm(range(n_sims)):
sample1 = np.random.uniform(0.2, 0.7, size=size)
sample2 = np.random.uniform(0.2, 0.7, size=size)
sample, indicator = k_sample_transform(sample1, sample2)
test = DCorr(which_test="unbiased")
p, p_meta = test.p_value(
sample, indicator, replication_factor=n_permutations, is_fast=False
)
p_vals[i] = p
plt.figure()
sns.distplot(p_vals)
plt.title("MGCPy DCorr, 2-sample under null, unbiased, not fast")
plt.xlabel("p-value")
plt.savefig("graspy-misc/profile_dcorr/mgcpy_dcorr.png", facecolor="w")
#%% mgcpy with is_fast=True
# p_vals = np.zeros(n_sims)
# for i in tqdm(range(n_sims)):
# sample1 = np.random.uniform(0.2, 0.7, size=size)
def test_dcorr():
# test the special case when one of the dataset has zero variance
X = np.array([1, 1, 1])[:, np.newaxis]
Y = np.array([1, 2, 3])[:, np.newaxis]
unbiased = DCorr(which_test='unbiased')
assert np.allclose(unbiased.test_statistic(X, Y)[0], 0)
dir_name = './mgcpy/independence_tests/unit_tests/dcorr/data/'
X = np.genfromtxt(dir_name + 'test_stat_X_mtx.csv', delimiter=',')
Y = np.genfromtxt(dir_name + 'test_stat_Y_mtx.csv', delimiter=',')
Y = Y[:, np.newaxis]
unbiased = DCorr(which_test='unbiased')
biased = DCorr(which_test='biased')
mantel = DCorr(which_test='mantel')
# test get_name
assert unbiased.get_name() == 'unbiased'
assert biased.get_name() == 'biased'
assert mantel.get_name() == 'mantel'
# test statistic
assert np.allclose(unbiased.test_statistic(X, Y)[0], 0.1174, atol=1e-4)
assert np.allclose(biased.test_statistic(X, Y)[0], 0.1548, atol=1e-4)
assert np.allclose(mantel.test_statistic(X, Y)[0], 0.2421, atol=1e-4)
# test statistic (fast versions)
assert np.allclose(unbiased.test_statistic(X, Y, is_fast=True)[0],
0.1562,
atol=1e-4)
assert np.allclose(biased.test_statistic(X, Y, is_fast=True)[0],
0.3974,
atol=1e-4)
assert np.allclose(mantel.test_statistic(X, Y, is_fast=True)[0],
0.3392,
atol=1e-4)
# additional test for mantel
X = np.genfromtxt(dir_name + 'mantel_test_stat_X_mtx.csv', delimiter=',')
Y = np.genfromtxt(dir_name + 'mantel_test_stat_Y_mtx.csv', delimiter=',')
X = X[:, np.newaxis]
Y = Y[:, np.newaxis]
assert np.allclose(mantel.test_statistic(X, Y)[0], 0.7115, atol=1e-4)
assert np.allclose(mantel.test_statistic(X, Y, is_fast=True)[0],
0.7552,
atol=1e-4) # faster version
'''
test p value
analytical p value for unbiased dcorr is compared with R package energy
other p values are compared with the permutation tests in mgc-paper
the value is the mean and atol is set to 4 times standard deviation
'''
X = np.genfromtxt(dir_name + 'pvalue_X_mtx.csv', delimiter=',')
Y = np.genfromtxt(dir_name + 'pvalue_Y_mtx.csv', delimiter=',')
Y = Y[:, np.newaxis]
# p value
assert np.allclose(unbiased.p_value(X, Y)[0], 0.0640, atol=0.03)
assert np.allclose(biased.p_value(X, Y)[0], 0.0510, atol=0.03)
assert np.allclose(mantel.p_value(X, Y)[0], 0.1020, atol=0.03)
# p value (faster versions)
assert np.allclose(unbiased.p_value(X, Y, is_fast=True)[0],
0.7429,
atol=0.03)
assert np.allclose(biased.p_value(X, Y, is_fast=True)[0], 0, atol=0.03)
assert np.allclose(mantel.p_value(X, Y, is_fast=True)[0], 0, atol=0.03)