def sample_filter(category1, category2, averaging, nrefits):
if category2 > category1:
return False
return True
def func(category1, category2, averaging, nrefits):
start_time = time.time()
np.random.seed(10 * category1 + category2)
y_train1, y_train2 = get_categories(category1, category2)
np.random.seed()
htest = HTest(dataloader_workers=0,
verbose=1,
distribution="bernoulli",
averaging=averaging)
htest.fit(y_train1, y_train2, nrefits=nrefits)
elapsed_time = time.time() - start_time
return dict(
pvalue=htest.pvalue,
elapsed_time=elapsed_time,
)
do_simulation_study(to_sample,
func,
db,
ResultVAECIFARHTest,
sample_filter=sample_filter)
if ncomparisons == 1 and averaging == "median":
return False
return True
def func(distribution,
no_instances,
dissimilarity,
ncomparisons,
averaging):
def data_gen(size, dim, mu):
res = np.linspace(0.2, 0.9, dim)
res = stats.lognorm.rvs(res, scale=2, size=(size, dim))
res -= stats.lognorm.rvs(0.5, scale=2, size=(size, 1))
res += stats.norm.rvs(loc=mu, scale=2, size=(size, 1))
return res
start_time = time.time()
y_train0 = data_gen(no_instances, 10, 0)
y_train1 = data_gen(no_instances, 10, dissimilarity)
htest = HTest(dataloader_workers=0, verbose=1, averaging=averaging)
htest.fit(y_train0, y_train1, 10000, ncomparisons=ncomparisons)
elapsed_time = time.time() - start_time
return dict(
pvalue=htest.pvalue,
elapsed_time=elapsed_time,
)
do_simulation_study(to_sample, func, db, ResultVAEHTest,
max_count=200, sample_filter=sample_filter)
retrain_permutations = retrain_permutations,
estimator = estimator,
method = method,
)
elif estimator == "rf":
nn_obj = NNPTest(
y_train = y_train,
x_train = np.delete(x_train_n, feature_testedd, 1),
x_to_permutate = x_train_n[:, feature_testedd],
retrain_permutations = retrain_permutations,
estimator = "rf",
method = method,
n_estimators = 300,
)
elif estimator == "linear":
nn_obj = NNPTest(
y_train = y_train,
x_train = np.delete(x_train_n, feature_testedd, 1),
x_to_permutate = x_train_n[:, feature_testedd],
retrain_permutations = retrain_permutations,
estimator = "linear",
method = method,
)
return dict(
pvalue=nn_obj.pvalue, elapsed_time=nn_obj.elapsed_time,
)
do_simulation_study(to_sample, func, db, ResultRealData, max_count=1,
sample_filter=sample_filter)
from cifar_compare_db_structure import ResultVAECIFARCompare, db
from vaecompare import Compare
from sstudy import do_simulation_study
from utils import get_categories
to_sample = dict(
category1 = range(10),
category2 = range(10),
)
def sample_filter(category1, category2):
if category2 > category1:
return False
return True
def func(category1, category2):
start_time = time.time()
y_train1, y_train2 = get_categories(category1, category2)
compare = Compare(dataloader_workers=0, verbose=2,
distribution="bernoulli")
compare.fit(y_train1, y_train2, 10000)
elapsed_time = time.time() - start_time
return dict(
samples=pickle.dumps(compare.samples),
elapsed_time=elapsed_time,
)
do_simulation_study(to_sample, func, db, ResultVAECIFARCompare,
sample_filter=sample_filter, max_count=90)