def perform_me_test(train_miss_impute,
test_miss_impute,
train_full,
test_full,
alpha,
test_locs_miss=None,
gwidth_miss=None,
test_locs_full=None,
gwidth_full=None):
me_result = np.zeros(2)
op = {
'n_test_locs': 10, # number of test locations to optimize
'max_iter': 200, # maximum number of gradient ascent iterations
'locs_step_size': 1.0, # step size for the test locations (features)
'gwidth_step_size': 0.1, # step size for the Gaussian width
'tol_fun':
1e-4, # stop if the objective does not increase more than this.
'seed': 0 # random seed
}
sb_data_miss_impute = TSTData(train_miss_impute, test_miss_impute)
train_miss_impute_sb, dumy = sb_data_miss_impute.split_tr_te(
tr_proportion=1, seed=1)
dumy, test_miss_impute_sb = sb_data_miss_impute.split_tr_te(
tr_proportion=0, seed=1)
#half_size = int(train_miss_impute.shape[0]/2)
#train_miss_impute_sb = TSTData(train_miss_impute[:half_size], train_miss_impute[half_size:half_size*2])
#test_miss_impute_sb = TSTData(train_miss_impute, test_miss_impute)
if test_locs_miss is None:
test_locs_miss, gwidth_miss, info = tst.MeanEmbeddingTest.optimize_locs_width(
train_miss_impute_sb, alpha, **op)
met_opt = tst.MeanEmbeddingTest(test_locs_miss, gwidth_miss, alpha)
test_result = met_opt.perform_test(test_miss_impute_sb)
if test_result['h0_rejected']:
me_result[0] = 1
sb_data_full = TSTData(train_full, test_full)
train_full_sb, dumy = sb_data_full.split_tr_te(tr_proportion=1, seed=1)
dumy, test_full_sb = sb_data_full.split_tr_te(tr_proportion=0, seed=1)
if test_locs_full is None:
test_locs_full, gwidth_full, info = tst.MeanEmbeddingTest.optimize_locs_width(
train_full_sb, alpha, **op)
met_opt = tst.MeanEmbeddingTest(test_locs_full, gwidth_full, alpha)
test_result = met_opt.perform_test(test_full_sb)
if test_result['h0_rejected']:
me_result[1] = 1
return me_result, test_locs_miss, gwidth_miss, test_locs_full, gwidth_full
def perform_mmd_test(train_miss_impute,
test_miss_impute,
train_full,
test_full,
alpha,
mmd_miss_impute=None,
mmd_full=None):
mmd_result = np.zeros(2)
sb_data_miss_impute = TSTData(train_miss_impute, test_miss_impute)
if mmd_miss_impute is None:
print('ini')
x, y = sb_data_miss_impute.xy()
dist_mat_miss_impute = metrics.pairwise_distances(x, y)
the_kernel = kernel.KGauss(dist_mat_miss_impute.std())
mmd_miss_impute = tst.QuadMMDTest(the_kernel, alpha=alpha)
test_result = mmd_miss_impute.perform_test(sb_data_miss_impute)
if test_result['h0_rejected']:
mmd_result[0] = 1
sb_data_full = TSTData(train_full, test_full)
if mmd_full is None:
x, y = sb_data_full.xy()
dist_mat_full = metrics.pairwise_distances(x, y)
the_kernel = kernel.KGauss(dist_mat_full.std())
mmd_full = tst.QuadMMDTest(the_kernel, alpha=alpha)
test_result = mmd_full.perform_test(sb_data_full)
if test_result['h0_rejected']:
mmd_result[1] = 1
return mmd_result, mmd_miss_impute, mmd_full
def perform_mmd_test(train_miss_impute,
test_miss_impute,
train_full,
test_full,
alpha,
mmd_miss_impute=None,
mmd_full=None):
mmd_result = np.zeros(2)
sb_data_miss_impute = TSTData(train_miss_impute, test_miss_impute)
if mmd_miss_impute is None:
print('ini')
the_kernel = kernel.KGauss(sb_data_miss_impute.mean_std())
mmd_miss_impute = tst.QuadMMDTest(the_kernel, alpha=alpha)
test_result = mmd_miss_impute.perform_test(sb_data_miss_impute)
if test_result['h0_rejected']:
mmd_result[0] = 1
sb_data_full = TSTData(train_full, test_full)
if mmd_full is None:
the_kernel = kernel.KGauss(sb_data_full.mean_std())
mmd_full = tst.QuadMMDTest(the_kernel, alpha=alpha)
test_result = mmd_full.perform_test(sb_data_full)
if test_result['h0_rejected']:
mmd_result[1] = 1
return mmd_result, mmd_miss_impute, mmd_full