Merge pull request #361 from jakobrunge/developer

Developer

tests/test_PairwiseMultCI.py

@@ -0,0 +1,98 @@
+"""
+Tests for independence_tests.py.
+"""
+from __future__ import print_function
+import numpy as np
+import pytest
+from scipy import stats
+
+from tigramite.independence_tests.parcorr import ParCorr
+from tigramite.independence_tests.robust_parcorr import RobustParCorr
+from tigramite.independence_tests.pairwise_CI import PairwiseMultCI
+from tigramite.independence_tests.cmiknn import CMIknn
+from tigramite.independence_tests.regressionCI import RegressionCI
+# PairwiseMultCI TESTING ##################################################
+@pytest.fixture(params=[
+ # Generate PairwiseMultCI test instances
+ # test PairwiseMultCI for different combinations of:
+ # significance (of the univariate tests), alpha_pre, pre_step_sample_fraction, cond_ind_test, cond_ind_test_thres, cond_ind_test_thres_pre
+ ('fixed_thres', 0.5, 0.2, ParCorr, 1, 0.5),
+ ('fixed_thres', 0.5, 0.5, ParCorr, 1, 0.5),
+ ('fixed_thres', 0.8, 0.2, ParCorr, 1, 0.5),
+ ('fixed_thres', 0.8, 0.5, ParCorr, 1, 0.5),
+ ('analytic', 0.5, 0.2, ParCorr, None, None),
+ ('shuffle_test', 0.5, 0.2, ParCorr, None, None),
+ ('analytic', 0.5, 0.5, ParCorr, None, None),
+ ('shuffle_test', 0.5, 0.5, ParCorr, None, None),
+ ('analytic', 0.8, 0.2, ParCorr, None, None),
+ ('shuffle_test', 0.8, 0.2, ParCorr, None, None),
+ ('analytic', 0.8, 0.5, ParCorr, None, None),
+ ('shuffle_test', 0.8, 0.5, ParCorr, None, None),
+ ('analytic', 0.5, 0.2, RobustParCorr, None, None),
+ ('shuffle_test', 0.5, 0.2, RobustParCorr, None, None),
+ ('analytic', 0.5, 0.5, RobustParCorr, None, None),
+ ('shuffle_test', 0.5, 0.5, RobustParCorr, None, None),
+ ('analytic', 0.8, 0.2, RobustParCorr, None, None),
+ ('shuffle_test', 0.8, 0.2, RobustParCorr, None, None),
+ ('analytic', 0.8, 0.5, RobustParCorr, None, None),
+ ('shuffle_test', 0.8, 0.5, RobustParCorr, None, None),
+ ('shuffle_test', 0.5, 0.2, CMIknn, None, None),
+ ('shuffle_test', 0.5, 0.5, CMIknn, None, None),
+ ('shuffle_test', 0.8, 0.2, CMIknn, None, None),
+ ('shuffle_test', 0.8, 0.5, CMIknn, None, None),
+])
+
+def pairwise_mult_ci(request):
+ # Unpack the parameters
+ sig, alpha_pre, pre_step_sample_fraction, cond_ind_test, cond_ind_test_thres, cond_ind_test_thres_pre = request.param
+ # Generate the par_corr_wls independence test
+ if sig != "fixed_thres":
+ return PairwiseMultCI(cond_ind_test = cond_ind_test(significance = sig),
+ alpha_pre = alpha_pre,
+ pre_step_sample_fraction = pre_step_sample_fraction)
+ else:
+ return PairwiseMultCI(cond_ind_test = cond_ind_test(significance = sig),
+ alpha_pre = None,
+ pre_step_sample_fraction=pre_step_sample_fraction,
+ significance= sig,
+ fixed_thres_pre = 2
+ )
+
+@pytest.fixture(params=[
+ # Generate PairwiseMultCI test instances
+ # test PairwiseMultCI for different combinations of:
+ # seed, true_dep, T (=sample size)
+ (123, 0, 100),
+ (123, 0, 1000),
+ (123, 0.2, 100),
+ (123, 0.2, 1000),
+ (123, 0.5, 100),
+ (123, 0, 1000),
+ (46, 0, 100),
+ (46, 0, 1000),
+ (46, 0.2, 100),
+ (46, 0.2, 1000),
+ (46, 0.5, 100),
+ (46, 0, 1000),
+])
+
+def data_sample_c_cc_c(request):
+ # Set the random seed
+ seed, true_dep, T = request.param
+ np.random.seed(seed)
+ x = np.random.normal(0, 1, T).reshape(T, 1)
+ y1 = np.random.normal(0, 1, T).reshape(T, 1)
+ y2 = true_dep * x + y1 + 0.3 * np.random.normal(0, 1, T).reshape(T, 1)
+ y = np.hstack((y1,y2))
+ z = np.random.normal(0, 1, T).reshape(T, 1)
+ # Return data xyz
+ return x, y, z
+
+def test_pairwise_mult_ci(pairwise_mult_ci, data_sample_c_cc_c):
+ # Get the data sample values
+ x, y, z = data_sample_c_cc_c
+ # Get the analytic significance
+ test_result = pairwise_mult_ci.run_test_raw(x = x, y = y, z = z, alpha_or_thres=1)
+ val = test_result[0]
+ pval = test_result[1]
+ np.testing.assert_allclose(pval, 0.5, atol=0.5)

tigramite/independence_tests/gpdc.py

@@ -27,6 +27,9 @@ class GaussProcReg():
  a \*.npz file containing the null distribution for different sample sizes.
  This file can then be supplied as null_dist_filename.
 
+ Assumes one-dimensional X, Y. But can be combined with PairwiseMultCI to
+ obtain a test for multivariate X, Y.
+
  Parameters
  ----------
  null_samples : int

tigramite/independence_tests/gpdc_torch.py

@@ -25,6 +25,9 @@ class GaussProcRegTorch():
  a \*.npz file containing the null distribution for different sample sizes.
  This file can then be supplied as null_dist_filename.
 
+ Assumes one-dimensional X, Y. But can be combined with PairwiseMultCI to
+ obtain a test for multivariate X, Y.
+
  Parameters
  ----------
  null_samples : int

tigramite/independence_tests/gsquared.py

@@ -23,7 +23,8 @@ class Gsquared(CondIndTest):
  above 1000-2000 samples (depends on data). For smaller sample sizes use the
  CMIsymb class which includes a local permutation test.
 
- Assumes one-dimensional X, Y.
+ Assumes one-dimensional X, Y. But can be combined with PairwiseMultCI to
+ obtain a test for multivariate X, Y.
 
  This method requires the scipy.stats package.