Merge pull request #433 from jakobrunge/developer

fixed issue with get_measure for variable data type

Author: jakobrunge

setup.py

@@ -61,7 +61,7 @@ def run(self):
 # Run the setup
 setup(
     name="tigramite",
-    version="5.2.6.5",
+    version="5.2.6.6",
     packages=["tigramite", "tigramite.independence_tests", "tigramite.toymodels"],
     license="GNU General Public License v3.0",
     description="Tigramite causal inference for time series",

tigramite/data_processing.py

@@ -1574,18 +1574,18 @@ def nonlin_f(x): return (x + 5. * x**2 * np.exp(-x**2 / 20.))
 
     print(frame.T)
 
-    X=[(0, 0)]
-    Y=[(0, 0)]
-    Z=[(0, -3)]
-    tau_max=5
-    frame.construct_array(X, Y, Z, tau_max,
-                        extraZ=None,
-                        mask=None,
-                        mask_type=None,
-                        return_cleaned_xyz=False,
-                        do_checks=True,
-                        cut_off='2xtau_max',
-                        verbosity=4)
-
-    print(ordinal_patt_array(data, array_mask=None, dim=2, step=1,
-                        weights=False, verbosity=0)[0])
+    # X=[(0, 0)]
+    # Y=[(0, 0)]
+    # Z=[(0, -3)]
+    # tau_max=5
+    # frame.construct_array(X, Y, Z, tau_max,
+    #                     extraZ=None,
+    #                     mask=None,
+    #                     mask_type=None,
+    #                     return_cleaned_xyz=False,
+    #                     do_checks=True,
+    #                     cut_off='2xtau_max',
+    #                     verbosity=4)
+
+    # print(ordinal_patt_array(data, array_mask=None, dim=2, step=1,
+    #                     weights=False, verbosity=0)[0])

tigramite/independence_tests/independence_tests_base.py

@@ -837,15 +837,45 @@ def get_measure(self, X, Y, Z=None, tau_max=0,
             The test statistic value.
 
         """
-        # Make the array
-        array, xyz, (X, Y, Z), _ = self._get_array(X=X, Y=Y, Z=Z, tau_max=tau_max,
-                                            remove_constant_data=False)
-        D, T = array.shape
-        # Check it is valid
-        if np.isnan(array).sum() != 0:
+
+        if self.significance == 'fixed_thres' and alpha_or_thres is None:
+            raise ValueError("significance == 'fixed_thres' requires setting alpha_or_thres")
+
+        # Get the array to test on
+        (array, xyz, XYZ, data_type, 
+         nonzero_array, nonzero_xyz, nonzero_XYZ, nonzero_data_type) = self._get_array(
+                                            X=X, Y=Y, Z=Z, tau_max=tau_max,
+                                            remove_constant_data=True, 
+                                            verbosity=self.verbosity)
+        X, Y, Z = XYZ
+        nonzero_X, nonzero_Y, nonzero_Z = nonzero_XYZ
+
+        # Record the dimensions
+        # dim, T = array.shape
+
+        # Ensure it is a valid array
+        if np.any(np.isnan(array)):
             raise ValueError("nans in the array!")
-        # Return the dependence measure
-        return self._get_dependence_measure_recycle(X, Y, Z, xyz, array)
+
+        # If all X or all Y are zero, then return pval=1, val=0, dependent=False
+        if len(nonzero_X) == 0 or len(nonzero_Y) == 0:
+            val = 0.
+        else:
+            # Get the dependence measure, reycling residuals if need be
+            val = self._get_dependence_measure_recycle(nonzero_X, nonzero_Y, nonzero_Z, 
+                                        nonzero_xyz, nonzero_array, nonzero_data_type)
+          
+        return val
+
+        # # Make the array
+        # array, xyz, (X, Y, Z), _ = self._get_array(X=X, Y=Y, Z=Z, tau_max=tau_max,
+        #                                     remove_constant_data=False)
+        # D, T = array.shape
+        # # Check it is valid
+        # if np.isnan(array).sum() != 0:
+        #     raise ValueError("nans in the array!")
+        # # Return the dependence measure
+        # return self._get_dependence_measure_recycle(X, Y, Z, xyz, array)
 
     def get_confidence(self, X, Y, Z=None, tau_max=0,
                        data_type=None):

tigramite/pcmci.py

@@ -2855,7 +2855,6 @@ def _run_pcalg_test(self, graph, i, abstau, j, S, lagged_parents, max_conds_py,
             val, pval, dependent = self.cond_ind_test.run_test(X=[(i, -abstau)], Y=[(j, 0)],
                                                 Z=Z, tau_max=tau_max,
                                                 alpha_or_thres=alpha_or_thres,
-                                                # verbosity=self.verbosity
                                                 )
 
         return val, pval, Z, dependent
@@ -3925,7 +3924,8 @@ def _optimize_pcmciplus_alpha(self,
 
 if __name__ == '__main__':
     from tigramite.independence_tests.parcorr import ParCorr
-    from tigramite.independence_tests.cmiknn import CMIknn
+    from tigramite.independence_tests.regression_ci import RegressionCI
+    # from tigramite.independence_tests.cmiknn import CMIknn
 
     import tigramite.data_processing as pp
     from tigramite.toymodels import structural_causal_processes as toys
@@ -4009,39 +4009,20 @@ def _optimize_pcmciplus_alpha(self,
     # data, _ = toys.structural_causal_process(links_coeffs, T=T, seed=3)
     # T, N = data.shape
 
-    # # Initialize dataframe object
-    # dataframe = pp.DataFrame(data)
-    # pcmci = PCMCI(
-    #     dataframe=dataframe, 
-    #     cond_ind_test=ParCorr(),
-    #     verbosity=0)
-
-    # multidata[0][40:100, :] = 999.
-
-    # dataframe = pp.DataFrame(multidata, analysis_mode='multiple',
-    #         missing_flag = 999.,
-    #         time_offsets = {0:50, 1:0}
-    #          # reference_points=list(range(500, 1000))
-    #          ) 
-
-    # pcmci = PCMCI(dataframe=dataframe, 
-    #     cond_ind_test=ParCorr(verbosity=0), verbosity=0)
 
-    # # results = pcmci.run_pcmciplus(tau_max=1)
+    multidata = np.random.randn(10, 100, 5)
+    data_type = np.zeros((10, 100, 5), dtype='bool')
+    data_type[:,:,:3] = True
 
-    # results = pcmci.run_sliding_window_of(
-    #     window_step=499, window_length=500,
-    #     method='run_pcmciplus', method_args={'tau_max':1, 
-    #     'link_assumptions':{
-    #     0: {(0, -1): '-->'},
-    #     1: {(1, -1): '-->', (0, -1): '-!>'},
-    #     }
-    #     })
+    dataframe = pp.DataFrame(multidata, 
+        data_type=data_type,
+        analysis_mode='multiple',
+            missing_flag = 999.,
+            time_offsets = {0:50, 1:0}
+             # reference_points=list(range(500, 1000))
+             ) 
 
-    # # tp.plot_graph(results['graph'])
-    # print(multidata[0].shape, multidata[1].shape)
-    # print(results['window_results']['val_matrix'])
-    # print(results['window_results']['val_matrix'][0][0,1])
-    # print(results['window_results']['val_matrix'][1][0,1])
+    pcmci = PCMCI(dataframe=dataframe, 
+        cond_ind_test=RegressionCI(verbosity=0), verbosity=0)
 
-    # plt.show()
+    # results = pcmci.run_pcmciplus(tau_max=1)

tigramite/toymodels/structural_causal_processes.py

@@ -738,7 +738,7 @@ def structural_causal_process(links, T, noises=None,
         elif isinstance(intervention_type, str):
             intervention_type = {j:intervention_type for j in intervention}
         for j in intervention.keys():
-            if len(intervention[j]) != T:
+            if len(np.atleast_1d(intervention[j])) != T:
                 raise ValueError("intervention array for j=%s must be of length T = %d" %(j, T))
             if j not in intervention_type.keys():        
                 raise ValueError("intervention_type dictionary must contain entry for %s" %(j))