Merge pull request #17 from WFHong/WFHong

fix some bugs

Author: miumiujiang12138

cdmir/discovery/funtional_based/LearningHierarchicalStructure/Causal_Discovery_in_LHM.py

@@ -8,9 +8,8 @@
 import Utils
 
 
-
 def Causal_Discovery_LHM(data, alpha=0.01):
-    '''
+    """
     Function: Causal Discovery in Linear Latent Hierarchical Structure
     Parameter
         data: DataFrame (pandas)
@@ -23,82 +22,72 @@ def Causal_Discovery_LHM(data, alpha=0.01):
         Graph (selected)
             the Causal graph of hierarchical structure
 
-    '''
-    #Initize Variable
-    Current_Clusters=[]
-    PureClusters=[]
-    GeneralPureClusters=[]
-    ImpureClusters=[]
+    """
+    # Initize Variable
+    Current_Clusters = []
+    PureClusters = []
+    GeneralPureClusters = []
+    ImpureClusters = []
 
-    AllCausalCluster=[]
+    AllCausalCluster = []
 
-    LatentIndex={}
+    LatentIndex = {}
     LatentNum = 1
     Ora_data = data.copy()
 
-
     while True:
 
         '''Begin recursive procedure'''
 
-        #Phase I: Finding causal cluster and judge the purity or impurity
-        Current_Clusters, PureClusters, ImpureClusters, PClusters=FindCausalCluster.FindCausalCluster(data, PureClusters, ImpureClusters, alpha)
+        # Phase I: Finding causal cluster and judge the purity or impurity
+        Current_Clusters, PureClusters, ImpureClusters, PClusters = FindCausalCluster.FindCausalCluster(data,
+                                                                                                        PureClusters,
+                                                                                                        ImpureClusters,
+                                                                                                        alpha)
         AllCausalCluster = Utils.ExtendList(AllCausalCluster, PClusters)
         AllCausalCluster = Utils.ExtendList(AllCausalCluster, Current_Clusters)
 
-        #debug
+        # debug
         print('Finished Finding Causal Cluster: ', Current_Clusters, PureClusters, ImpureClusters, PClusters)
 
+        # Phase II: Check merge rule for the learned clusters and update record variables
 
-        #Phase II: Check merge rule for the learned clusters and update record variables
-
-        Merge_Results, PureClusters, ImpureClusters, AllCausalCluster, GeneralPureClusters, LatentIndex = MC.MergeCausalCluster(Current_Clusters, PureClusters, ImpureClusters, AllCausalCluster, GeneralPureClusters, LatentIndex, data, Ora_data, alpha)
+        Merge_Results, PureClusters, ImpureClusters, AllCausalCluster, GeneralPureClusters, LatentIndex = MC.MergeCausalCluster(
+            Current_Clusters, PureClusters, ImpureClusters, AllCausalCluster, GeneralPureClusters, LatentIndex, data,
+            Ora_data, alpha)
 
         MergeCluster = Merge_Results[0]
         EarlyLearningImpureClusters = Merge_Results[1]
         EarlyLearningRemoveClusters = Merge_Results[2]
         IntroduceLatent_PureClusters = Merge_Results[3]
         RemainingVariables = Merge_Results[4]
 
-        print('Merge_Results: ',Merge_Results)
+        print('Merge_Results: ', Merge_Results)
         print(LatentIndex)
 
-
-
         if len(MergeCluster) == 0 and len(IntroduceLatent_PureClusters) == 0:
             print('This is nothing be learned !')
-            if len(RemainingVariables) == 0 and len(EarlyLearningImpureClusters) !=0:
+            if len(RemainingVariables) == 0 and len(EarlyLearningImpureClusters) != 0:
                 print('There are something wrong! In the merger Results !!', EarlyLearningImpureClusters)
                 exit(-1)
             elif len(RemainingVariables) <= 3:
                 print('Recursive Procedure Finished ! The structure is identified up to a Markov equivalent class.')
                 print(LatentIndex, ImpureClusters)
                 break
 
-
-
-        #Phase III: Introduce latent variable into the graph and update the actived data set
+        # Phase III: Introduce latent variable into the graph and update the actived data set
         data, LatentNum, LatentIndex = UpdataData.UpdataData(Merge_Results, LatentNum, LatentIndex, data)
 
-        if len(data) <=1:
+        if len(data) <= 1:
             print(LatentIndex, ImpureClusters)
             break
 
         print(data, LatentNum, LatentIndex, ImpureClusters)
 
         '''End recursive procedure'''
 
-    MakeGraph.Make_graph_Impure(LatentIndex,ImpureClusters)
+    MakeGraph.Make_graph_Impure(LatentIndex, ImpureClusters)
 
-    #Phase IV: orientation causal direction among latent variable, including latent measured
+    # Phase IV: orientation causal direction among latent variable, including latent measured
     Orientation.Orientation_Cluster(Ora_data, LatentIndex, PureClusters, AllCausalCluster)
-    #ImpureOrder = Orientation.Orientation_ImpureCluster(Ora_data, LatentIndex, PureClusters, AllCausalCluster, ImpureClusters)
-
-
-
-
-
-
-
-
-
+    # ImpureOrder = Orientation.Orientation_ImpureCluster(Ora_data, LatentIndex, PureClusters, AllCausalCluster, ImpureClusters)

cdmir/discovery/funtional_based/LearningHierarchicalStructure/FindCausalCluster.py

@@ -7,8 +7,9 @@
 import TetradMethod
 import VanishedTest as VT
 
+
 def FindCausalCluster(data, PureCluster, ImpureCluster, alpha=0.01):
-    '''
+    """
     Function: Find Causal Cluster from the actived data set and check the purity
     Parameter
         data: DataFrame
@@ -29,55 +30,48 @@ def FindCausalCluster(data, PureCluster, ImpureCluster, alpha=0.01):
             Update ImpureCluster set by adding the current learned clusters
         PClusters: List
             Return only two element cluster
-    '''
-    LearnedClusters=[]
-    PClusters=[]
-    indexs=list(data.columns) #all observed data in current procedure
-    B=indexs.copy() #remain variables
-    ClusterLength=2
+    """
+    LearnedClusters = []
+    PClusters = []
+    indexs = list(data.columns)  # all observed data in current procedure
+    B = indexs.copy()  # remain variables
+    ClusterLength = 2
     for S in itertools.combinations(list(B), ClusterLength):
         if TetradMethod.CheckCausalCluster(list(S), data, alpha):
             LearnedClusters.append(S)
             PClusters.append(S)
-            B=set(B)-set(S)
-
-    #overlap merge, updata the causal cluster and add into PureCluster
-    #only recall the overlap merge function and check whether the cluster with more than three elements
-    LearnedClusters=Overlap_Merge.merge_list(LearnedClusters)
+            B = set(B) - set(S)
 
+    # overlap merge, updata the causal cluster and add into PureCluster
+    # only recall the overlap merge function and check whether the cluster with more than three elements
+    LearnedClusters = Overlap_Merge.merge_list(LearnedClusters)
 
     for S in LearnedClusters:
-        if len(S) > 2: #Overlap merged cluster add into Purecluster
+        if len(S) > 2:  # Overlap merged cluster add into Purecluster
             PureCluster.append(S)
-        else: # run the identifying pure cluster function
-            Pure_flag=TetradMethod.JudgePureCluster(S, data, alpha)
+        else:  # run the identifying pure cluster function
+            Pure_flag = TetradMethod.JudgePureCluster(S, data, alpha)
             if Pure_flag:
                 PureCluster.append(S)
             else:
                 ImpureCluster.append(S)
 
-    ClusterLength +=1
+    ClusterLength += 1
     while len(B) >= ClusterLength and len(indexs) > (ClusterLength + 2):
         for S in itertools.combinations(list(B), ClusterLength):
-            S=list(S)
+            S = list(S)
             if TetradMethod.CheckCausalCluster(list(S), data, alpha):
                 LearnedClusters.append(S)
                 ImpureCluster.append(S)
-                B=set(B)-set(S)
-        ClusterLength +=1
+                B = set(B) - set(S)
+        ClusterLength += 1
 
     return LearnedClusters, PureCluster, ImpureCluster, PClusters
 
 
-
-
-
-
-
-
-
 def main():
     pass
 
+
 if __name__ == '__main__':
     main()

cdmir/discovery/funtional_based/LearningHierarchicalStructure/GIN2.py

@@ -1,53 +1,56 @@
 import numpy as np
 import pandas as pd
-import cdmir.discovery.funtional_based.LearningHierarchicalStructure.indTest.HSIC2 as hsic
+import lingam.hsic as hsic
 
 import cdmir.discovery.funtional_based.LearningHierarchicalStructure.indTest.independence as ID
 
 
-#GIN by fast HSIC
-#X=['X1','X2']
-#Z=['X3']
-#data.type=Pandas.DataFrame
-def GIN(X,Z,data,alpha=0.05):
-    omega = getomega(data,X,Z)
-    tdata= data[X]
-    #print(tdata.T)
+# GIN by fast HSIC
+# X=['X1','X2']
+# Z=['X3']
+# data.type=Pandas.DataFrame
+def GIN(X, Z, data, alpha=0.05):
+    omega = getomega(data, X, Z)
+    tdata = data[X]
     result = np.dot(omega, tdata.T)
     for i in Z:
         temp = np.array(data[i])
-        flag =hsic.test(result.T,temp,alpha)
 
-        if not flag:#not false == ture  ---> if false
-            #print(X,Z,flag)
+        pvalue = hsic.hsic_test_gamma(result.T, temp, alpha)
+
+        if pvalue > alpha:
+            flag = True
+        else:
+            flag = False
+
+        if not flag:
             return False
 
     return True
 
 
-#mthod 1: estimating mutual information by k nearest neighbors (density estimation)
-#mthod 2: estimating mutual information by sklearn package
-def GIN_MI(X,Z,data,method='1'):
-    omega = getomega(data,X,Z)
-    tdata= data[X]
+# mthod 1: estimating mutual information by k nearest neighbors (density estimation)
+# mthod 2: estimating mutual information by sklearn package
+def GIN_MI(X, Z, data, method='1'):
+    omega = getomega(data, X, Z)
+    tdata = data[X]
     result = np.dot(omega, tdata.T)
-    MIS=0
+    MIS = 0
     for i in Z:
 
         temp = np.array(data[i])
-        if method =='1':
-            mi=ID.independent(result.T,temp)
+        if method == '1':
+            mi = ID.independent(result.T, temp)
         else:
-            mi=ID.independent11(result.T,temp)
-        MIS+=mi
-    MIS = MIS/len(Z)
+            mi = ID.independent11(result.T, temp)
+        MIS += mi
+    MIS = MIS / len(Z)
 
     return MIS
 
 
-
-def getomega(data,X,Z):
-    cov_m =np.cov(data,rowvar=False)
+def getomega(data, X, Z):
+    cov_m = np.cov(data, rowvar=False)
     col = list(data.columns)
     Xlist = []
     Zlist = []
@@ -58,13 +61,12 @@ def getomega(data,X,Z):
         t = col.index(i)
         Zlist.append(t)
     B = cov_m[Xlist]
-    B = B[:,Zlist]
+    B = B[:, Zlist]
     A = B.T
-    u,s,v = np.linalg.svd(A)
+    u, s, v = np.linalg.svd(A)
     lens = len(X)
-    omega =v.T[:,lens-1]
-    omegalen=len(omega)
-    omega=omega.reshape(1,omegalen)
+    omega = v.T[:, lens - 1]
+    omegalen = len(omega)
+    omega = omega.reshape(1, omegalen)
 
     return omega
-