generation of the BID

mar.net

@@ -41,13 +41,13 @@ node Ic {
 }
 
 potential (A) {
-   data = (  0.561407 0.438593);
+   data = (  0.6 0.4);
 }
 
 potential ( B | A   ) {
    data = 
-   ((   0.515132   0.484868)   % A=0   
-   (   0.371633   0.628367));   % A=1   
+   ((   0.5        0.5)   % A=0   
+   (    0.2        0.8));      % A=1   
 }
 
 potential ( C | B  ) {

mcc.py

@@ -4,6 +4,13 @@ import psycopg2
 import sys
 import argparse
 
+def is_var_indicator(v):
+    return not is_var(v)
+
+def is_var(v):
+    return not v.lower().startswith('i')
+
+
 parser = argparse.ArgumentParser()
 parser.add_argument("BNFile", help="Bayesian network file")
 parser.add_argument("DBSize", help="size of the database")
@@ -38,7 +45,6 @@ cursor.execute(deletion_sql)
 # creating the table to store the 
 col_def = map(lambda name : "{} int NOT NULL".format(name), var_names)
 creation_sql = "CREATE TABLE {}({});".format(table_name, ",".join(col_def))
-print(creation_sql)
 cursor.execute(creation_sql) 
 
 # loading the CSV data to the table
@@ -48,30 +54,84 @@ with open(out_file, "r") as file:
 
 missing_selects = []
 for v in var_names:
-    if not v.lower().startswith('i') and v[1:] not in var_names:
+    if is_var(v) and v[1:] not in var_names:
         missing_selects.append("case when i{}=1 then NULL ELSE {} END".format(v,v))
-    elif not v.lower().startswith('i'):
+    elif is_var(v):
         missing_selects.append(v)
 
 missing_data_table = "{}_star".format(table_name)
 deletion_sql = "DROP TABLE IF EXISTS {};".format(missing_data_table)
 cursor.execute(deletion_sql)
-missing_data_sql = "CREATE TABLE {} AS SELECT {} FROM {}".format(missing_data_table, ",".join(missing_selects), table_name)
+missing_selects_text = ",".join(missing_selects)
+missing_data_sql = "CREATE TABLE {} AS SELECT {}, md5(CAST(({}) AS text)) AS superblock FROM {}".format(missing_data_table, missing_selects_text, missing_selects_text, table_name)
 print(missing_data_sql)
 cursor.execute(missing_data_sql)
 
+# superblock definitions
+superblock_table = "{}_sb".format(table_name)
+deletion_sql = "DROP TABLE IF EXISTS {};".format(superblock_table)
+cursor.execute(deletion_sql)
+
+vars = list(filter(is_var, var_names))
+col_def = map(lambda name : "{} int NOT NULL".format(name), vars)
+creation_sql = "CREATE TABLE {}(superblock text, {}, prob float NOT NULL);".format(superblock_table, ",".join(col_def))
+cursor.execute(creation_sql) 
 
 
+cursor.execute("SELECT DISTINCT * FROM {}".format(missing_data_table))
+ie = gum.LazyPropagation(bn)
+
+insert_sql = "INSERT INTO {}({}, superblock, prob) VALUES({},%s,%s)".format(superblock_table, ",".join(vars), ",".join(["%s" for _ in vars]))
+
+def instantiation_to_list(inst, row, vars):
+    l = []
+    i = 0
+    inst_vars = list(map(lambda v : v.name(), inst.variablesSequence()))
+    for v in vars:
+        if v in inst_vars:
+            l.append(inst[v])
+        else:
+            l.append(row[i])
+        i+=1
+    return l
+
+for row in cursor.fetchall():
+    null_vars = []
+    null_pos  = []
+    i = 0
+    for v in vars:
+        if row[i] is None:
+            null_vars.append(v)
+            null_pos.append(i)
+            ie.addTarget(v)
+        else:
+            ie.addEvidence(v, row[i])
+        i+=1
+    if len(null_vars) > 0:
+        potentiel = ie.jointPosterior(set(null_vars)) if len(vars) > len(null_vars) else ie.evidenceJointImpact(vars,{})
+        for i in potentiel.loopIn():
+            inserted_row = instantiation_to_list(i, row, vars)
+            inserted_row.append(row[-1])
+            inserted_row.append(potentiel.get(i))
+            cursor.execute(insert_sql, inserted_row)
+    ie.eraseAllTargets()
+    ie.eraseAllEvidence()
+
+
+
+
+
+    
 ###### Imputation
 
-import numpy as np
-import pandas as pd
-from sklearn.impute import KNNImputer
+# import numpy as np
+# import pandas as pd
+# from sklearn.impute import KNNImputer
 
-cursor.execute("SELECT * FROM {}".format(missing_data_table))
-tuples_list = cursor.fetchall()
-df = pd.DataFrame(tuples_list)
+# cursor.execute("SELECT * FROM {}".format(missing_data_table))
+# tuples_list = cursor.fetchall()
+# df = pd.DataFrame(tuples_list)
 
-imputer = KNNImputer(n_neighbors=2)
-impute = imputer.fit_transform(df)
-print(impute)
+# imputer = KNNImputer(n_neighbors=10)
+# impute = imputer.fit_transform(df)
+# print(impute)

mnar.net

+
+net {
+  name = mnar;
+  software = "aGrUM 1.17.1";
+  node_size = (50 50);
+}
+
+node B {
+   states = (0 1 );
+   label = "B";
+   ID = "B";
+}
+
+node A {
+   states = (0 1 );
+   label = "A";
+   ID = "A";
+}
+
+node Ia {
+   states = (0 1 );
+   label = "Ia";
+   ID = "Ia";
+}
+
+node Ib {
+   states = (0 1 );
+   label = "Ib";
+   ID = "Ib";
+}
+
+potential ( B | A   ) {
+   data = 
+   ((   0.515132   0.484868)   % A=0   
+   (   0.371633   0.628367));   % A=1   
+}
+
+potential (A) {
+   data = (  0.561407 0.438593);
+}
+
+potential (Ib) {
+   data = (  0.8 0.2);
+}
+
+potential ( Ia | B   ) {
+   data = 
+   ((   0.664707   0.335293)   % B=0   
+   (   0.344864   0.655136));   % B=1   
+}
+
+
+