pyAgrum-nightly 2.3.1.9.dev202512261765915415__cp310-abi3-macosx_10_15_x86_64.whl

pyagrum/lib/bn_vs_bn.py

@@ -0,0 +1,605 @@
+############################################################################
+#   This file is part of the aGrUM/pyAgrum library.                        #
+#                                                                          #
+#   Copyright (c) 2005-2025 by                                             #
+#       - Pierre-Henri WUILLEMIN(_at_LIP6)                                 #
+#       - Christophe GONZALES(_at_AMU)                                     #
+#                                                                          #
+#   The aGrUM/pyAgrum library is free software; you can redistribute it    #
+#   and/or modify it under the terms of either :                           #
+#                                                                          #
+#    - the GNU Lesser General Public License as published by               #
+#      the Free Software Foundation, either version 3 of the License,      #
+#      or (at your option) any later version,                              #
+#    - the MIT license (MIT),                                              #
+#    - or both in dual license, as here.                                   #
+#                                                                          #
+#   (see https://agrum.gitlab.io/articles/dual-licenses-lgplv3mit.html)    #
+#                                                                          #
+#   This aGrUM/pyAgrum library is distributed in the hope that it will be  #
+#   useful, but WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,          #
+#   INCLUDING BUT NOT LIMITED TO THE WARRANTIES MERCHANTABILITY or FITNESS #
+#   FOR A PARTICULAR PURPOSE  AND NONINFRINGEMENT. IN NO EVENT SHALL THE   #
+#   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER #
+#   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,        #
+#   ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR  #
+#   OTHER DEALINGS IN THE SOFTWARE.                                        #
+#                                                                          #
+#   See LICENCES for more details.                                         #
+#                                                                          #
+#   SPDX-FileCopyrightText: Copyright 2005-2025                            #
+#       - Pierre-Henri WUILLEMIN(_at_LIP6)                                 #
+#       - Christophe GONZALES(_at_AMU)                                     #
+#   SPDX-License-Identifier: LGPL-3.0-or-later OR MIT                      #
+#                                                                          #
+#   Contact  : info_at_agrum_dot_org                                       #
+#   homepage : http://agrum.gitlab.io                                      #
+#   gitlab   : https://gitlab.com/agrumery/agrum                           #
+#                                                                          #
+############################################################################
+
+"""
+The purpose of this module is to provide tools for comaring different BNs.
+"""
+
+import os
+import math
+from itertools import product, combinations
+
+import pydot as dot
+
+import pyagrum as gum
+import pyagrum.lib.bn2graph as ggr
+import pyagrum.lib.utils as gutils
+
+
+STRUCTURAL_HAMMING = "structural hamming"
+PURE_HAMMING = "hamming"
+
+
+class GraphicalBNComparator:
+  """
+  BNGraphicalComparator allows to compare in multiple way 2 BNs...
+  The smallest assumption is that the names of the variables are the same in
+  the 2 BNs. But some comparisons will have also to check the type and
+  domainSize of the variables.
+
+  The bns have not exactly the  same role : _bn1 is rather the referent model
+  for the comparison whereas _bn2 is the compared one to the referent model.
+
+  Parameters
+  ----------
+  bn1 : str or pyagrum.BayesNet
+    a BN or a filename for reference
+  bn2 : str or pyagrum.BayesNet
+    another BN or antoher filename for comparison
+  """
+
+  def __init__(self, bn1: str | gum.BayesNet, bn2: str | gum.BayesNet, delta=1e-6):
+    self.DELTA_ERROR = delta
+    if isinstance(bn1, str):
+      self._bn1 = gum.loadBN(bn1)
+      self._bn1.setProperty("name", '"' + os.path.basename(self._bn1.property("name") + '"'))
+    else:
+      self._bn1 = bn1
+
+    if isinstance(bn2, str):
+      self._bn2 = gum.loadBN(bn2)
+      self._bn2.setProperty("name", '"' + os.path.basename(self._bn2.property("name") + '"'))
+    else:
+      self._bn2 = bn2
+
+    s1: set[str] = set(self._bn1.names())
+    s2: set[str] = set(self._bn2.names())
+
+    if s1 != s2:
+      raise ValueError(
+        "The 2 BNs are not comparable! There are names not present in the 2 BNs : " + str(s1.symmetric_difference(s2))
+      )
+
+  def _compareBNVariables(self):
+    """
+    Checks if the two BNs have the same set of variables
+
+    Returns
+    -------
+    str
+      'OK' if the BNs have composed of the same variables, indicates problematic variables otherwise
+
+    """
+    # it is assumed (checked by the constructor) that _bn1 and _bn2 share the same set of variable names
+    for i in self._bn1.nodes():
+      v1 = self._bn1.variable(i)
+      v2 = self._bn2.variableFromName(v1.name())
+      if v2.domainSize() != v1.domainSize():
+        return v1.name() + " has not the same domain size in the two bns"
+
+    return "OK"
+
+  @staticmethod
+  def _parents_name(bn, n):
+    return {bn.variable(p).name() for p in bn.parents(n)}
+
+  def _compareBNParents(self):
+    """
+    Returns
+    -------
+    str
+      'OK' if _bn2 have (at least) the same variable as b1 and their parents are the same.
+
+    """
+    for id1 in self._bn1.nodes():
+      id2 = self._bn2.idFromName(self._bn1.variable(id1).name())
+
+      p1 = self._parents_name(self._bn1, id1)
+      p2 = self._parents_name(self._bn2, id2)
+      if p1 != p2:
+        return (
+          self._bn1.variable(id1).name()
+          + " has different parents in the two bns whose names are in "
+          + str(p1.symmetric_difference(p2))
+        )
+
+    return "OK"
+
+  def _compareTensors(self, pot1, pot2):
+    """
+    Compare 2 tensors one in each Bayesian network
+
+    Parameters
+    ----------
+    pot1 : pyagrum.Tensor
+      one of b1's cpts
+    pot2 : pyagrum.Tensor
+      one of _bn2's cpts
+
+    Returns
+    -------
+    str
+      'OK' if CPTs are the same
+
+    Raises
+    ------
+    gum.KeyError
+      If cpts are not from the same variable
+    """
+    I1 = gum.Instantiation(pot1)
+    I2 = gum.Instantiation(pot2)
+    I1.setFirst()
+    while not I1.end():
+      I2.fromdict(I1.todict())  # copy value on the base of names
+      if abs(pot1.get(I1) - pot2.get(I2)) > self.DELTA_ERROR:
+        return "Different CPTs for " + pot1.variable(0).name()
+      I1 += 1
+    return "OK"
+
+  def _compareBNCPT(self):
+    """
+    Returns
+    -------
+    str
+      'OK' if _bn2 have (at least) the same variable as b1 and their cpts are the same
+    """
+    for i in self._bn1.nodes():
+      res = self._compareTensors(self._bn1.cpt(i), self._bn2.cpt(self._bn1.variable(i).name()))
+      if res != "OK":
+        return res
+    return "OK"
+
+  def equivalentBNs(self):
+    """
+    Check if the 2 BNs are equivalent :
+
+    * same variables
+    * same graphical structure
+    * same parameters
+
+    Returns
+    -------
+    str
+      "OK" if bn are the same, a description of the error otherwise
+
+    """
+
+    ret = self._compareBNVariables()
+    if ret != "OK":
+      return ret
+
+    ret = self._compareBNParents()
+    if ret != "OK":
+      return ret
+
+    ret = self._compareBNCPT()
+    return ret
+
+  def dotDiff(self):
+    """Return a pydot graph that compares the arcs of _bn1 (reference) with those of self._bn2.
+    full black line: the arc is common for both
+    full red line: the arc is common but inverted in _bn2
+    dotted black line: the arc is added in _bn2
+    dotted red line: the arc is removed in _bn2
+
+    Warning
+    -------
+    if pydot is not installed, this function just returns None
+
+    Returns
+    -------
+    pydot.Dot
+      the result dot graph or None if pydot can not be imported
+    """
+    return graphDiff(self._bn1, self._bn2)
+
+  def skeletonScores(self):
+    """
+    Compute Precision, Recall, F-score for skeletons of self._bn2 compared to self._bn1
+
+    precision and recall are computed considering BN1 as the reference
+
+    Fscor is 2*(recall* precision)/(recall+precision) and is the weighted average of Precision and Recall.
+
+    dist2opt=square root of (1-precision)^2+(1-recall)^2 and represents the euclidian distance to the ideal point (precision=1, recall=1)
+
+    Returns
+    -------
+    dict[str,double]
+      A dictionnary containing 'precision', 'recall', 'fscore', 'dist2opt' and so on.
+    """
+    # t: True, f: False, p: Positive, n: Negative
+    count = {"tp": 0, "tn": 0, "fp": 0, "fn": 0}
+
+    # We look at all combination
+    listVariables = self._bn1.names()
+
+    # Loop on pairs of variables
+    for head, tail in combinations(listVariables, 2):
+      idHead_1 = self._bn1.idFromName(head)
+      idTail_1 = self._bn1.idFromName(tail)
+
+      idHead_2 = self._bn2.idFromName(head)
+      idTail_2 = self._bn2.idFromName(tail)
+
+      if self._bn1.dag().existsArc(idHead_1, idTail_1) or self._bn1.dag().existsArc(
+        idTail_1, idHead_1
+      ):  # Check edge node1-node2
+        if self._bn2.dag().existsArc(idHead_2, idTail_2) or self._bn2.dag().existsArc(idTail_2, idHead_2):  # if edge:
+          count["tp"] += 1
+        else:  # If no edge:
+          count["fn"] += 1
+      else:  # Check if no edge
+        if self._bn2.dag().existsArc(idHead_2, idTail_2) or self._bn2.dag().existsArc(idTail_2, idHead_2):  # If edge
+          count["fp"] += 1
+        else:  # If no arc
+          count["tn"] += 1
+
+    # Compute the scores
+    if count["tp"] + count["fn"] != 0:
+      recall = (1.0 * count["tp"]) / (count["tp"] + count["fn"])
+    else:
+      recall = 0.0
+
+    if count["tp"] + count["fp"] != 0:
+      precision = (1.0 * count["tp"]) / (count["tp"] + count["fp"])
+    else:
+      precision = 0.0
+
+    if precision + recall != 0.0:
+      Fscore = (2 * recall * precision) / (recall + precision)
+    else:
+      Fscore = 0.0
+
+    return {
+      "count": count,
+      "recall": recall,
+      "precision": precision,
+      "fscore": Fscore,
+      "dist2opt": math.sqrt((1 - precision) ** 2 + (1 - recall) ** 2),
+    }
+
+  def scores(self):
+    """
+    Compute Precision, Recall, F-score for self._bn2 compared to self._bn1
+
+    precision and recall are computed considering BN1 as the reference
+
+    Fscore is 2*(recall* precision)/(recall+precision) and is the weighted average of Precision and Recall.
+
+    dist2opt=square root of (1-precision)^2+(1-recall)^2 and represents the euclidian distance to the ideal point (precision=1, recall=1)
+
+    Returns
+    -------
+    dict[str,double]
+      A dictionnary containing 'precision', 'recall', 'fscore', 'dist2opt' and so on.
+    """
+    # t: True, f: False, p: Positive, n: Negative
+    count = {"tp": 0, "tn": 0, "fp": 0, "fn": 0}
+
+    # We look at all combination
+    listVariables = self._bn1.names()
+
+    # Loop on oriented pairs of variables
+    for head, tail in product(listVariables, listVariables):
+      if head != tail:
+        idHead_1 = self._bn1.idFromName(head)
+        idTail_1 = self._bn1.idFromName(tail)
+
+        idHead_2 = self._bn2.idFromName(head)
+        idTail_2 = self._bn2.idFromName(tail)
+
+        if self._bn1.dag().existsArc(idHead_1, idTail_1):  # Check arcs head->tail
+          if self._bn2.dag().existsArc(idHead_2, idTail_2):  # if arc:
+            count["tp"] += 1
+          else:  # If no arc:
+            count["fn"] += 1
+        else:  # Check if no arc
+          if self._bn2.dag().existsArc(idHead_2, idTail_2):  # If arc
+            count["fp"] += 1
+          else:  # If no arc
+            count["tn"] += 1
+
+    # Compute the scores
+    if count["tp"] + count["fn"] != 0:
+      recall = (1.0 * count["tp"]) / (count["tp"] + count["fn"])
+    else:
+      recall = 0.0
+
+    if count["tp"] + count["fp"] != 0:
+      precision = (1.0 * count["tp"]) / (count["tp"] + count["fp"])
+    else:
+      precision = 0.0
+
+    if precision + recall != 0.0:
+      Fscore = (2 * recall * precision) / (recall + precision)
+    else:
+      Fscore = 0.0
+
+    return {
+      "count": count,
+      "recall": recall,
+      "precision": precision,
+      "fscore": Fscore,
+      "dist2opt": math.sqrt((1 - precision) ** 2 + (1 - recall) ** 2),
+    }
+
+  def hamming(self):
+    """
+    Compute hamming and structural hamming distance
+
+    Hamming distance is the difference of edges comparing the 2 skeletons, and Structural Hamming difference is the
+    difference comparing the cpdags,	including the arcs' orientation.
+
+    Returns
+    -------
+    dict[double,double]
+      A dictionary containing PURE_HAMMING,STRUCTURAL_HAMMING
+    """
+    # convert graphs to cpdags
+    cpdag1 = gum.EssentialGraph(self._bn1).pdag()
+    cpdag2 = gum.EssentialGraph(self._bn2).pdag()
+
+    # We look at all combinations
+    listVariables = self._bn1.names()
+    hamming_dico = {PURE_HAMMING: 0, STRUCTURAL_HAMMING: 0}
+
+    for head, tail in combinations(listVariables, 2):
+      idHead_1 = self._bn1.idFromName(head)
+      idTail_1 = self._bn1.idFromName(tail)
+
+      idHead_2 = self._bn2.idFromName(head)
+      idTail_2 = self._bn2.idFromName(tail)
+
+      if cpdag1.existsArc(idHead_1, idTail_1):  # Check arcs head->tail
+        if cpdag2.existsArc(idTail_2, idHead_2) or cpdag2.existsEdge(idTail_2, idHead_2):
+          hamming_dico[STRUCTURAL_HAMMING] += 1
+        elif (
+          not cpdag2.existsArc(idTail_2, idHead_2)
+          and not cpdag2.existsArc(idHead_2, idTail_2)
+          and not cpdag2.existsEdge(idTail_2, idHead_2)
+        ):
+          hamming_dico[STRUCTURAL_HAMMING] += 1
+          hamming_dico[PURE_HAMMING] += 1
+
+      elif cpdag1.existsArc(idTail_1, idHead_1):  # Check arcs tail->head
+        if cpdag2.existsArc(idHead_2, idTail_2) or cpdag2.existsEdge(idTail_2, idHead_2):
+          hamming_dico[STRUCTURAL_HAMMING] += 1
+        elif (
+          not cpdag2.existsArc(idTail_2, idHead_2)
+          and not cpdag2.existsArc(idHead_2, idTail_2)
+          and not cpdag2.existsEdge(idTail_2, idHead_2)
+        ):
+          hamming_dico[STRUCTURAL_HAMMING] += 1
+          hamming_dico[PURE_HAMMING] += 1
+
+      elif cpdag1.existsEdge(idTail_1, idHead_1):  # Check edge
+        if cpdag2.existsArc(idHead_2, idTail_2) or cpdag2.existsArc(idTail_2, idHead_2):
+          hamming_dico[STRUCTURAL_HAMMING] += 1
+        elif (
+          not cpdag2.existsArc(idTail_2, idHead_2)
+          and not cpdag2.existsArc(idHead_2, idTail_2)
+          and not cpdag2.existsEdge(idTail_2, idHead_2)
+        ):
+          hamming_dico[STRUCTURAL_HAMMING] += 1
+          hamming_dico[PURE_HAMMING] += 1
+          # check no edge or arc on the ref graph, and yes on the other graph
+
+      elif (
+        cpdag2.existsArc(idHead_2, idTail_2)
+        or cpdag2.existsEdge(idHead_2, idTail_2)
+        or cpdag2.existsArc(idTail_2, idHead_2)
+      ):
+        hamming_dico[STRUCTURAL_HAMMING] += 1
+        hamming_dico[PURE_HAMMING] += 1
+
+    return hamming_dico
+
+
+def graphDiff(bnref, bncmp, noStyle=False):
+  """Return a pydot graph that compares the arcs of bnref to bncmp.
+  graphDiff allows bncmp to have less nodes than bnref. (this is not the case in GraphicalBNComparator.dotDiff())
+
+  if noStyle is False use 4 styles (fixed in pyagrum.config) :
+    - the arc is common for both
+    - the arc is common but inverted in _bn2
+    - the arc is added in _bn2
+    - the arc is removed in _bn2
+
+  See graphDiffLegend() to add a legend to the graph.
+  Warning
+  -------
+  if pydot is not installed, this function just returns None
+
+  Returns
+  -------
+  pydot.Dot
+    the result dot graph or None if pydot can not be imported
+  """
+  g = ggr.BN2dot(bnref)
+  positions = gutils.dot_layout(g)
+
+  res = dot.Dot(graph_type="digraph", bgcolor="transparent", layout="fdp", splines=True)
+  for i1 in bnref.nodes():
+    if bnref.variable(i1).name() in bncmp.names():
+      res.add_node(
+        dot.Node(
+          f'"{bnref.variable(i1).name()}"',
+          style="filled",
+          fillcolor=gum.config["notebook", "graphdiff_correct_color"],
+          color=gutils.getBlackInTheme(),
+        )
+      )
+    else:
+      if not noStyle:
+        res.add_node(
+          dot.Node(
+            f'"{bnref.variable(i1).name()}"',
+            style="dashed",
+            fillcolor=gum.config["notebook", "graphdiff_correct_color"],
+            color=gutils.getBlackInTheme(),
+          )
+        )
+  if noStyle:
+    for i1, i2 in bncmp.arcs():
+      n1 = bncmp.variable(i1).name()
+      n2 = bncmp.variable(i2).name()
+      res.add_edge(
+        dot.Edge(
+          f'"{n1}"',
+          f'"{n2}"',
+          style=gum.config["notebook", "graphdiff_correct_style"],
+          color=gum.config["notebook", "graphdiff_correct_color"],
+        )
+      )
+
+  else:
+    for i1, i2 in bnref.arcs():
+      n1 = bnref.variable(i1).name()
+      n2 = bnref.variable(i2).name()
+
+      # a node is missing
+      if not (n1 in bncmp.names() and n2 in bncmp.names()):
+        res.add_edge(
+          dot.Edge(
+            f'"{n1}"',
+            f'"{n2}"',
+            style=gum.config["notebook", "graphdiff_missing_style"],
+            color=gum.config["notebook", "graphdiff_missing_color"],
+          )
+        )
+        continue
+
+      if bncmp.existsArc(n1, n2):  # arc is OK in BN2
+        res.add_edge(
+          dot.Edge(
+            f'"{n1}"',
+            f'"{n2}"',
+            style=gum.config["notebook", "graphdiff_correct_style"],
+            color=gum.config["notebook", "graphdiff_correct_color"],
+          )
+        )
+      elif bncmp.existsArc(n2, n1):  # arc is reversed in BN2
+        res.add_edge(dot.Edge(f'"{n1}"', f'"{n2}"', style="invis"))
+        res.add_edge(
+          dot.Edge(
+            f'"{n2}"',
+            f'"{n1}"',
+            style=gum.config["notebook", "graphdiff_reversed_style"],
+            color=gum.config["notebook", "graphdiff_reversed_color"],
+          )
+        )
+      else:  # arc is missing in BN2
+        res.add_edge(
+          dot.Edge(
+            f'"{n1}"',
+            f'"{n2}"',
+            style=gum.config["notebook", "graphdiff_missing_style"],
+            color=gum.config["notebook", "graphdiff_missing_color"],
+          )
+        )
+
+    for i1, i2 in bncmp.arcs():
+      n1 = bncmp.variable(i1).name()
+      n2 = bncmp.variable(i2).name()
+      if not bnref.existsArc(n1, n2) and not bnref.existsArc(n2, n1):  # arc only in BN2
+        res.add_edge(
+          dot.Edge(
+            f'"{n1}"',
+            f'"{n2}"',
+            style=gum.config["notebook", "graphdiff_overflow_style"],
+            color=gum.config["notebook", "graphdiff_overflow_color"],
+          )
+        )
+
+    gutils.apply_dot_layout(res, positions)
+
+  return res
+
+
+def graphDiffLegend():
+  try:
+    # pydot is optional
+    # pylint: disable=import-outside-toplevel
+    import pydot as dot
+  except ImportError:
+    return None
+
+  res = dot.Dot(graph_type="digraph", bgcolor="transparent", rankdir="LR")
+  for i in "abcdefgh":
+    res.add_node(dot.Node(i, style="invis"))
+  res.add_edge(
+    dot.Edge(
+      "a",
+      "b",
+      label="overflow",
+      style=gum.config["notebook", "graphdiff_overflow_style"],
+      color=gum.config["notebook", "graphdiff_overflow_color"],
+    )
+  )
+  res.add_edge(
+    dot.Edge(
+      "c",
+      "d",
+      label="Missing",
+      style=gum.config["notebook", "graphdiff_missing_style"],
+      color=gum.config["notebook", "graphdiff_missing_color"],
+    )
+  )
+  res.add_edge(
+    dot.Edge(
+      "e",
+      "f",
+      label="reversed",
+      style=gum.config["notebook", "graphdiff_reversed_style"],
+      color=gum.config["notebook", "graphdiff_reversed_color"],
+    )
+  )
+  res.add_edge(
+    dot.Edge(
+      "g",
+      "h",
+      label="Correct",
+      style=gum.config["notebook", "graphdiff_correct_style"],
+      color=gum.config["notebook", "graphdiff_correct_color"],
+    )
+  )
+
+  return res