From 3868bb1c48d1eb243a2a113ae5e4a4575b2895de Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Petr=C3=B4nio=20C=C3=A2ndido?= <petronio.candido@gmail.com>
Date: Fri, 22 Mar 2019 15:07:44 -0300
Subject: [PATCH] Interval forecasting on MVFTS and WMVFTS

---
 pyFTS/models/hofts.py               | 12 --------
 pyFTS/models/multivariate/mvfts.py  | 48 +++++++++++++++++++++++++++--
 pyFTS/models/multivariate/wmvfts.py | 19 ++++++++++--
 3 files changed, 63 insertions(+), 16 deletions(-)

diff --git a/pyFTS/models/hofts.py b/pyFTS/models/hofts.py
index 40d21c2..9e395a9 100644
--- a/pyFTS/models/hofts.py
+++ b/pyFTS/models/hofts.py
@@ -73,24 +73,12 @@ class WeightedHighOrderFLRG(flrg.FLRG):
         return self.midpoint
 
     def get_lower(self, sets):
-        """
-        Returns the lower bound value for the RHS fuzzy sets
-
-        :param sets: fuzzy sets
-        :return: lower bound value
-        """
         if self.lower is None:
             lw = np.array([sets[s].lower for s in self.RHS.keys()])
             self.lower = lw.dot(self.weights())
         return self.lower
 
     def get_upper(self, sets):
-        """
-        Returns the upper bound value for the RHS fuzzy sets
-
-        :param sets: fuzzy sets
-        :return: upper bound value
-        """
         if self.upper is None:
             up = np.array([sets[s].upper for s in self.RHS.keys()])
             self.upper = up.dot(self.weights())
diff --git a/pyFTS/models/multivariate/mvfts.py b/pyFTS/models/multivariate/mvfts.py
index b94adab..5121f63 100644
--- a/pyFTS/models/multivariate/mvfts.py
+++ b/pyFTS/models/multivariate/mvfts.py
@@ -145,8 +145,9 @@ class MVFTS(fts.FTS):
                         mvs.append(0.)
                         mps.append(0.)
                 else:
-                    mvs.append(self.flrgs[flrg.get_key()].get_membership(data_point, self.explanatory_variables))
-                    mps.append(self.flrgs[flrg.get_key()].get_midpoint(self.target_variable.partitioner.sets))
+                    _flrg = self.flrgs[flrg.get_key()]
+                    mvs.append(_flrg.get_membership(data_point, self.explanatory_variables))
+                    mps.append(_flrg.get_midpoint(self.target_variable.partitioner.sets))
 
             mv = np.array(mvs)
             mp = np.array(mps)
@@ -201,6 +202,49 @@ class MVFTS(fts.FTS):
 
         return ret
 
+    def forecast_interval(self, data, **kwargs):
+        ret = []
+        ndata = self.apply_transformations(data)
+        c = 0
+        for index, row in ndata.iterrows() if isinstance(ndata, pd.DataFrame) else enumerate(ndata):
+            data_point = self.format_data(row)
+            flrs = self.generate_lhs_flrs(data_point)
+            mvs = []
+            ups = []
+            los = []
+            for flr in flrs:
+                flrg = mvflrg.FLRG(lhs=flr.LHS)
+                if flrg.get_key() not in self.flrgs:
+                    #Naïve approach is applied when no rules were found
+                    if self.target_variable.name in flrg.LHS:
+                        fs = flrg.LHS[self.target_variable.name]
+                        fset = self.target_variable.partitioner.sets[fs]
+                        up = fset.upper
+                        lo = fset.lower
+                        mv = fset.membership(data_point[self.target_variable.name])
+                        mvs.append(mv)
+                        ups.append(up)
+                        los.append(lo)
+                    else:
+                        mvs.append(0.)
+                        ups.append(0.)
+                        los.append(0.)
+                else:
+                    _flrg = self.flrgs[flrg.get_key()]
+                    mvs.append(_flrg.get_membership(data_point, self.explanatory_variables))
+                    ups.append(_flrg.get_upper(self.target_variable.partitioner.sets))
+                    los.append(_flrg.get_lower(self.target_variable.partitioner.sets))
+
+            mv = np.array(mvs)
+            up = np.dot(mv, np.array(ups).T) / np.sum(mv)
+            lo = np.dot(mv, np.array(los).T) / np.sum(mv)
+
+            ret.append([lo, up])
+
+        ret = self.target_variable.apply_inverse_transformations(ret,
+                                                           params=data[self.target_variable.data_label].values)
+        return ret
+
 
     def clone_parameters(self, model):
         super(MVFTS, self).clone_parameters(model)
diff --git a/pyFTS/models/multivariate/wmvfts.py b/pyFTS/models/multivariate/wmvfts.py
index 156fc5d..a969ef7 100644
--- a/pyFTS/models/multivariate/wmvfts.py
+++ b/pyFTS/models/multivariate/wmvfts.py
@@ -33,8 +33,23 @@ class WeightedFLRG(mvflrg.FLRG):
         return self.w
 
     def get_midpoint(self, sets):
-        mp = np.array([sets[c].centroid for c in self.RHS.keys()])
-        return mp.dot(self.weights())
+        if self.midpoint is None:
+            mp = np.array([sets[c].centroid for c in self.RHS.keys()])
+            self.midpoint = mp.dot(self.weights())
+
+        return self.midpoint
+
+    def get_lower(self, sets):
+        if self.lower is None:
+            lw = np.array([sets[s].lower for s in self.RHS.keys()])
+            self.lower = lw.dot(self.weights())
+        return self.lower
+
+    def get_upper(self, sets):
+        if self.upper is None:
+            up = np.array([sets[s].upper for s in self.RHS.keys()])
+            self.upper = up.dot(self.weights())
+        return self.upper
 
 
     def __str__(self):