Implementation of forecast_ahead in TimeVariant and IncrementalEnsemble methods

pyFTS/models/incremental/IncrementalEnsemble.py

@@ -63,6 +63,15 @@ class IncrementalEnsembleFTS(ensemble.EnsembleFTS):
 
     def forecast(self, data, **kwargs):
         l = len(data)
+        no_update = kwargs.get('no_update', False)
+        if no_update:
+            ret = []
+            for k in np.arange(self.order, l+1):
+                sample = data[k-self.order: k]
+                tmp = self.get_models_forecasts(sample)
+                point = self.get_point(tmp)
+                ret.append(point)
+            return ret
 
         data_window = []
 
@@ -88,6 +97,27 @@ class IncrementalEnsembleFTS(ensemble.EnsembleFTS):
 
         return ret
 
+    def forecast_ahead(self, data, steps, **kwargs):
+        if len(data) < self.order:
+            return data
+
+        if isinstance(data, np.ndarray):
+            data = data.tolist()
+
+        start = kwargs.get('start_at',0)
+
+        ret = data[:start+self.order]
+        for k in np.arange(start+self.order, steps+start+self.order):
+            tmp = self.forecast(ret[k-self.order:k], no_update=True, **kwargs)
+
+            if isinstance(tmp,(list, np.ndarray)):
+                tmp = tmp[-1]
+
+            ret.append(tmp)
+            data.append(tmp)
+
+        return ret[-steps:]
+
 
 
 

pyFTS/models/incremental/TimeVariant.py

@@ -55,6 +55,11 @@ class Retrainer(fts.FTS):
     def forecast(self, data, **kwargs):
         l = len(data)
 
+        no_update = kwargs.get('no_update',False)
+
+        if no_update:
+            return self.model.predict(data, **kwargs)
+
         horizon = self.window_length + self.order
 
         ret = []
@@ -73,13 +78,34 @@ class Retrainer(fts.FTS):
 
         return ret
 
+    def forecast_ahead(self, data, steps, **kwargs):
+        if len(data) < self.order:
+            return data
+
+        if isinstance(data, np.ndarray):
+            data = data.tolist()
+
+        start = kwargs.get('start_at',0)
+
+        ret = data[:start+self.order]
+        for k in np.arange(start+self.order, steps+start+self.order):
+            tmp = self.forecast(ret[k-self.order:k], no_update=True, **kwargs)
+
+            if isinstance(tmp,(list, np.ndarray)):
+                tmp = tmp[-1]
+
+            ret.append(tmp)
+            data.append(tmp)
+
+        return ret[-steps:]
+
     def offset(self):
         return self.max_lag
 
     def __str__(self):
         """String representation of the model"""
 
-        return str(this.model)
+        return str(self.model)
 
     def __len__(self):
         """

pyFTS/tests/nonstationary.py

@@ -23,10 +23,44 @@ from pyFTS.common import Transformations
 from pyFTS.partitioners import Grid, Util as pUtil
 from pyFTS.benchmarks import benchmarks as bchmk
 from pyFTS.models import chen, hofts, pwfts, hwang
+from pyFTS.models.incremental import TimeVariant, IncrementalEnsemble
 
 train = dataset[:1000]
 test = dataset[1000:]
 
+window = 100
+
+batch = 10
+
+num_models = 3
+
+model1 = TimeVariant.Retrainer(partitioner_method=Grid.GridPartitioner, partitioner_params={'npart': 35},
+                               fts_method=pwfts.ProbabilisticWeightedFTS, fts_params={}, order=1,
+                               batch_size=batch, window_length=window * num_models)
+
+model2 = IncrementalEnsemble.IncrementalEnsembleFTS(partitioner_method=Grid.GridPartitioner,
+                                                    partitioner_params={'npart': 35},
+                                                    fts_method=pwfts.ProbabilisticWeightedFTS, fts_params={}, order=1,
+                                                    batch_size=int(batch / 3), window_length=window,
+                                                    num_models=num_models)
+
+model1.fit(train)
+model2.fit(train)
+
+print(len(test))
+'''
+forecasts1 = model1.predict(test[:-10])
+print(len(forecasts1))
+forecasts1 = model1.predict(test[-10:], steps_ahead=10)
+print(len(forecasts1))
+'''
+forecasts2 = model2.predict(test[:-10])
+print(len(forecasts2))
+forecasts2 = model2.predict(test[-10:], steps_ahead=10)
+print(len(forecasts2))
+
+
+'''
 from pyFTS.models.nonstationary import partitioners as nspart, nsfts, honsfts
 fs = nspart.simplenonstationary_gridpartitioner_builder(data=train,npart=35,transformation=None)
 print(fs)
@@ -36,3 +70,4 @@ model.fit(train)
 print(model)
 forecasts = model.predict(test)
 #print(forecasts)
+'''