Python 3.11 support #227. Reorganized short list / competition logs.

antoinecarme · Mar 9, 2023 · e3e5466 · e3e5466
1 parent 7227165
commit e3e5466
Show file tree

Hide file tree

Showing 3 changed files with 46 additions and 15 deletions.
diff --git a/pyaf/TS/ModelSelection_Voting.py b/pyaf/TS/ModelSelection_Voting.py
@@ -39,6 +39,7 @@ def collectPerformanceIndices_ModelSelection(self, iSignal, iSigDecs) :
         logger = tsutil.get_pyaf_logger();
         lTimer = tsutil.cTimer(("MODEL_SELECTION", {"Signal" : iSignal, "Transformations" : sorted(list(iSigDecs.keys()))}))
         lVotingScores = self.compute_voting_scores(iSigDecs)
+        lCriterion = self.mOptions.mModelSelection_Criterion
         rows_list = []
         lPerfsByModel = {}
         for (lName, sigdec) in iSigDecs.items():
@@ -47,26 +48,38 @@ def collectPerformanceIndices_ModelSelection(self, iSignal, iSigDecs) :
                 lTranformName = sigdec.mSignal;
                 lDecompType = model[1];
                 lModelFormula = model
-                lModelCategory = value[0][2].get_model_category()
+                tsmodel = value[0][2]
+                lModelCategory = tsmodel.get_model_category()
                 lSplit = value[0][2].mTimeInfo.mOptions.mCustomSplit
                 #  value format : self.mPerfsByModel[lModel.mOutName] = [lModel, lComplexity, lFitPerf , lForecastPerf, lTestPerf];
                 lComplexity = value[1];
-                lFitPerf = value[2];
-                lForecastPerf = value[3];
-                lTestPerf = value[4];
+                model_perfs = tsmodel.get_perfs_summary()
+                lFitPerf = model_perfs["Fit"];
+                lForecastPerf = model_perfs["Forecast"]
+                lTestPerf = model_perfs["Test"]
+                H = tsmodel.mTimeInfo.mHorizon
+
                 lVoting = lVotingScores[ lModelFormula[3] ]
                 row = [lSplit, lTranformName, lDecompType, lModelFormula[3], lModelFormula, lModelCategory, lComplexity,
-                       lFitPerf,
-                       lForecastPerf,
-                       lTestPerf, lVoting]
+                       lFitPerf[1].getCriterionValue(lCriterion),
+                       lFitPerf[H].getCriterionValue(lCriterion), 
+                       lForecastPerf[1].getCriterionValue(lCriterion),
+                       lForecastPerf[H].getCriterionValue(lCriterion),
+                       lTestPerf[1].getCriterionValue(lCriterion),
+                       lTestPerf[H].getCriterionValue(lCriterion),
+                       lVoting]
                 rows_list.append(row);
 
         self.mTrPerfDetails =  pd.DataFrame(rows_list, columns=
                                             ('Split', 'Transformation', 'DecompositionType',
                                              'Model', 'DetailedFormula', 'Category', 'Complexity',
-                                             'Fit' + self.mOptions.mModelSelection_Criterion,
-                                             'Forecast' + self.mOptions.mModelSelection_Criterion,
-                                             'Test' + self.mOptions.mModelSelection_Criterion, "Voting")) 
+                                             'Fit_' + self.mOptions.mModelSelection_Criterion + "_1",
+                                             'Fit_' + self.mOptions.mModelSelection_Criterion + "_H",
+                                             'Forecast_' + self.mOptions.mModelSelection_Criterion + "_1",
+                                             'Forecast_' + self.mOptions.mModelSelection_Criterion + "_H",
+                                             'Test_' + self.mOptions.mModelSelection_Criterion + "_1",
+                                             'Test_' + self.mOptions.mModelSelection_Criterion + "_H",
+                                             "Voting")) 
         # print(self.mTrPerfDetails.head(self.mTrPerfDetails.shape[0]));
         lIndicator = 'Voting';
         lBestPerf = self.mTrPerfDetails[ lIndicator ].max();
@@ -88,8 +101,8 @@ def collectPerformanceIndices_ModelSelection(self, iSignal, iSigDecs) :
         # print("BEST_MODEL", lBestName, lBestModel)
         self.mBestModel = lBestModel
         self.mPerfsByModel = lPerfsByModel
-        self.mModelShortList = lInterestingModels[['Transformation', 'DecompositionType', 'Model', lIndicator, 'Complexity']] 
-        print(self.mModelShortList.head());
+        self.mModelShortList = lInterestingModels[['Transformation', 'DecompositionType', 'Model', lIndicator, 'Complexity', 'Forecast_' + self.mOptions.mModelSelection_Criterion + "_1",  'Forecast_' + self.mOptions.mModelSelection_Criterion + "_H"]] 
+        # print(self.mModelShortList.head());
         return (iSignal, lPerfsByModel, lBestModel, self.mModelShortList)
 
 

diff --git a/pyaf/TS/SignalDecomposition.py b/pyaf/TS/SignalDecomposition.py
@@ -96,8 +96,6 @@ def train(self , iInputDS, iTimes, iSignals, iHorizons, iExogenousData = None):
         self.mModelShortListBySignal = lTrainer.mModelShortList
         # some backward compatibility
         lFirstSignal = self.mSignals[0] 
-        self.mTrPerfDetails = lTrainer.mTrPerfDetails[lFirstSignal]
-        self.mModelShortList = lTrainer.mModelShortList[lFirstSignal]
         self.mBestModel = self.mBestModels[lFirstSignal]
         self.mTrainingTime = lTimer.get_elapsed_time()
 
@@ -116,12 +114,22 @@ def getModelFormula(self):
             lFormula[lSignal] = self.mBestModel.getFormula();
         return lFormula;
 
+    def get_competition_details(self):
+        logger = tsutil.get_pyaf_logger();
+        for lSignal in self.mSignals:
+            logger.info("COMPETITION_DETAIL_START '" + lSignal + "'");
+            lShortList_Dict = self.mModelShortListBySignal[lSignal].to_dict(orient = 'index')
+            # print(lShortList_Dict)
+            for (k, v) in lShortList_Dict.items():
+                logger.info("COMPETITION_DETAIL_SHORT_LIST '" + lSignal + "' " + str(k) + " " + str(v));
+            logger.info("COMPETITION_DETAIL_END '" + lSignal + "'");
 
     def getModelInfo(self):
         for lSignal in self.mSignals:
             self.mBestModels[lSignal].getInfo()
         logger = tsutil.get_pyaf_logger();
         logger.info("TRAINING_TIME_IN_SECONDS " + str(self.mTrainingTime));
+        self.get_competition_details()
 
     def to_dict(self, iWithOptions = False):
         dict1 = {}

diff --git a/pyaf/TS/TimeSeriesModel.py b/pyaf/TS/TimeSeriesModel.py
@@ -87,7 +87,17 @@ def updatePerfs(self, compute_all_indicators = False):
         self.mForecastPerfs = lPredictionIntervalsEstimator.mForecastPerformances
         self.mTestPerfs = lPredictionIntervalsEstimator.mTestPerformances
 
-
+    def get_perfs_summary(self):
+        output = {"Fit" : {}, "Forecast" : {}, "Test" : {}}
+        lForecastColumn = str(self.mOriginalSignal) + "_Forecast";
+        lCriterion = self.mTimeInfo.mOptions.mModelSelection_Criterion
+        for h in [1, self.mTimeInfo.mHorizon]:
+            lHorizonName = lForecastColumn + "_" + str(h);
+            output["Fit"][h] = self.mFitPerfs[lHorizonName]
+            output["Forecast"][h] = self.mForecastPerfs[lHorizonName]
+            output["Test"][h] = self.mTestPerfs[lHorizonName]
+        return output
+
     def aggregate_criteria(self, criteria):
         lAggregated = criteria[0]
         return lAggregated