Revize 76072df0

Re #8597 Implementation of AbstractClassificationPredictor, SdcaMEClassifier

Server/ServerApp/Predictor/PredictionController.cs

@@ -93 +93 @@
             this.FeatureExtractor = new FeatureExtractor(this.DataParser, this.Configuration);
 
             DirectoryInfo di = new DirectoryInfo(Configuration.ModelDataPath);
+            _log.Debug($"Looking for directory: {Configuration.ModelDataPath}");
+            if (! di.Exists)
+            {
+                _log.Info($"Creating the model data directory: {Configuration.ModelDataPath}");
+                di.Create();
+            }
             FileInfo[] files = di.GetFiles();
-            if (Array.FindAll(files, f => Regex.IsMatch(f.Name, @"[-]?\d+_\d+.zip")).GroupBy(f => f.Name.Split("_.".ToCharArray())[0]).OrderBy(f => DateTime.FromBinary(Convert.ToInt64(f.Key))).Any())
+            if (Array.FindAll(files, f => Regex.IsMatch(f.Name, @"[-]?\d+_\d+.zip")).GroupBy(f => f.Name.Split("_.".ToCharArray())[0]).OrderBy(f => DateTime.FromBinary(Convert.ToInt64(f.Key))).Count() == Configuration.PredictorCount)
             {
                 _log.Info("Found existing predictors, loading the newest.");
-                this.Load(Array.FindAll(files, f => Regex.IsMatch(f.Name, @"[-]?\d+_\d+.zip")).GroupBy(f => f.Name.Split("_.".ToCharArray())[0]).OrderBy(f => DateTime.FromBinary(Convert.ToInt64(f.Key))).Last().Select(f => f.Name.Split("_".ToCharArray())[0]).First());
+                var predictorID = Array.FindAll(files, f => Regex.IsMatch(f.Name, @"[-]?\d+_\d+.zip")).GroupBy(f => f.Name.Split("_.".ToCharArray())[0]).OrderBy(f => DateTime.FromBinary(Convert.ToInt64(f.Key))).Last().Select(f => f.Name.Split("_".ToCharArray())[0]).First();
+                this.Load(predictorID);
             }
             else
             {
                 _log.Info("No predictors found, creating new ones");
                 for (int i = 0; i < this.Configuration.PredictorCount; i++)
                 {
-                    predictors[i] = new NaiveBayesClassifier();
+                    predictors[i] = new SdcaMaximumEntropyClassifier();
+                }
+                lock (predictorsLock)
+                {
+                    this.Predictors = predictors;
                 }
             }
-            lock (predictorsLock)
-            {
-                this.Predictors = predictors;
-            }
+            
             PredictorConfiguration.SaveConfig(PredictorConfiguration.DEFAULT_CONFIG_PATH, Configuration);
         }
         public List<string> GetPredictors()
@@ -120 +128 @@
 
         public int Load(string predictorID)
         {
-            DirectoryInfo di = new DirectoryInfo(Configuration.ModelDataPath);
+            DirectoryInfo di = new DirectoryInfo(Configuration.ModelDataPath); 
+            if (!di.Exists)
+            {
+                _log.Warn("The model data directory could not be found.");
+                return 2;
+            }
             FileInfo[] files = di.GetFiles($"{predictorID}_*.zip");
             if (Array.FindAll(files, f => Regex.IsMatch(f.Name, $@"{predictorID}_\d+.zip")).Any()){
                 IPredictor[] newPredictors = new IPredictor[this.Configuration.PredictorCount];
@@ -128 +141 @@
                 {
                     for (int i = 0; i < this.Configuration.PredictorCount; i++)
                     {
-                        newPredictors[i] = new NaiveBayesClassifier(Array.Find(files, f => Regex.IsMatch(f.Name, $"{predictorID}_{i}.zip")).FullName);
+                        newPredictors[i] = new SdcaMaximumEntropyClassifier(Array.Find(files, f => Regex.IsMatch(f.Name, $"{predictorID}_{i}.zip")).FullName);
                     }
                     files = di.GetFiles($"{predictorID}.txt");
                     var dataFilenames = File.ReadLines(files[0].FullName); 
@@ -154 +167 @@
         public void Save()
         {
             DirectoryInfo di = new DirectoryInfo(Configuration.ModelDataPath);
+            if (!di.Exists)
+            {
+                _log.Warn("The model data directory could not be found.");
+                return;
+            }
 
             lock (predictorsLock)
             {
@@ -168 +186 @@
         public int Rollback()
         {
             DirectoryInfo di = new DirectoryInfo(Configuration.ModelDataPath);
+            if (!di.Exists)
+            {
+                _log.Warn("The model data directory could not be found.");
+                return 2;
+            }
             FileInfo[] files = di.GetFiles();
-            if (Array.FindAll(files, f => Regex.IsMatch(f.Name, @"[-]?\d+_\d+.zip")).GroupBy(f => f.Name.Split("_.".ToCharArray())[0]).OrderBy(f => DateTime.FromBinary(Convert.ToInt64(f.Key))).Any())
+            if (Array.FindAll(files, f => Regex.IsMatch(f.Name, @"[-]?\d+_\d+.zip")).GroupBy(f => f.Name.Split("_.".ToCharArray())[0]).OrderBy(f => DateTime.FromBinary(Convert.ToInt64(f.Key))).Count() == Configuration.PredictorCount)
             {
                 string RollbackedPredictorID = Array.FindAll(files, f => Regex.IsMatch(f.Name, @"[-]?\d+_\d+.zip")).GroupBy(f => f.Name.Split("_.".ToCharArray())[0]).OrderBy(f => DateTime.FromBinary(Convert.ToInt64(f.Key))).Last().Select(f => f.Name.Split("_".ToCharArray())[0]).First();
                 this.Delete(this.PredictorID);
@@ -184 +207 @@
         private void Delete(string predictorID)
         {
             DirectoryInfo di = new DirectoryInfo(Configuration.ModelDataPath);
+            if (!di.Exists)
+            {
+                _log.Warn("The model data directory could not be found.");
+                return;
+            }
 
             for (int i = 0; i < this.Configuration.PredictorCount; i++)
             {