Better shit-detection and mitigation when training

caliGraph.py

@@ -823,11 +823,12 @@ def evaluateFitness(books, debugPrint=False):
     fit = sum(errSq)/len(errSq) + 0.005*regressionLoss + 0.2*boundsLoss/len(ratedBooks) - 1.0*sum(linSepLoss)/len(linSepLoss)
     return fit, gradient
 
-def train(gamma = 1, full=True):
+def train(initGamma = 1, full=True):
     global weights
     if full:
         for wt in weights:
             weights[wt] = random.random()
+    gamma = initGamma
     books = loadBooksFromDB()
     bestWeights = copy.copy(weights)
     mse, gradient = evaluateFitness(books)
@@ -849,12 +850,22 @@ def train(gamma = 1, full=True):
         if mse < best_mse:
             saveWeights(weights)
             bestWeights = copy.copy(weights)
-        else:
+            best_mse = mse
+        if mse > last_mse:
             stagLen += 1
-        if stagLen == 3 or mse > 100:
-            stagLen = -2
-            for wt in weights:
-                weights[wt] = random.random()
+        else:
+            stagLen = 0
+        if stagLen == 4 or mse > 50:
+            print("#")
+            stagLen = 0
+            gamma = initGamma
+            if random.random() < 0.50:
+                for wt in weights:
+                    weights[wt] = random.random()
+            else:
+                weights = copy.copy(bestWeights)
+                for wt in weights:
+                    weights[wt] *= 0.975+0.05*random.random()
     print('Done.')
 
 def saveWeights(weights):