Tweaked training

caliGraph.py

@@ -682,7 +682,10 @@ def evaluateFitness():
         rating = G.nodes[m]['rating']
         G.nodes[m]['rating'] = None
         mu, std = genScores(G, books)
-        errSq.append((rating - G.nodes[m]['score'])**2)
+        if G.nodes[m]['score'] > rating: # over estimated
+            errSq.append(((rating - G.nodes[m]['score'])**2)*1.5)
+        else:
+            errSq.append((rating - G.nodes[m]['score'])**2)
         G.nodes[m]['rating'] = rating
     return sum(errSq) / len(errSq)
 
@@ -697,11 +700,17 @@ def train(gamma = 0.1):
     while True:
         print({'mse': best_mse, 'w': weights, 'gamma': gamma})
         weights = copy.copy(bestWeights)
-        weights[attr] += delta
+        if gamma < 0.01 and random.random() < 0.5:
+            gamma = 0.01
+            weights[attr] = -1+random.random()*2
+        else:
+            weights[attr] += delta
+        if attr not in ['sigma, mu']:
+            weights[attr] = min(max(0, weight[attr]), 1.5)
         mse = evaluateFitness()
         if mse < best_mse: # got better
             saveWeights(weights)
-            gamma *= 1.1
+            gamma *= 1.75
             bestWeights = copy.copy(weights)
             best_mse = mse
             delta *= 2