From 5a4e48d86c8935851dda76c2d583bdc384b45665 Mon Sep 17 00:00:00 2001
From: Dominik Roth <dominik.roth.dev@gmail.com>
Date: Fri, 24 Sep 2021 19:12:09 +0200
Subject: [PATCH] Added a regression-loss (push weights towards 1)

---
 caliGraph.py | 22 ++++++++++++----------
 1 file changed, 12 insertions(+), 10 deletions(-)

diff --git a/caliGraph.py b/caliGraph.py
index c370ac6..3d82115 100755
--- a/caliGraph.py
+++ b/caliGraph.py
@@ -341,7 +341,7 @@ def scoreOpinions(G, globMu, globStd):
 def scoreUnread(G, globMu, globStd):
     for n in list(G.nodes):
         feedbacks = [globMu]
-        weights = [getWeightForType('mu')]
+        ws = [getWeightForType('mu')]
         node = G.nodes[n]
         if node['t'] == 'book':
             if node['rating'] == None:
@@ -352,16 +352,16 @@ def scoreUnread(G, globMu, globStd):
                         w = getWeightForType(adjNode['t'], G[n][adj]['weight'] if 'weight' in G[n][adj] else 1)
                         for fb in adjNode['feedbacks']:
                             feedbacks.append(fb)
-                            weights.append(w)
+                            ws.append(w)
                 if len(feedbacks):
                     node['mean'], node['std'] = norm.fit(feedbacks)
                     node['se'] = globStd / math.sqrt(len(feedbacks))
                     feedbacks.append(node['std'])
-                    weights.append(getWeightForType('sigma'))
+                    ws.append(getWeightForType('sigma'))
                     feedbacks.append(node['se'])
-                    weights.append(getWeightForType('se'))
-                    node['score'] = sum([fb*w for fb, w in zip(feedbacks, weights)])/len(feedbacks)
-                    #node['score'] = sum([fb*w for fb, w in zip(feedbacks, weights)])/sum(weights)
+                    ws.append(getWeightForType('se'))
+                    #node['score'] = sum([fb*w for fb, w in zip(feedbacks, weights)])/len(feedbacks)
+                    node['score'] = sum([fb*w for fb, w in zip(feedbacks, ws)])/len(feedbacks)
                 else:
                     node['score'] = globMu + errorFac*globStd + len(feedbacks)*0.0000000001
                 if 'series' in node:
@@ -683,6 +683,7 @@ def waveFlow(G, node, n, dist, menge, firstEdge=False):
             waveFlow(G, node, m, dist, menge, firstEdge=firstEdge)
 
 def evaluateFitness(books):
+    global weights
     G = buildBookGraph(books)
     graphAddAuthors(G, books)
     graphAddRecommenders(G, books)
@@ -703,7 +704,8 @@ def evaluateFitness(books):
         if rating > 10.0:
             errSq[-1] *= 1.5
         G.nodes[m]['rating'] = rating
-    return sum(errSq) / len(errSq)
+    regressionLoss = sum([(1-w)**2 for w in weights.values()])
+    return sum(errSq)/len(errSq) + regressionLoss/1000
 
 def train(gamma = 0.1):
     global weights
@@ -720,15 +722,15 @@ def train(gamma = 0.1):
         if gamma < 0.01:
             while random.random() < 0.5:
                 attr = random.choice(w)
-                weights[attr] = -0.1+random.random()*1.5
+                weights[attr] = -0.1+random.random()*1.5+random.random()
         else:
             weights[attr] += delta
         if attr not in ['sigma', 'mu', 'se']:
-            weights[attr] = min(max(0, weights[attr]), 3)
+            weights[attr] = min(max(0.0, weights[attr]), 3.0)
         mse = evaluateFitness(books)
         if mse < best_mse: # got better
             saveWeights(weights)
-            gamma = max(gamma*1.75, 0.001)
+            gamma = min(max(gamma*1.75, 0.001), 0.5)
             bestWeights = copy.copy(weights)
             best_mse = mse
             delta *= 2