From 8fe5c4eee79c73b0409d4e305e18b51b790e028c Mon Sep 17 00:00:00 2001
From: Alexander Schmidt <alex@treefish.org>
Date: Mon, 2 Nov 2020 22:39:52 +0100
Subject: [PATCH] updating qtable

---
 src/qtable.py | 23 ++++++++++++++---------
 1 file changed, 14 insertions(+), 9 deletions(-)

diff --git a/src/qtable.py b/src/qtable.py
index fb1f390..5dda444 100755
--- a/src/qtable.py
+++ b/src/qtable.py
@@ -6,9 +6,12 @@ import sys
 
 from game import Game
 
+learning_rate = 0.1
+discount_factor = 1.0
+
 states_dim = 147456 # 2^12 * 6^2
 actions_dim = 637 # 12+1 * (6+1)^2
-num_episodes = 10000000
+num_episodes = 10000000000
 
 def find_state_qid(shutable, diced):
     qid = 0
@@ -40,7 +43,7 @@ def select_option(opts, qs):
             return (opt_qid_pair[0], opt_qid_pair[1])
     return (None, None)
 
-Q = np.zeros([states_dim, actions_dim])
+Q = np.ones([states_dim, actions_dim])
 
 running_score = [0.0, 0.0]
 
@@ -54,15 +57,17 @@ for i in range(num_episodes):
             old_score = g.get_score()
             g.shut(opt)
             g.dice()
-            reward = g.get_score() - old_score
+            reward = (g.get_score() - old_score) / 12.0
             new_state_qid = find_state_qid(g.get_shutable(), g.get_diced())
-            lr = 0.1
-            gamma = 0.99
-            Q[state_qid, opt_qid] = Q[state_qid, opt_qid] + \
-                lr * (reward + gamma * np.max(Q[new_state_qid, :]) - Q[state_qid, opt_qid])
+            Q[state_qid, opt_qid] += \
+                learning_rate * (reward
+                                 + discount_factor * np.max(Q[new_state_qid, :])
+                                 - Q[state_qid, opt_qid])
             state_qid = new_state_qid
-    running_score[0] *= 0.999
+        else:
+            Q[state_qid, opt_qid] = 0
+    running_score[0] *= 0.99999999
     running_score[0] += g.get_score()
-    running_score[1] *= 0.999
+    running_score[1] *= 0.99999999
     running_score[1] += 1.0
     print( "%d: %f" % (i, running_score[0]/running_score[1]) )
-- 
2.39.5