From: Alexander Schmidt <alex@treefish.org>
Date: Tue, 3 Nov 2020 07:38:00 +0000 (+0100)
Subject: fixed final q
X-Git-Url: http://git.treefish.org/~alex/shutbox.git/commitdiff_plain/9e3585ba3617d2af399e6a18bfd45b046f7c5e1e?ds=inline

fixed final q
---

diff --git a/src/qtable.py b/src/qtable.py
index 0804187..5bb1c51 100755
--- a/src/qtable.py
+++ b/src/qtable.py
@@ -6,7 +6,7 @@ import sys
 
 from game import Game
 
-learning_rate = 0.1
+learning_rate = 1.0
 discount_factor = 1.0
 
 states_dim = 36864 # 2^10 * 6^2
@@ -34,7 +34,6 @@ def select_option(opts, qs):
         opt_qid = find_option_qid(opt)
         opt_qid_pairs.append( [opt, opt_qid] )
         opt_qsum += qs[opt_qid]
-    #random.shuffle(opt_qid_pairs)
     ran_pt = random.uniform(0.0, opt_qsum)
     decision_pt = 0.0
     for opt_qid_pair in opt_qid_pairs:
@@ -64,7 +63,7 @@ for i in range(num_episodes):
                                  + discount_factor * np.max(Q[new_state_qid, :])
                                  - Q[state_qid, opt_qid])
             state_qid = new_state_qid
-    Q[state_qid, opt_qid] = 0
+    Q[state_qid, :] = 0
     running_score[0] *= 0.99999999
     running_score[0] += g.get_score()
     running_score[1] *= 0.99999999