Merge pull request dennybritz#194 from alek5k/dqn-fixes

fixed shape descriptions for neural network input layer

Author: dennybritz

DQN/Deep Q Learning Solution.ipynb

@@ -117,7 +117,7 @@
     "        \"\"\"\n",
     "\n",
     "        # Placeholders for our input\n",
-    "        # Our input are 4 RGB frames of shape 160, 160 each\n",
+    "        # Our input are 4 grayscale frames of shape 84, 84 each\n",
     "        self.X_pl = tf.placeholder(shape=[None, 84, 84, 4], dtype=tf.uint8, name=\"X\")\n",
     "        # The TD target value\n",
     "        self.y_pl = tf.placeholder(shape=[None], dtype=tf.float32, name=\"y\")\n",
@@ -166,7 +166,7 @@
     "\n",
     "        Args:\n",
     "          sess: Tensorflow session\n",
-    "          s: State input of shape [batch_size, 4, 160, 160, 3]\n",
+    "          s: State input of shape [batch_size, 4, 84, 84, 1]\n",
     "\n",
     "        Returns:\n",
     "          Tensor of shape [batch_size, NUM_VALID_ACTIONS] containing the estimated \n",
@@ -180,7 +180,7 @@
     "\n",
     "        Args:\n",
     "          sess: Tensorflow session object\n",
-    "          s: State input of shape [batch_size, 4, 160, 160, 3]\n",
+    "          s: State input of shape [batch_size, 4, 84, 84, 1]\n",
     "          a: Chosen actions of shape [batch_size]\n",
     "          y: Targets of shape [batch_size]\n",
     "\n",

DQN/Deep Q Learning.ipynb

@@ -110,7 +110,7 @@
     "        \"\"\"\n",
     "\n",
     "        # Placeholders for our input\n",
-    "        # Our input are 4 RGB frames of shape 160, 160 each\n",
+    "        # Our input are 4 grayscale frames of shape 84, 84 each\n",
     "        self.X_pl = tf.placeholder(shape=[None, 84, 84, 4], dtype=tf.uint8, name=\"X\")\n",
     "        # The TD target value\n",
     "        self.y_pl = tf.placeholder(shape=[None], dtype=tf.float32, name=\"y\")\n",
@@ -160,7 +160,7 @@
     "\n",
     "        Args:\n",
     "          sess: Tensorflow session\n",
-    "          s: State input of shape [batch_size, 4, 160, 160, 3]\n",
+    "          s: State input of shape [batch_size, 4, 84, 84, 1]\n",
     "\n",
     "        Returns:\n",
     "          Tensor of shape [batch_size, NUM_VALID_ACTIONS] containing the estimated \n",
@@ -174,7 +174,7 @@
     "\n",
     "        Args:\n",
     "          sess: Tensorflow session object\n",
-    "          s: State input of shape [batch_size, 4, 160, 160, 3]\n",
+    "          s: State input of shape [batch_size, 4, 84, 84, 1]\n",
     "          a: Chosen actions of shape [batch_size]\n",
     "          y: Targets of shape [batch_size]\n",
     "\n",

DQN/Double DQN Solution.ipynb

@@ -109,7 +109,7 @@
     "        \"\"\"\n",
     "\n",
     "        # Placeholders for our input\n",
-    "        # Our input are 4 RGB frames of shape 160, 160 each\n",
+    "        # Our input are 4 grayscale frames of shape 84, 84 each\n",
     "        self.X_pl = tf.placeholder(shape=[None, 84, 84, 4], dtype=tf.uint8, name=\"X\")\n",
     "        # The TD target value\n",
     "        self.y_pl = tf.placeholder(shape=[None], dtype=tf.float32, name=\"y\")\n",
@@ -136,7 +136,7 @@
     "\n",
     "        Args:\n",
     "          sess: Tensorflow session\n",
-    "          s: State input of shape [batch_size, 4, 160, 160, 3]\n",
+    "          s: State input of shape [batch_size, 4, 84, 84, 1]\n",
     "\n",
     "        Returns:\n",
     "          Tensor of shape [batch_size, NUM_VALID_ACTIONS] containing the estimated \n",
@@ -150,7 +150,7 @@
     "\n",
     "        Args:\n",
     "          sess: Tensorflow session object\n",
-    "          s: State input of shape [batch_size, 4, 160, 160, 3]\n",
+    "          s: State input of shape [batch_size, 4, 84, 84, 1]\n",
     "          a: Chosen actions of shape [batch_size]\n",
     "          y: Targets of shape [batch_size]\n",
     "\n",