Update lab-12-0-rnn_basics.ipynb

qoocrab · kkweon · commit dbbe3977de10 · 2019-01-17T09:08:19.000-08:00
1. tf.contrib.rnn.BasicRNNCell will be removed in a future version. so changed to tf.keras.layers.SimpleRNNCell.
2. tf.contrib.rnn.BasicLSTMCell will be removed in a future version. so changed to tf.nn.rnn_cell.LSTMCell.
diff --git a/lab-12-0-rnn_basics.ipynb b/lab-12-0-rnn_basics.ipynb
@@ -63,7 +63,7 @@
     "with tf.variable_scope('one_cell') as scope:\n",
     "    # One cell RNN input_dim (4) -> output_dim (2)\n",
     "    hidden_size = 2\n",
-    "    cell = tf.contrib.rnn.BasicRNNCell(num_units=hidden_size)\n",
+    "    cell = tf.keras.layers.SimpleRNNCell(units=hidden_size)\n",
     "    print(cell.output_size, cell.state_size)\n",
     "\n",
     "    x_data = np.array([[h]], dtype=np.float32) # x_data = [[[1,0,0,0]]]\n",
@@ -110,7 +110,7 @@
     "with tf.variable_scope('two_sequances') as scope:\n",
     "    # One cell RNN input_dim (4) -> output_dim (2). sequence: 5\n",
     "    hidden_size = 2\n",
-    "    cell = tf.contrib.rnn.BasicRNNCell(num_units=hidden_size)\n",
+    "    cell = tf.keras.layers.SimpleRNNCell(units=hidden_size)\n",
     "    x_data = np.array([[h, e, l, l, o]], dtype=np.float32)\n",
     "    print(x_data.shape)\n",
     "    pp.pprint(x_data)\n",
@@ -184,7 +184,7 @@
     "    pp.pprint(x_data)\n",
     "    \n",
     "    hidden_size = 2\n",
-    "    cell = rnn.BasicLSTMCell(num_units=hidden_size, state_is_tuple=True)\n",
+    "    cell = tf.nn.rnn_cell.LSTMCell(num_units=hidden_size, state_is_tuple=True)\n",
     "    outputs, _states = tf.nn.dynamic_rnn(\n",
     "        cell, x_data, dtype=tf.float32)\n",
     "    sess.run(tf.global_variables_initializer())\n",
@@ -249,7 +249,7 @@
     "    pp.pprint(x_data)\n",
     "    \n",
     "    hidden_size = 2\n",
-    "    cell = rnn.BasicLSTMCell(num_units=hidden_size, state_is_tuple=True)\n",
+    "    cell = tf.nn.rnn_cell.LSTMCell(num_units=hidden_size, state_is_tuple=True)\n",
     "    outputs, _states = tf.nn.dynamic_rnn(\n",
     "        cell, x_data, sequence_length=[5,3,4], dtype=tf.float32)\n",
     "    sess.run(tf.global_variables_initializer())\n",
@@ -314,7 +314,7 @@
     "    \n",
     "    # One cell RNN input_dim (4) -> output_dim (5). sequence: 5, batch: 3\n",
     "    hidden_size=2\n",
-    "    cell = rnn.BasicLSTMCell(num_units=hidden_size, state_is_tuple=True)\n",
+    "    cell = tf.nn.rnn_cell.LSTMCell(num_units=hidden_size, state_is_tuple=True)\n",
     "    initial_state = cell.zero_state(batch_size, tf.float32)\n",
     "    outputs, _states = tf.nn.dynamic_rnn(cell, x_data,\n",
     "                                         initial_state=initial_state, dtype=tf.float32)\n",
@@ -412,7 +412,7 @@
    "source": [
     "with tf.variable_scope('generated_data') as scope:\n",
     "    # One cell RNN input_dim (3) -> output_dim (5). sequence: 5, batch: 3\n",
-    "    cell = rnn.BasicLSTMCell(num_units=5, state_is_tuple=True)\n",
+    "    cell = tf.nn.rnn_cell.LSTMCell(num_units=5, state_is_tuple=True)\n",
     "    initial_state = cell.zero_state(batch_size, tf.float32)\n",
     "    outputs, _states = tf.nn.dynamic_rnn(cell, x_data,\n",
     "                                         initial_state=initial_state, dtype=tf.float32)\n",
@@ -470,7 +470,7 @@
    "source": [
     "with tf.variable_scope('MultiRNNCell') as scope:\n",
     "    # Make rnn\n",
-    "    cell = rnn.BasicLSTMCell(num_units=5, state_is_tuple=True)\n",
+    "    cell = tf.nn.rnn_cell.LSTMCell(num_units=5, state_is_tuple=True)\n",
     "    cell = rnn.MultiRNNCell([cell] * 3, state_is_tuple=True) # 3 layers\n",
     "\n",
     "    # rnn in/out\n",
@@ -529,7 +529,7 @@
    ],
    "source": [
     "with tf.variable_scope('dynamic_rnn') as scope:\n",
-    "    cell = rnn.BasicLSTMCell(num_units=5, state_is_tuple=True)\n",
+    "    cell = tf.nn.rnn_cell.LSTMCell(num_units=5, state_is_tuple=True)\n",
     "    outputs, _states = tf.nn.dynamic_rnn(cell, x_data, dtype=tf.float32,\n",
     "                                         sequence_length=[1, 3, 2])\n",
     "    # lentgh 1 for batch 1, lentgh 2 for batch 2\n",
@@ -642,8 +642,8 @@
    "source": [
     "with tf.variable_scope('bi-directional') as scope:\n",
     "    # bi-directional rnn\n",
-    "    cell_fw = rnn.BasicLSTMCell(num_units=5, state_is_tuple=True)\n",
-    "    cell_bw = rnn.BasicLSTMCell(num_units=5, state_is_tuple=True)\n",
+    "    cell_fw = tf.nn.rnn_cell.LSTMCell(num_units=5, state_is_tuple=True)\n",
+    "    cell_bw = tf.nn.rnn_cell.LSTMCell(num_units=5, state_is_tuple=True)\n",
     "\n",
     "    outputs, states = tf.nn.bidirectional_dynamic_rnn(cell_fw, cell_bw, x_data,\n",
     "                                                      sequence_length=[2, 3, 1],\n",
