int32 타입으로 변경, Estimator API 사용, tf.keras.datasets.mnist 사용

10_introduction_to_artificial_neural_networks.ipynb

@@ -115,7 +115,7 @@
     "X = iris.data[:, (2, 3)]  # 꽃잎 길이, 꽃잎 너비\n",
     "y = (iris.target == 0).astype(np.int)\n",
     "\n",
-    "per_clf = Perceptron(max_iter=5, random_state=42)\n",
+    "per_clf = Perceptron(max_iter=100, random_state=42)\n",
     "per_clf.fit(X, y)\n",
     "\n",
     "y_pred = per_clf.predict([[2, 0.5]])"
@@ -322,13 +322,6 @@
     "# MNIST를 위한 FNN"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## tf.learn를 사용"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": 10,
@@ -338,10 +331,94 @@
     "import tensorflow as tf"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "경고: `tf.examples.tutorials.mnist`은 삭제될 예정이므로 대신 `tf.keras.datasets.mnist`를 사용하겠습니다. `tf.contrib.learn` API는 `tf.estimators`와 `tf.feature_columns`로 옮겨졌고 상당히 많이 바뀌었습니다. 특히 `infer_real_valued_columns_from_input()` 함수와 `SKCompat` 클래스가 없습니다."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "(X_train, y_train), (X_test, y_test) = tf.keras.datasets.mnist.load_data()\n",
+    "X_train = X_train.astype(np.float32).reshape(-1, 28*28) / 255.0\n",
+    "X_test = X_test.astype(np.float32).reshape(-1, 28*28) / 255.0\n",
+    "y_train = y_train.astype(np.int32)\n",
+    "y_test = y_test.astype(np.int32)\n",
+    "X_valid, X_train = X_train[:5000], X_train[5000:]\n",
+    "y_valid, y_train = y_train[:5000], y_train[5000:]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Estimator API를 사용 (이전엔 `tf.contrib.learn`)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "feature_cols = [tf.feature_column.numeric_column(\"X\", shape=[28 * 28])]\n",
+    "dnn_clf = tf.estimator.DNNClassifier(hidden_units=[300,100], n_classes=10,\n",
+    "                                     feature_columns=feature_cols)\n",
+    "\n",
+    "input_fn = tf.estimator.inputs.numpy_input_fn(\n",
+    "    x={\"X\": X_train}, y=y_train, num_epochs=40, batch_size=50, shuffle=True)\n",
+    "dnn_clf.train(input_fn=input_fn)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "test_input_fn = tf.estimator.inputs.numpy_input_fn(\n",
+    "    x={\"X\": X_test}, y=y_test, shuffle=False)\n",
+    "eval_results = dnn_clf.evaluate(input_fn=test_input_fn)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "eval_results"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "y_pred_iter = dnn_clf.predict(input_fn=test_input_fn)\n",
+    "y_pred = list(y_pred_iter)\n",
+    "y_pred[0]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## tf.learn를 사용"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 11,
-   "metadata": {},
+   "metadata": {
+    "scrolled": true
+   },
    "outputs": [
     {
      "name": "stdout",
@@ -359,9 +436,8 @@
     }
    ],
    "source": [
-    "from tensorflow.examples.tutorials.mnist import input_data\n",
+    "# from tensorflow.examples.tutorials.mnist import input_data\n",
-    "tf.logging.set_verbosity(tf.logging.ERROR)  # deprecated 경고 메세지를 출력하지 않기 위해 \n",
+    "# mnist = input_data.read_data_sets(\"/tmp/data/\")"
-    "mnist = input_data.read_data_sets(\"/tmp/data/\")"
    ]
   },
   {
@@ -370,16 +446,18 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "X_train = mnist.train.images\n",
+    "# X_train = mnist.train.images\n",
-    "X_test = mnist.test.images\n",
+    "# X_test = mnist.test.images\n",
-    "y_train = mnist.train.labels.astype(\"int\")\n",
+    "# y_train = mnist.train.labels.astype(\"int\")\n",
-    "y_test = mnist.test.labels.astype(\"int\")"
+    "# y_test = mnist.test.labels.astype(\"int\")"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 13,
-   "metadata": {},
+   "metadata": {
+    "collapsed": true
+   },
    "outputs": [
     {
      "name": "stdout",
@@ -1241,8 +1319,6 @@
     }
    ],
    "source": [
-    "import tensorflow as tf\n",
-    "\n",
     "config = tf.contrib.learn.RunConfig(tf_random_seed=42) # 책에는 없음\n",
     "\n",
     "feature_cols = tf.contrib.learn.infer_real_valued_columns_from_input(X_train)\n",
@@ -1341,7 +1417,7 @@
     "reset_graph()\n",
     "\n",
     "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
-    "y = tf.placeholder(tf.int64, shape=(None), name=\"y\")"
+    "y = tf.placeholder(tf.int32, shape=(None), name=\"y\")"
    ]
   },
   {
@@ -1434,6 +1510,20 @@
     "batch_size = 50"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def shuffle_batch(X, y, batch_size):\n",
+    "    rnd_idx = np.random.permutation(len(X))\n",
+    "    n_batches = len(X) // batch_size\n",
+    "    for batch_idx in np.array_split(rnd_idx, n_batches):\n",
+    "        X_batch, y_batch = X[batch_idx], y[batch_idx]\n",
+    "        yield X_batch, y_batch"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 25,
@@ -1490,13 +1580,11 @@
     "with tf.Session() as sess:\n",
     "    init.run()\n",
     "    for epoch in range(n_epochs):\n",
-    "        for iteration in range(mnist.train.num_examples // batch_size):\n",
+    "        for X_batch, y_batch in shuffle_batch(X_train, y_train, batch_size):\n",
-    "            X_batch, y_batch = mnist.train.next_batch(batch_size)\n",
     "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
-    "        acc_train = accuracy.eval(feed_dict={X: X_batch, y: y_batch})\n",
+    "        acc_batch = accuracy.eval(feed_dict={X: X_batch, y: y_batch})\n",
-    "        acc_val = accuracy.eval(feed_dict={X: mnist.validation.images,\n",
+    "        acc_valid = accuracy.eval(feed_dict={X: X_valid, y: y_valid})\n",
-    "                                            y: mnist.validation.labels})\n",
+    "        print(epoch, \"배치 데이터 정확도:\", acc_batch, \"검증 세트 정확도:\", acc_valid)\n",
-    "        print(epoch, \"훈련 정확도:\", acc_train, \"검증 세트 정확도:\", acc_val)\n",
     "\n",
     "    save_path = saver.save(sess, \"./my_model_final.ckpt\")"
    ]
@@ -1517,7 +1605,7 @@
    "source": [
     "with tf.Session() as sess:\n",
     "    saver.restore(sess, \"./my_model_final.ckpt\") # 또는 save_path를 사용합니다\n",
-    "    X_new_scaled = mnist.test.images[:20]\n",
+    "    X_new_scaled = X_test[:20]\n",
     "    Z = logits.eval(feed_dict={X: X_new_scaled})\n",
     "    y_pred = np.argmax(Z, axis=1)"
    ]
@@ -1538,7 +1626,7 @@
    ],
    "source": [
     "print(\"예측 클래스:\", y_pred)\n",
-    "print(\"진짜 클래스:\", mnist.test.labels[:20])"
+    "print(\"진짜 클래스:\", y_test[:20])"
    ]
   },
   {
@@ -1547,6 +1635,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "# 크롬 브라우저는 확인되었지만 다른 브라우저에서는 작동되지 않을 수 있습니다.\n",
+    "\n",
     "from IPython.display import clear_output, Image, display, HTML\n",
     "\n",
     "def strip_consts(graph_def, max_const_size=32):\n",
@@ -1647,7 +1737,7 @@
     "reset_graph()\n",
     "\n",
     "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
-    "y = tf.placeholder(tf.int64, shape=(None), name=\"y\") "
+    "y = tf.placeholder(tf.int32, shape=(None), name=\"y\") "
    ]
   },
   {
@@ -1661,7 +1751,8 @@
     "                              activation=tf.nn.relu)\n",
     "    hidden2 = tf.layers.dense(hidden1, n_hidden2, name=\"hidden2\",\n",
     "                              activation=tf.nn.relu)\n",
-    "    logits = tf.layers.dense(hidden2, n_outputs, name=\"outputs\")"
+    "    logits = tf.layers.dense(hidden2, n_outputs, name=\"outputs\")\n",
+    "    y_proba = tf.nn.softmax(logits)"
    ]
   },
   {
@@ -1748,12 +1839,11 @@
     "with tf.Session() as sess:\n",
     "    init.run()\n",
     "    for epoch in range(n_epochs):\n",
-    "        for iteration in range(mnist.train.num_examples // batch_size):\n",
+    "        for X_batch, y_batch in shuffle_batch(X_train, y_train, batch_size):\n",
-    "            X_batch, y_batch = mnist.train.next_batch(batch_size)\n",
     "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
-    "        acc_train = accuracy.eval(feed_dict={X: X_batch, y: y_batch})\n",
+    "        acc_batch = accuracy.eval(feed_dict={X: X_batch, y: y_batch})\n",
-    "        acc_test = accuracy.eval(feed_dict={X: mnist.test.images, y: mnist.test.labels})\n",
+    "        acc_valid = accuracy.eval(feed_dict={X: X_valid, y: y_valid})\n",
-    "        print(epoch, \"훈련 정확도:\", acc_train, \"테스트 정확도:\", acc_test)\n",
+    "        print(epoch, \"배치 데이터 정확도:\", acc_batch, \"검증 세트 정확도:\", acc_valid)\n",
     "\n",
     "    save_path = saver.save(sess, \"./my_model_final.ckpt\")"
    ]
@@ -1859,7 +1949,7 @@
     "reset_graph()\n",
     "\n",
     "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
-    "y = tf.placeholder(tf.int64, shape=(None), name=\"y\") "
+    "y = tf.placeholder(tf.int32, shape=(None), name=\"y\") "
    ]
   },
   {
@@ -1976,7 +2066,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "잠시만요! 조기 종료를 구현하는 것이 좋겠죠? 이렇게 하려면 검증 세트가 필요합니다. 다행히 텐서플로의 `input_data()` 함수에서 반환되는 데이터셋은 이미 훈련 세트(이미 섞여있는 60,000개의 샘플), 검증 세트(5,000개의 샘플), 테스트 세트(5,000개의 샘플)로 나뉘어져 있습니다. 손쉽게 `X_valid`와 `y_valid`를 정의할 수 있습니다:"
+    "잠시만요! 조기 종료를 구현하는 것이 좋겠죠? 이렇게 하려면 검증 세트가 필요합니다."
    ]
   },
   {
@@ -2060,8 +2150,7 @@
     "        sess.run(init)\n",
     "\n",
     "    for epoch in range(start_epoch, n_epochs):\n",
-    "        for iteration in range(mnist.train.num_examples // batch_size):\n",
+    "        for X_batch, y_batch in shuffle_batch(X_train, y_train, batch_size):\n",
-    "            X_batch, y_batch = mnist.train.next_batch(batch_size)\n",
     "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
     "        accuracy_val, loss_val, accuracy_summary_str, loss_summary_str = sess.run([accuracy, loss, accuracy_summary, loss_summary], feed_dict={X: X_valid, y: y_valid})\n",
     "        file_writer.add_summary(accuracy_summary_str, epoch)\n",