7.1 keras 函数式API.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_8 (Dense)              (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "dense_9 (Dense)              (None, 32)                1056      \n",
      "_________________________________________________________________\n",
      "dense_10 (Dense)             (None, 10)                330       \n",
      "=================================================================\n",
      "Total params: 3,466\n",
      "Trainable params: 3,466\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_3 (InputLayer)         (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_11 (Dense)             (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "dense_12 (Dense)             (None, 32)                1056      \n",
      "_________________________________________________________________\n",
      "dense_13 (Dense)             (None, 10)                330       \n",
      "=================================================================\n",
      "Total params: 3,466\n",
      "Trainable params: 3,466\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential, Model\n",
    "from keras import layers\n",
    "from keras import Input\n",
    "seq_model = Sequential()\n",
    "seq_model.add(layers.Dense(32, activation='relu', input_shape=(64,)))\n",
    "seq_model.add(layers.Dense(32, activation='relu'))\n",
    "seq_model.add(layers.Dense(10, activation='softmax'))\n",
    "seq_model.summary()\n",
    "\n",
    "input_tensor = Input(shape=(64,))\n",
    "x = layers.Dense(32, activation='relu')(input_tensor)\n",
    "x = layers.Dense(32, activation='relu')(x)\n",
    "output_tensor = layers.Dense(10, activation='softmax')(x)\n",
    "model = Model(input_tensor, output_tensor)\n",
    "model.summary() "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "7.1.2 多输入模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 用函数式 API 实现双输入问答模型\n",
    "from keras.models import Model\n",
    "from keras import layers\n",
    "from keras import Input\n",
    "\n",
    "text_vocabulary_size = 10000\n",
    "question_vocabulary_size = 10000\n",
    "answer_vocabulary_size = 500\n",
    "\n",
    "text_input = Input(shape=(None,), dtype='int32', name='text')\n",
    "embedded_text = layers.Embedding(text_vocabulary_size, 64)(text_input)\n",
    "encoded_text = layers.LSTM(32)(embedded_text)\n",
    "\n",
    "question_input = Input(shape=(None,),dtype='int32',name='question')\n",
    "embedded_question = layers.Embedding(question_vocabulary_size, 32)(question_input)\n",
    "encoded_question = layers.LSTM(16)(embedded_question)\n",
    "\n",
    "concatenated = layers.concatenate([encoded_text, encoded_question],axis=-1)\n",
    "\n",
    "answer = layers.Dense(answer_vocabulary_size,activation='softmax')(concatenated) \n",
    "\n",
    "model = Model([text_input, question_input], answer)\n",
    "model.compile(optimizer='rmsprop',\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=['acc'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/fc/anaconda3/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.cast instead.\n",
      "Epoch 1/10\n",
      "1000/1000 [==============================] - 3s 3ms/step - loss: 6.2146 - acc: 1.0000e-03\n",
      "Epoch 2/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 6.1993 - acc: 0.0470\n",
      "Epoch 3/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 6.1706 - acc: 0.0490\n",
      "Epoch 4/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 6.0672 - acc: 0.0070\n",
      "Epoch 5/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 6.0123 - acc: 0.0060\n",
      "Epoch 6/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.9708 - acc: 0.0060\n",
      "Epoch 7/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.8838 - acc: 0.0080\n",
      "Epoch 8/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.7881 - acc: 0.0080\n",
      "Epoch 9/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.6971 - acc: 0.0110\n",
      "Epoch 10/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.6464 - acc: 0.0200\n",
      "Epoch 1/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.5614 - acc: 0.0290\n",
      "Epoch 2/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.5050 - acc: 0.0370\n",
      "Epoch 3/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.4402 - acc: 0.0400\n",
      "Epoch 4/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.3807 - acc: 0.0390\n",
      "Epoch 5/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.3238 - acc: 0.0470\n",
      "Epoch 6/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.2718 - acc: 0.0570\n",
      "Epoch 7/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.2134 - acc: 0.0720\n",
      "Epoch 8/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.1811 - acc: 0.0760\n",
      "Epoch 9/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.1432 - acc: 0.0730\n",
      "Epoch 10/10\n",
      "1000/1000 [==============================] - 1s 1ms/step - loss: 5.0804 - acc: 0.0860\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f91254b6908>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 将数据输入到多输入模型中\n",
    "import numpy as np\n",
    "from keras import utils\n",
    "\n",
    "num_samples = 1000\n",
    "max_length = 100\n",
    "\n",
    "text = np.random.randint(1, text_vocabulary_size,size=(num_samples, max_length))\n",
    "question = np.random.randint(1, question_vocabulary_size,size=(num_samples, max_length))\n",
    "answers = np.random.randint(answer_vocabulary_size, size=(num_samples))\n",
    "answers = utils.to_categorical(answers, answer_vocabulary_size)\n",
    "\n",
    "model.fit([text, question], answers, epochs=10, batch_size=128)\n",
    "model.fit({'text': text, 'question': question}, answers,\n",
    "          epochs=10, batch_size=128)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "7.1.3 多输出模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 用函数式 API 实现一个三输出模型\n",
    "from keras import layers\n",
    "from keras import Input\n",
    "from keras.models import Model\n",
    "\n",
    "vocabulary_size = 50000\n",
    "num_income_groups = 10\n",
    "\n",
    "posts_input = Input(shape=(None,), dtype='int32', name='posts')\n",
    "embedded_posts = layers.Embedding(256, vocabulary_size)(posts_input)\n",
    "\n",
    "x = layers.Conv1D(128, 5, activation='relu')(embedded_posts)\n",
    "x = layers.MaxPooling1D(5)(x)\n",
    "x = layers.Conv1D(256, 5, activation='relu')(x)\n",
    "x = layers.Conv1D(256, 5, activation='relu')(x)\n",
    "x = layers.MaxPooling1D(5)(x)\n",
    "x = layers.Conv1D(256, 5, activation='relu')(x)\n",
    "x = layers.Conv1D(256, 5, activation='relu')(x)\n",
    "x = layers.GlobalMaxPooling1D()(x)\n",
    "x = layers.Dense(128, activation='relu')(x)\n",
    "\n",
    "age_prediction = layers.Dense(1, name='age')(x)\n",
    "income_prediction = layers.Dense(num_income_groups,\n",
    "                                 activation='softmax',\n",
    "                                 name='income')(x)\n",
    "gender_prediction = layers.Dense(1, activation='sigmoid', name='gender')(x)\n",
    "model = Model(posts_input,[age_prediction, income_prediction, gender_prediction])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 多输出模型的编译选项：多重损失\n",
    "model.compile(optimizer='rmsprop',\n",
    "              loss=['mse', 'categorical_crossentropy', 'binary_crossentropy'])\n",
    "\n",
    "model.compile(optimizer='rmsprop',\n",
    "              loss={'age': 'mse',\n",
    "                    'income': 'categorical_crossentropy',\n",
    "                    'gender': 'binary_crossentropy'})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 多输出模型的编译选项：损失加权\n",
    "model.compile(optimizer='rmsprop',\n",
    " loss=['mse', 'categorical_crossentropy', 'binary_crossentropy'],\n",
    " loss_weights=[0.25, 1., 10.])\n",
    "model.compile(optimizer='rmsprop',\n",
    " loss={'age': 'mse',\n",
    " 'income': 'categorical_crossentropy',\n",
    " 'gender': 'binary_crossentropy'},\n",
    " loss_weights={'age': 0.25,\n",
    " 'income': 1.,\n",
    " 'gender': 10.})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 将数据输入到多输出模型中\n",
    "model.fit(posts, [age_targets, income_targets, gender_targets],\n",
    " epochs=10, batch_size=64)\n",
    "model.fit(posts, {'age': age_targets,\n",
    " 'income': income_targets,\n",
    " 'gender': gender_targets},\n",
    " epochs=10, batch_size=64) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}