{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using Theano backend.\n",
      "Using gpu device 0: GeForce GT625M (CNMeM is disabled, cuDNN not available)\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Activation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## sequential"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "1000/1000 [==============================] - 0s - loss: 2.9005 - acc: 0.5010     \n",
      "Epoch 2/10\n",
      "1000/1000 [==============================] - 0s - loss: 4.4020 - acc: 0.3610     \n",
      "Epoch 3/10\n",
      "1000/1000 [==============================] - 0s - loss: 7.2061 - acc: 0.5130     \n",
      "Epoch 4/10\n",
      "1000/1000 [==============================] - 0s - loss: 7.7480 - acc: 0.5140     \n",
      "Epoch 5/10\n",
      "1000/1000 [==============================] - 0s - loss: 7.7480 - acc: 0.5140     \n",
      "Epoch 6/10\n",
      "1000/1000 [==============================] - 0s - loss: 7.7480 - acc: 0.5140     \n",
      "Epoch 7/10\n",
      "1000/1000 [==============================] - 0s - loss: 7.7480 - acc: 0.5140     \n",
      "Epoch 8/10\n",
      "1000/1000 [==============================] - 0s - loss: 7.7480 - acc: 0.5140     \n",
      "Epoch 9/10\n",
      "1000/1000 [==============================] - 0s - loss: 7.7480 - acc: 0.5140     \n",
      "Epoch 10/10\n",
      "1000/1000 [==============================] - 0s - loss: 7.7480 - acc: 0.5140     \n",
      "CPU times: user 1.7 s, sys: 416 ms, total: 2.12 s\n",
      "Wall time: 1min 14s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Dense(1,input_dim=784, activation='tanh'))\n",
    "model.compile(loss='binary_crossentropy', optimizer='sgd', metrics=['accuracy'])\n",
    "\n",
    "data = np.random.random((1000,784))\n",
    "labels = np.random.randint(2,size=(1000,1))\n",
    "\n",
    "model.fit(data,labels,nb_epoch=10,batch_size=32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "1000/1000 [==============================] - 0s - loss: 4.2552 - acc: 0.4020     \n",
      "Epoch 2/10\n",
      "1000/1000 [==============================] - 0s - loss: 5.8849 - acc: 0.3460     \n",
      "Epoch 3/10\n",
      "1000/1000 [==============================] - 0s - loss: 3.8482 - acc: 0.4740     \n",
      "Epoch 4/10\n",
      "1000/1000 [==============================] - 0s - loss: 2.6821 - acc: 0.5090     \n",
      "Epoch 5/10\n",
      "1000/1000 [==============================] - 0s - loss: 2.6482 - acc: 0.4820     \n",
      "Epoch 6/10\n",
      "1000/1000 [==============================] - 0s - loss: 2.8464 - acc: 0.4740     \n",
      "Epoch 7/10\n",
      "1000/1000 [==============================] - 0s - loss: 3.3461 - acc: 0.4530     \n",
      "Epoch 8/10\n",
      "1000/1000 [==============================] - 0s - loss: 3.3146 - acc: 0.4630     \n",
      "Epoch 9/10\n",
      "1000/1000 [==============================] - 0s - loss: 3.6025 - acc: 0.4620     \n",
      "Epoch 10/10\n",
      "1000/1000 [==============================] - 0s - loss: 2.3228 - acc: 0.5060     \n",
      "CPU times: user 1.46 s, sys: 212 ms, total: 1.68 s\n",
      "Wall time: 17.8 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Dense(1,input_dim=784, activation='linear'))\n",
    "model.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=['accuracy'])\n",
    "\n",
    "data = np.random.random((1000,784))\n",
    "labels = np.random.randint(2,size=(1000,1))\n",
    "\n",
    "model.fit(data,labels,nb_epoch=10,batch_size=32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 2/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 3/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 4/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 5/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 6/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 7/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 8/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 9/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 10/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 11/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 12/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 13/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 14/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 15/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 16/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 17/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 18/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 19/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 20/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 21/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 22/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 23/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 24/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 25/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 26/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 27/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 28/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 29/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 30/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 31/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 32/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 33/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 34/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 35/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 36/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 37/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 38/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 39/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 40/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 41/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 42/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 43/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 44/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 45/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 46/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 47/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 48/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 49/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "Epoch 50/50\n",
      "1000/1000 [==============================] - 0s - loss: 7.8817 - acc: 0.5110     \n",
      "CPU times: user 3.9 s, sys: 612 ms, total: 4.52 s\n",
      "Wall time: 14.7 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Dense(1,input_dim=784, activation='relu'))\n",
    "model.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=['accuracy'])\n",
    "\n",
    "data = np.random.random((1000,784))\n",
    "labels = np.random.randint(2,size=(1000,1))\n",
    "\n",
    "model.fit(data,labels,nb_epoch=50,batch_size=32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.7277 - acc: 0.5130     \n",
      "Epoch 2/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.7125 - acc: 0.5300     \n",
      "Epoch 3/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.7040 - acc: 0.5030     \n",
      "Epoch 4/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.7012 - acc: 0.5450     \n",
      "Epoch 5/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6945 - acc: 0.5410     \n",
      "Epoch 6/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6837 - acc: 0.5570     \n",
      "Epoch 7/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6777 - acc: 0.5680     \n",
      "Epoch 8/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6756 - acc: 0.5870     \n",
      "Epoch 9/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6645 - acc: 0.6030     \n",
      "Epoch 10/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6560 - acc: 0.6110     \n",
      "CPU times: user 1.5 s, sys: 188 ms, total: 1.68 s\n",
      "Wall time: 11.3 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Dense(1,input_dim=784, activation='sigmoid'))\n",
    "model.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=['accuracy'])\n",
    "\n",
    "data = np.random.random((1000,784))\n",
    "labels = np.random.randint(2,size=(1000,1))\n",
    "\n",
    "model.fit(data,labels,nb_epoch=10,batch_size=32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.7120 - acc: 0.5130     \n",
      "Epoch 2/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6932 - acc: 0.5540     \n",
      "Epoch 3/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6892 - acc: 0.5550     \n",
      "Epoch 4/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6840 - acc: 0.5790     \n",
      "Epoch 5/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6763 - acc: 0.5750     \n",
      "Epoch 6/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6795 - acc: 0.5670     \n",
      "Epoch 7/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6643 - acc: 0.5960     \n",
      "Epoch 8/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6600 - acc: 0.5970     \n",
      "Epoch 9/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6530 - acc: 0.6080     \n",
      "Epoch 10/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6490 - acc: 0.6230     \n",
      "CPU times: user 1.54 s, sys: 220 ms, total: 1.76 s\n",
      "Wall time: 13.7 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Dense(1,input_dim=784, activation='hard_sigmoid'))\n",
    "model.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=['accuracy'])\n",
    "\n",
    "data = np.random.random((1000,784))\n",
    "labels = np.random.randint(2,size=(1000,1))\n",
    "\n",
    "model.fit(data,labels,nb_epoch=10,batch_size=32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.8269 - acc: 0.4830     \n",
      "Epoch 2/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.7506 - acc: 0.5540     \n",
      "Epoch 3/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6852 - acc: 0.6180     \n",
      "Epoch 4/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6625 - acc: 0.6510     \n",
      "Epoch 5/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6564 - acc: 0.6460     \n",
      "Epoch 6/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6392 - acc: 0.6700     \n",
      "Epoch 7/10\n",
      "1000/1000 [==============================] - 1s - loss: 0.6369 - acc: 0.6810     \n",
      "Epoch 8/10\n",
      "1000/1000 [==============================] - 1s - loss: 0.6046 - acc: 0.7180     \n",
      "Epoch 9/10\n",
      "1000/1000 [==============================] - 0s - loss: 0.6090 - acc: 0.7220     \n",
      "Epoch 10/10\n",
      "1000/1000 [==============================] - 1s - loss: 0.6131 - acc: 0.7120     \n",
      "CPU times: user 9.64 s, sys: 1.76 s, total: 11.4 s\n",
      "Wall time: 26.9 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "from keras.regularizers import l1,l2,l1l2, activity_l2\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Dense(1,input_dim=784, activation='sigmoid', W_regularizer=l2()))\n",
    "# model.add(Dense(1,input_dim=784, activation='sigmoid'))\n",
    "\n",
    "model.compile(loss='binary_crossentropy', optimizer='sgd', metrics=['accuracy'])\n",
    "\n",
    "data = np.random.random((1000,784))\n",
    "labels = np.random.randint(2,size=(1000,1))\n",
    "\n",
    "model.fit(data,labels,nb_epoch=10,batch_size=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1000, 784)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "labels.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0],\n",
       "       [0],\n",
       "       [1],\n",
       "       [1],\n",
       "       [0]])"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 [==============================] - 0s\n",
      "1/1 [==============================] - 0s\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(array([[ 0.13410421]], dtype=float32), array([[0]], dtype=int32))"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test = np.random.random(784).reshape(1,-1)\n",
    "proba = model.predict_proba(test)\n",
    "classes = model.predict_classes(test)\n",
    "proba,classes"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## merge layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}