a.ipnb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/name/anaconda2/lib/python2.7/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from __future__ import print_function\n",
    "\n",
    "import keras\n",
    "from keras.models import Sequential\n",
    "from keras.models import Model\n",
    "from keras.layers import Input\n",
    "from keras.layers import Dense\n",
    "from keras.layers import LSTM\n",
    "from keras.layers import GRU\n",
    "from keras.layers import SimpleRNN\n",
    "from keras.layers import Dropout\n",
    "from keras.layers import concatenate\n",
    "from keras import losses\n",
    "from keras import regularizers\n",
    "from keras.constraints import min_max_norm\n",
    "import h5py\n",
    "\n",
    "from keras.constraints import Constraint\n",
    "from keras import backend as K\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def my_crossentropy(y_true, y_pred):\n",
    "    return K.mean(2*K.abs(y_true-0.5) * K.binary_crossentropy(y_pred, y_true), axis=-1)\n",
    "\n",
    "def mymask(y_true):\n",
    "    return K.minimum(y_true+1., 1.)\n",
    "\n",
    "def msse(y_true, y_pred):\n",
    "    return K.mean(mymask(y_true) * K.square(K.sqrt(y_pred) - K.sqrt(y_true)), axis=-1)\n",
    "\n",
    "def mycost(y_true, y_pred):\n",
    "    return K.mean(mymask(y_true) * (10*K.square(K.square(K.sqrt(y_pred) - K.sqrt(y_true))) + K.square(K.sqrt(y_pred) - K.sqrt(y_true)) + 0.01*K.binary_crossentropy(y_pred, y_true)), axis=-1)\n",
    "\n",
    "def my_accuracy(y_true, y_pred):\n",
    "    return K.mean(2*K.abs(y_true-0.5) * K.equal(y_true, K.round(y_pred)), axis=-1)\n",
    "\n",
    "class WeightClip(Constraint):\n",
    "    '''Clips the weights incident to each hidden unit to be inside a range\n",
    "    '''\n",
    "    def __init__(self, c=2):\n",
    "        self.c = c\n",
    "\n",
    "    def __call__(self, p):\n",
    "        return K.clip(p, -self.c, self.c)\n",
    "\n",
    "    def get_config(self):\n",
    "        return {'name': self.__class__.__name__,\n",
    "            'c': self.c}\n",
    "\n",
    "reg = 0.000001\n",
    "constraint = WeightClip(0.499)"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "main_input = Input(shape=(None, 42), name='main_input')\n",
    "tmp = Dense(24, activation='tanh', name='input_dense', kernel_constraint=constraint, bias_constraint=constraint)(main_input)\n",
    "vad_gru = GRU(24, activation='tanh', recurrent_activation='sigmoid', return_sequences=True, name='vad_gru', kernel_regularizer=regularizers.l2(reg), recurrent_regularizer=regularizers.l2(reg), kernel_constraint=constraint, recurrent_constraint=constraint, bias_constraint=constraint)(tmp)\n",
    "vad_output = Dense(1, activation='sigmoid', name='vad_output', kernel_constraint=constraint, bias_constraint=constraint)(vad_gru)\n",
    "noise_input = keras.layers.concatenate([tmp, vad_gru, main_input])\n",
    "noise_gru = GRU(48, activation='relu', recurrent_activation='sigmoid', return_sequences=True, name='noise_gru', kernel_regularizer=regularizers.l2(reg), recurrent_regularizer=regularizers.l2(reg), kernel_constraint=constraint, recurrent_constraint=constraint, bias_constraint=constraint)(noise_input)\n",
    "denoise_input = keras.layers.concatenate([vad_gru, noise_gru, main_input])\n",
    "\n",
    "denoise_gru = GRU(96, activation='tanh', recurrent_activation='sigmoid', return_sequences=True, name='denoise_gru', kernel_regularizer=regularizers.l2(reg), recurrent_regularizer=regularizers.l2(reg), kernel_constraint=constraint, recurrent_constraint=constraint, bias_constraint=constraint)(denoise_input)\n",
    "\n",
    "denoise_output = Dense(22, activation='sigmoid', name='denoise_output', kernel_constraint=constraint, bias_constraint=constraint)(denoise_gru)\n",
    "\n",
    "model = Model(inputs=main_input, outputs=[denoise_output, vad_output])\n",
    "\n",
    "model.compile(loss=[mycost, my_crossentropy],\n",
    "              metrics=[msse],\n",
    "              optimizer='adam', loss_weights=[10, 0.5])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "main_input (InputLayer)         (None, None, 42)     0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_dense (Dense)             (None, None, 24)     1032        main_input[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "vad_gru (GRU)                   (None, None, 24)     3528        input_dense[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_1 (Concatenate)     (None, None, 90)     0           input_dense[0][0]                \n",
      "                                                                 vad_gru[0][0]                    \n",
      "                                                                 main_input[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "noise_gru (GRU)                 (None, None, 48)     20016       concatenate_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_2 (Concatenate)     (None, None, 114)    0           vad_gru[0][0]                    \n",
      "                                                                 noise_gru[0][0]                  \n",
      "                                                                 main_input[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "denoise_gru (GRU)               (None, None, 96)     60768       concatenate_2[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "denoise_output (Dense)          (None, None, 22)     2134        denoise_gru[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "vad_output (Dense)              (None, None, 1)      25          vad_gru[0][0]                    \n",
      "==================================================================================================\n",
      "Total params: 87,503\n",
      "Trainable params: 87,503\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  }
 ],
 "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": "2.7.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}