use mxnet.nd.gather_nd to create matrix for fertility bias (Cohn et al, 2016)

fertility bias window construction.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 470,
   "metadata": {},
   "outputs": [],
   "source": [
    "import mxnet as mx\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 471,
   "metadata": {},
   "outputs": [],
   "source": [
    "k=1\n",
    "n=4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 472,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_indices(k, n):\n",
    "    index_center = mx.nd.zeros((2*k+1, n))\n",
    "    zeros = mx.nd.zeros((n, 2*k+1))\n",
    "    for i in range(0,2*k+1):\n",
    "        value = np.arange(i, i+n)\n",
    "        index_center[i][:] = value[:]\n",
    "    index_center= index_center.transpose()\n",
    "    return mx.nd.stack(zeros, index_center, zeros)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 473,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2, 4, 1)"
      ]
     },
     "execution_count": 473,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "attention = mx.ndarray.array([[[1], [2], [3], [4]],[[5], [6], [7], [8]]])\n",
    "attention.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 474,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "\n",
       "[[[0.]\n",
       "  [1.]\n",
       "  [2.]\n",
       "  [3.]\n",
       "  [4.]\n",
       "  [0.]]\n",
       "\n",
       " [[0.]\n",
       "  [5.]\n",
       "  [6.]\n",
       "  [7.]\n",
       "  [8.]\n",
       "  [0.]]]\n",
       "<NDArray 2x6x1 @cpu(0)>"
      ]
     },
     "execution_count": 474,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# mxnet's padding only works for 4- and 5-dimensional data\n",
    "attention_padded = np.pad(attention.asnumpy(), [(0,0), (k,k), (0,0)], \"constant\")\n",
    "attention_padded = mx.nd.array(attention_padded)\n",
    "attention_padded "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 475,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "\n",
       "[[0. 1. 2.]\n",
       " [1. 2. 3.]\n",
       " [2. 3. 4.]\n",
       " [3. 4. 0.]]\n",
       "<NDArray 4x3 @cpu(0)>"
      ]
     },
     "execution_count": 475,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mx.ndarray.gather_nd(attention_padded, create_indices(k,n))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}