Step6_bigramLanguageModel.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#This notebook applies a large-vocabulary bigram language model (50k words) implemented in Kaldi to the RNN outputs.\n",
    "#It also measures character/word error rates after the language model has been applied. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import scipy.io\n",
    "import os\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "#point this towards the top level dataset directory\n",
    "rootDir = os.path.expanduser('~') + '/handwritingBCIData/'\n",
    "\n",
    "#point this towards the code directory\n",
    "repoDir = os.getcwd() + '/'\n",
    "\n",
    "#defines which train/test partition to use\n",
    "cvPart = 'HeldOutTrials'\n",
    "\n",
    "#defines which datasets to process\n",
    "dataDirs = ['t5.2019.05.08','t5.2019.11.25','t5.2019.12.09','t5.2019.12.11','t5.2019.12.18',\n",
    "            't5.2019.12.20','t5.2020.01.06','t5.2020.01.08','t5.2020.01.13','t5.2020.01.15']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "t5.2019.05.08\n",
      "t5.2019.11.25\n",
      "t5.2019.12.09\n",
      "t5.2019.12.11\n",
      "t5.2019.12.18\n",
      "t5.2019.12.20\n",
      "t5.2020.01.06\n",
      "t5.2020.01.08\n",
      "t5.2020.01.13\n",
      "t5.2020.01.15\n"
     ]
    }
   ],
   "source": [
    "#First we have to convert the RNN outputs to a format that Kaldi can read. Here we make a Kaldi-readable matrix\n",
    "#for each sentence that contains the RNN outputs. \n",
    "from characterDefinitions import getHandwritingCharacterDefinitions\n",
    "from rnnEval import rnnOutputToKaldiMatrices\n",
    "\n",
    "charDef = getHandwritingCharacterDefinitions()\n",
    "allErrCounts = []\n",
    "outDir = rootDir+'RNNTrainingSteps/Step5_RNNInference/' + cvPart\n",
    "\n",
    "for x in range(len(dataDirs)):\n",
    "    print(dataDirs[x])\n",
    "    \n",
    "    outputs = scipy.io.loadmat(outDir + '/' + dataDirs[x] + '_inferenceOutputs.mat')\n",
    "    sentenceDat = scipy.io.loadmat(rootDir+'Datasets/'+dataDirs[x]+'/sentences.mat')\n",
    "    \n",
    "    #define where to save the Kaldi matrices\n",
    "    kaldiDir = rootDir+'RNNTrainingSteps/Step6_ApplyBigramLM/' + cvPart + '/KaldiMatrices/' + dataDirs[x] + '/'\n",
    "    if not os.path.isdir(kaldiDir):\n",
    "        os.mkdir(kaldiDir)\n",
    "\n",
    "    #This function converts the RNN output to kaldi matrices so the language model can process it.\n",
    "    #Note that as part of this process we use the RNN's character start signal output to create a fake 'CTC blank' signal\n",
    "    #for the language model (the language model was originally designed to be used with a CTC loss).\n",
    "    cProb = rnnOutputToKaldiMatrices(outputs['outputs'], \n",
    "                             sentenceDat['numTimeBinsPerSentence']/2+50, \n",
    "                             charDef, \n",
    "                             kaldiDir)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f1e7801cd30>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Visualize what a Kaldi matrix looks like. Note that the first row is the CTC-blank signal, which goes low only briefly whenever\n",
    "#there is a new character.\n",
    "plt.figure(figsize=(12,4))\n",
    "plt.imshow(cProb[122].T, aspect='auto', clim=[0,-15])\n",
    "plt.xlabel('Time Step')\n",
    "plt.ylabel('Character Log Probability\\n(31 characters + CTC blank)')\n",
    "plt.colorbar(fraction=0.046, pad=0.04)\n",
    "plt.title('Example Kaldi Probability Matrix')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "t5.2019.05.08\n",
      "t5.2019.11.25\n",
      "t5.2019.12.09\n",
      "t5.2019.12.11\n",
      "t5.2019.12.18\n",
      "t5.2019.12.20\n",
      "t5.2020.01.06\n",
      "t5.2020.01.08\n",
      "t5.2020.01.13\n",
      "t5.2020.01.15\n"
     ]
    }
   ],
   "source": [
    "#Now call kaldi to apply the language model to the probability matrices created above.\n",
    "#Note that you will need kaldi (https://github.com/kaldi-asr/kaldi) \n",
    "#AND you will need custom kaldi decoders from https://github.com/jpuigcerver/kaldi-decoders.\n",
    "from parallelBash import parallelBash\n",
    "import time\n",
    "import multiprocessing\n",
    "\n",
    "#Add kaldi and kaldi-decoders binaries to the path. This assumes that kaldi is located in the home directory.\n",
    "homeDir = os.path.expanduser('~/')\n",
    "os.environ['PATH'] += ':'+homeDir+'kaldi-decoders/bin'\n",
    "os.environ['PATH'] += ':'+homeDir+'kaldi/src/bin'\n",
    "os.environ['PATH'] += ':'+homeDir+'kaldi/src/lm'\n",
    "os.environ['PATH'] += ':'+homeDir+'kaldi/src/lmbin'\n",
    "os.environ['PATH'] += ':'+homeDir+'kaldi/src/fstbin'\n",
    "os.environ['PATH'] += ':'+homeDir+'kaldi/src/featbin'\n",
    "os.environ['PATH'] += ':'+homeDir+'kaldi/src/gmmbin'\n",
    "os.environ['PATH'] += ':'+homeDir+'kaldi/tools/openfst-1.6.7/bin'\n",
    "os.environ['PATH'] += ':'+homeDir+'kaldi/src/latbin'\n",
    "\n",
    "#make sure we have permission to execute all the scripts we'll be calling below\n",
    "os.system('chmod +x ' + repoDir+'kaldiLMScripts/bigramLmDecode.sh')\n",
    "os.system('chmod +x ' + repoDir+'kaldiLMScripts/parseOptions.inc.sh')\n",
    "os.system('chmod +x ' + repoDir+'kaldiLMScripts/remove_transcript_dummy_boundaries.sh')\n",
    "os.system('chmod +x ' + repoDir+'kaldiLMScripts/int2sym.pl')\n",
    "\n",
    "#for each dataset, we launch multiple instances of the kaldi decoder program in parallel, each operating on a single sentence\n",
    "for dataDir in dataDirs:\n",
    "    print(dataDir)\n",
    "\n",
    "    #where the language model files are stored\n",
    "    langModelDir = rootDir+'BigramLM'\n",
    "    \n",
    "    #where the kaldi probability matrices are saved (these are inputs for this step)\n",
    "    matsDir = rootDir+'RNNTrainingSteps/Step6_ApplyBigramLM/'+cvPart+'/KaldiMatrices/'+dataDir\n",
    "    \n",
    "    #where we should save the language model outputs (which are lists of candidate sentences, along with their scores)\n",
    "    outDir = rootDir+'RNNTrainingSteps/Step6_ApplyBigramLM/'+cvPart+'/KaldiOutput/'+dataDir\n",
    "    if not os.path.isdir(outDir):\n",
    "        os.mkdir(outDir)\n",
    "        \n",
    "    #clear previous LM outputs\n",
    "    os.system('rm ' + outDir + '/*')\n",
    "\n",
    "    #This next bit of code generates some temporary .sh files that help launch multiple decoding programs at once.\n",
    "    #we then launch the '_master.sh' file to kick everything off, and wait until the outputs are all there before continuing\n",
    "    #to the next dataset.\n",
    "    scriptFile = repoDir + 'kaldiLMScripts/bigramLmDecode.sh'\n",
    "    \n",
    "    if not os.path.isdir(rootDir + 'RNNTrainingSteps/bashScratch'):\n",
    "        os.mkdir(rootDir + 'RNNTrainingSteps/bashScratch')\n",
    "    bashFilePrefix = rootDir + 'RNNTrainingSteps/bashScratch/lmDecode'\n",
    "    \n",
    "    fileList = os.listdir(matsDir)\n",
    "    txtFiles = []\n",
    "    for f in fileList:\n",
    "        if f.endswith('.txt'):\n",
    "            txtFiles.append(f)\n",
    "            \n",
    "    nFiles = len(txtFiles)    \n",
    "    nParallelProcesses = int(multiprocessing.cpu_count()/2)\n",
    "\n",
    "    sentenceIdx = np.arange(0,nFiles).astype(np.int32)\n",
    "    \n",
    "    argList = []\n",
    "    for x in range(len(sentenceIdx)):\n",
    "        newArgs = {}\n",
    "        newArgs['acoustic_scale'] = 1.0 #1.0, 1.79\n",
    "        newArgs['beam'] = 65\n",
    "        newArgs['max_active'] = 5000 #500000 #5000000\n",
    "        newArgs['1_mainArg'] = langModelDir\n",
    "        newArgs['2_mainArg'] = matsDir + '/kaldiMat_'+str(sentenceIdx[x])+'.txt'\n",
    "        newArgs['3_mainArg'] = outDir + '/' + str(sentenceIdx[x]) + '_'\n",
    "\n",
    "        argList.append(newArgs)\n",
    "\n",
    "    #this utility function generates temporary .sh files for running parallel decoding\n",
    "    parallelBash(argList, scriptFile, bashFilePrefix, nParallelProcesses)\n",
    "        \n",
    "    #make sure we have permissions to execute everything\n",
    "    os.system('chmod +x ' + bashFilePrefix+'_master.sh')\n",
    "    for x in range(nParallelProcesses):\n",
    "        os.system('chmod +x ' + bashFilePrefix+'_'+str(x)+'.sh')\n",
    "            \n",
    "    #launching the master file starts everything\n",
    "    os.system(bashFilePrefix+'_master.sh')\n",
    "\n",
    "    #now we just sit and wait until we have all the outputs\n",
    "    numFilesInDir = 0\n",
    "    while numFilesInDir < nFiles*9:\n",
    "        numFilesInDir = int(len(os.listdir(outDir)))\n",
    "        time.sleep(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " --t5.2019.05.08-- \n",
      "#2\n",
      "True:    you want me to sing?\n",
      "Decoded: you want me to sing?\n",
      "\n",
      "#5\n",
      "True:    have you ever seen a large cat fold itself into a tiny shoe box?\n",
      "Decoded: have you ever seen a large cat fold itself into a tiny shoe box?\n",
      "\n",
      "#21\n",
      "True:    the jeep was thirsty so i stopped for gas on the edge of town.\n",
      "Decoded: the jeep was thirsty so i stopped for gas on the edge of town.\n",
      "\n",
      "#33\n",
      "True:    because less than a quarter mile from where i'm standing right now is where alla's body was dumped.\n",
      "Decoded: because less than a quarter mile from where i'm standing right now is where allah's body was dumped.\n",
      "\n",
      "#34\n",
      "True:    there are moments when gentle background singing brings the song close to gospel.\n",
      "Decoded: there are moments when gentle background singing brings the song close to gospel.\n",
      "\n",
      "#39\n",
      "True:    so you could literally see here's what every single person in the company gets paid.\n",
      "Decoded: so you could literally see here's what every single person in the company gets paid.\n",
      "\n",
      "#54\n",
      "True:    i told you we were going to bring this sucker home, and we brought it home.\n",
      "Decoded: i told you we were going to bring this sucker home, and we brought it home.\n",
      "\n",
      "#90\n",
      "True:    what rent do you pay for your housing?\n",
      "Decoded: what rent do you pay for your housing.\n",
      "\n",
      "#91\n",
      "True:    plainly, there is a clear and present need to secure information in the public and private sectors.\n",
      "Decoded: plainly, there is a clear and present need to secure information in the public and private sectors.\n",
      "\n",
      "#98\n",
      "True:    i thought i could fix it but got trapped by my ego and refusal to fail.\n",
      "Decoded: i thought i could fix it but got trapped by my ego and refusal to fail.\n",
      "\n",
      " --t5.2019.11.25-- \n",
      "#5\n",
      "True:    then it wobbled under her weight and tilted forwards.\n",
      "Decoded: then it wobble under her weight and tilted forwards.\n",
      "\n",
      "#21\n",
      "True:    and my brother's in hong kong all that week.\n",
      "Decoded: and my brother's in hong kong all that week.\n",
      "\n",
      "#31\n",
      "True:    then we slipped into the kitchen , got up on the sink, and hauled ourselves into the little attic.\n",
      "Decoded: then we slipped into the kitchen i got up on the sink, and braced ourselves into the little attic.\n",
      "\n",
      "#33\n",
      "True:    he didn't like being tied up to the women.\n",
      "Decoded: he didn't like being tied up to the women.\n",
      "\n",
      "#34\n",
      "True:    you can rent money.\n",
      "Decoded: you can rent money.\n",
      "\n",
      "#39\n",
      "True:    the only peace that you'll find here lies in a kind of pagan solitude.\n",
      "Decoded: the only peace that you'll find here lies in a kind of pagan solitude.\n",
      "\n",
      "#61\n",
      "True:    his pot was in his hand still, it preceded him like the torch of justice.\n",
      "Decoded: his pot was in his hand still, if preceded him like the torch of justice.\n",
      "\n",
      " --t5.2019.12.09-- \n",
      "#5\n",
      "True:    maybe get your horses to swim across.\n",
      "Decoded: may be get your horses to swim across.\n",
      "\n",
      "#21\n",
      "True:    i thought they were staying the night.\n",
      "Decoded: i thought they were staying the night.\n",
      "\n",
      "#31\n",
      "True:    it seemed a poor thing to have descended to.\n",
      "Decoded: it seemed a poor thing to have descended to.\n",
      "\n",
      "#33\n",
      "True:    his sudden talk of alibis made the whole thing sound serious.\n",
      "Decoded: this sudden talk of ali is made the whole thing sound serious.\n",
      "\n",
      "#34\n",
      "True:    is the policy still to sit it out and wait rather than try and force your way into the prison?\n",
      "Decoded: is the policy still to sit it out and wait rather than try and force your way into the prison?\n",
      "\n",
      "#39\n",
      "True:    i accept that it precludes class a development.\n",
      "Decoded: i accept that it precludes class a development.\n",
      "\n",
      "#68\n",
      "True:    at that moment she heard rapid footsteps.\n",
      "Decoded: at that moment she heard rapid footsteps.\n",
      "\n",
      " --t5.2019.12.11-- \n",
      "#5\n",
      "True:    but i have to hear the whole tale.\n",
      "Decoded: , but i have to hear the whole tale.\n",
      "\n",
      "#21\n",
      "True:    this fuzzy, probabilistic approach is a sensible one.\n",
      "Decoded: this fuzzy, probabilistic approach is a sensible one.\n",
      "\n",
      "#31\n",
      "True:    he needed to develop his mind.\n",
      "Decoded: he needed to develop his mind.\n",
      "\n",
      "#33\n",
      "True:    people here are also adopting a healthier diet.\n",
      "Decoded: people here are also adopting a healthier diet.\n",
      "\n",
      "#34\n",
      "True:    the royal navy had left its mark.\n",
      "Decoded: the royal navy had left its mark.\n",
      "\n",
      "#39\n",
      "True:    a long shower relaxed her still further.\n",
      "Decoded: a long shower relayed her still further.\n",
      "\n",
      "#61\n",
      "True:    thrown in nuts and carrots and blend.\n",
      "Decoded: thrown in nuts and carrots and blend.\n",
      "\n",
      " --t5.2019.12.18-- \n",
      "#5\n",
      "True:    he lifted his hand out of the water to smooth his hair.\n",
      "Decoded: he lifted his hand out of the water to smooth his hair.\n",
      "\n",
      "#21\n",
      "True:    the servants would know whether it was possible to get back down the valley.\n",
      "Decoded: the servants would know whether it was possible to get back down the valley.\n",
      "\n",
      "#31\n",
      "True:    why are you prepared to play such a desperate game, to put your whole future in jeopardy like this?\n",
      "Decoded: why are you prepared to play such a desperate game, to put your whole future in jeopardy like this?\n",
      "\n",
      "#33\n",
      "True:    you had round red cheeks, just like a wooden doll.\n",
      "Decoded: you had round red cheeks, just like a wooden doll.\n",
      "\n",
      "#34\n",
      "True:    i have seven apples, seven oranges and seven bananas.\n",
      "Decoded: i have severe apples, seven oranges and seven bananas.\n",
      "\n",
      "#39\n",
      "True:    grumpy wizards make a toxic brew for the jovial queen.\n",
      "Decoded: grumpy wizards make a toxic brew for the jovial queen.\n",
      "\n",
      "#61\n",
      "True:    he reckons his success on the field depends on learning the lingo.\n",
      "Decoded: he reckons his success on the field depends on learning the lingo.\n",
      "\n",
      " --t5.2019.12.20-- \n",
      "#5\n",
      "True:    i'm not tough enough to toe the line.\n",
      "Decoded: i'm not tough enough to toe the line.\n",
      "\n",
      "#21\n",
      "True:    a lot of old folk are getting caught and we must stamp it out.\n",
      "Decoded: a lot of old folk are getting caught and we must stamp it out.\n",
      "\n",
      "#31\n",
      "True:    these agents are now being tested in clinical trials with people at risk.\n",
      "Decoded: these agents are now being tested in clinical trials with people at risk.\n",
      "\n",
      "#33\n",
      "True:    the way he said it made the name sound like orchids and honey.\n",
      "Decoded: the way he said it made the name sound like orchids and honey.\n",
      "\n",
      "#34\n",
      "True:    pack my box with five dozen liquor jugs.\n",
      "Decoded: pack my box with five dozen liquor jugs.\n",
      "\n",
      "#39\n",
      "True:    state officials are always under threat, always being kidnapped.\n",
      "Decoded: state officials are always under threat, always being kidnapped.\n",
      "\n",
      "#61\n",
      "True:    that's when i threw up on the carpet.\n",
      "Decoded: that's when i threw up on the carpet.\n",
      "\n",
      " --t5.2020.01.06-- \n",
      "#5\n",
      "True:    they breathe air and they cannot swim.\n",
      "Decoded: they breathe air and they can not swim.\n",
      "\n",
      "#21\n",
      "True:    from the terrace, a broad flight of central steps led down to lawns and formal flowerbeds.\n",
      "Decoded: from the terrace, a broad flight of central steps led down to lawns and formal flower beds.\n",
      "\n",
      "#31\n",
      "True:    the corpses lay uncovered in the open on plastic sheets.\n",
      "Decoded: the corpses lay uncovered in the open on plastic sheets.\n",
      "\n",
      "#39\n",
      "True:    shame on us for not caring.\n",
      "Decoded: shame on us for not caring.\n",
      "\n",
      "#61\n",
      "True:    the other battalions were infected by the panic and also fled.\n",
      "Decoded: the other battalions were infected by the panic and also feed.\n",
      "\n",
      " --t5.2020.01.08-- \n",
      "#5\n",
      "True:    on the eve of the council the worst of horrors was revealed.\n",
      "Decoded: on the eve of the council the worst of horrors was revealed.\n",
      "\n",
      "#21\n",
      "True:    a small coal fire burned in the grate.\n",
      "Decoded: a small coal fire burned in the great.\n",
      "\n",
      "#31\n",
      "True:    the result is a lighter weight fabric, but one which still shows off the yarn.\n",
      "Decoded: the result is a lighter weight fabric, but one which still shows off the yarn.\n",
      "\n",
      "#33\n",
      "True:    perhaps they would accept his old currency.\n",
      "Decoded: perhaps they would accept his old currency.\n",
      "\n",
      "#34\n",
      "True:    already the grass was becoming crisp with frost.\n",
      "Decoded: already the grass was becoming crisp with frost.\n",
      "\n",
      "#39\n",
      "True:    it is the necessary foundation for growth of small businesses.\n",
      "Decoded: it is the necessary foundation for growth of small businesses.\n",
      "\n",
      "#61\n",
      "True:    celia constantly refers to her mother's drowning accident.\n",
      "Decoded: celia constantly referred to her mother's drowning accident.\n",
      "\n",
      " --t5.2020.01.13-- \n",
      "#22\n",
      "True:    crazy frederick bought many very exquisite opal jewels.\n",
      "Decoded: crazy frederick brought many very exquisite opal jewels.\n",
      "\n",
      "#25\n",
      "True:    sphinx of black quartz, judge my vow.\n",
      "Decoded: sphinx of black quartz, judge my vow.\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "#121\n",
      "True:    residents were warned to prepare flood defences.\n",
      "Decoded: residents were warned to prepare flood defences.\n",
      "\n",
      " --t5.2020.01.15-- \n",
      "#22\n",
      "True:    imagine a star with a mass ten times that of the sun.\n",
      "Decoded: imagine a star with a mass ten times that of the sun.\n",
      "\n",
      "#25\n",
      "True:    the elder boys are not deterred, however.\n",
      "Decoded: the older boys are not deterred, however.\n",
      "\n",
      "#121\n",
      "True:    you never used to swear, you know.\n",
      "Decoded: you never used to swear you know.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "#Now that we have all the language model outputs, load them up and compute character/word error rates\n",
    "from kaldiReadWrite import readKaldiLatticeFile, readKaldiAliFile\n",
    "import numpy as np\n",
    "import scipy.io\n",
    "from rnnEval import wer\n",
    "import warnings\n",
    "\n",
    "#this stops scipy.io.savemat from throwing a warning about empty entries\n",
    "warnings.simplefilter(action='ignore', category=FutureWarning)\n",
    "\n",
    "#keep track of the error counts for all validation sentences in 'valErrCounts' so we can summarize in the next cell\n",
    "valErrCounts = []\n",
    "\n",
    "for dataDir in dataDirs:\n",
    "    #process ALL sentences from this dataset (both train & test)\n",
    "    print(' --' + dataDir + '-- ')\n",
    "\n",
    "    sentenceDat = scipy.io.loadmat(rootDir+'Datasets/'+dataDir+'/sentences.mat')\n",
    "    cvPartFile = scipy.io.loadmat(rootDir+'RNNTrainingSteps/trainTestPartitions_'+cvPart+'.mat')\n",
    "    valIdx = cvPartFile[dataDir+'_test']\n",
    "    \n",
    "    allErrCounts = []\n",
    "    decSentences = []\n",
    "    \n",
    "    kaldiDir = rootDir+'RNNTrainingSteps/Step6_ApplyBigramLM/'+cvPart+'/KaldiOutput/'+dataDir \n",
    "    nFiles = int(len(os.listdir(kaldiDir))/9)   \n",
    "    \n",
    "    for fileIdx in range(nFiles):\n",
    "        nbestFile = kaldiDir+'/'+str(fileIdx)+'_transcript.txt'\n",
    "        acFile = kaldiDir+'/'+str(fileIdx)+'_best_acscore.ark'\n",
    "        lmFile = kaldiDir+'/'+str(fileIdx)+'_best_lmscore.ark'\n",
    "\n",
    "        nums, content = readKaldiLatticeFile(nbestFile, 'string')\n",
    "        _, acScore = readKaldiLatticeFile(acFile, 'numeric')\n",
    "        _, lmScore = readKaldiLatticeFile(lmFile, 'numeric')\n",
    "\n",
    "        #here we select the best-scoring sentence according to the formula 'acScore + 2.0*lmScore'\n",
    "        bestIdx = np.argmin(acScore + 2.0*lmScore)\n",
    "        decSent = content[bestIdx]\n",
    "        decSentences.append(decSent)\n",
    "        \n",
    "        trueText = sentenceDat['sentencePrompt'][fileIdx,0][0]\n",
    "        trueText = trueText.replace('>',' ')\n",
    "        trueText = trueText.replace('~','.')\n",
    "        trueText = trueText.replace('#','')\n",
    "        \n",
    "        #compute character/word error counts\n",
    "        charErrs = wer(trueText, decSent)\n",
    "        wordErrs = wer(trueText.split(), decSent.split())\n",
    "        allErrCounts.append(np.array([charErrs, len(trueText), wordErrs, len(trueText.split())]))\n",
    "\n",
    "        #print the language model outputs for all sentences in the held-out set\n",
    "        if fileIdx in valIdx:\n",
    "            print('#' + str(fileIdx))\n",
    "            print('True:    ' + trueText)\n",
    "            print('Decoded: ' + decSent)\n",
    "            print('')\n",
    "            \n",
    "            valErrCounts.append(np.array([charErrs, len(trueText), wordErrs, len(trueText.split())]))\n",
    "\n",
    "    #save error rates & decoded sentences for this dataset\n",
    "    concatCounts = np.stack(allErrCounts, axis=0)\n",
    "\n",
    "    saveDict = {}\n",
    "    saveDict['decSentences'] = decSentences\n",
    "    saveDict['trueSentences'] = sentenceDat['sentencePrompt']\n",
    "    saveDict['charCounts'] = concatCounts[:,1]\n",
    "    saveDict['charErrors'] = concatCounts[:,0]\n",
    "    saveDict['wordCounts'] = concatCounts[:,3]\n",
    "    saveDict['wordErrors'] = concatCounts[:,2]\n",
    "\n",
    "    scipy.io.savemat(rootDir + 'RNNTrainingSteps/Step6_ApplyBigramLM/' + cvPart + '/' + dataDir + '_errCounts.mat', saveDict)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Character error rate: 0.80%\n",
      "Word error rate: 3.64%\n"
     ]
    }
   ],
   "source": [
    "#Summarize character error rate and word error rate across all held-out trials from all datasets\n",
    "concatErrCounts = np.squeeze(np.stack(valErrCounts, axis=0))\n",
    "cer = 100*(np.sum(concatErrCounts[:,0]) / np.sum(concatErrCounts[:,1]))\n",
    "wer = 100*(np.sum(concatErrCounts[:,2]) / np.sum(concatErrCounts[:,3]))\n",
    "\n",
    "print('Character error rate: %1.2f%%' % float(cer))\n",
    "print('Word error rate: %1.2f%%' % float(wer))"
   ]
  },
  {
   "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": "3.6.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}