dlnd_tv_script_generation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# TV Script Generation\n",
    "In this project, you'll generate your own [Simpsons](https://en.wikipedia.org/wiki/The_Simpsons) TV scripts using RNNs.  You'll be using part of the [Simpsons dataset](https://www.kaggle.com/wcukierski/the-simpsons-by-the-data) of scripts from 27 seasons.  The Neural Network you'll build will generate a new TV script for a scene at [Moe's Tavern](https://simpsonswiki.com/wiki/Moe's_Tavern).\n",
    "## Get the Data\n",
    "The data is already provided for you.  You'll be using a subset of the original dataset.  It consists of only the scenes in Moe's Tavern.  This doesn't include other versions of the tavern, like \"Moe's Cavern\", \"Flaming Moe's\", \"Uncle Moe's Family Feed-Bag\", etc.."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL\n",
    "\"\"\"\n",
    "import helper\n",
    "\n",
    "data_dir = './data/simpsons/moes_tavern_lines.txt'\n",
    "text = helper.load_data(data_dir)\n",
    "# Ignore notice, since we don't use it for analysing the data\n",
    "text = text[81:]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Explore the Data\n",
    "Play around with `view_sentence_range` to view different parts of the data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset Stats\n",
      "Roughly the number of unique words: 11492\n",
      "Number of scenes: 262\n",
      "Average number of sentences in each scene: 15.251908396946565\n",
      "Number of lines: 4258\n",
      "Average number of words in each line: 11.50164396430249\n",
      "\n",
      "The sentences 30 to 100:\n",
      "Moe_Szyslak: What's the matter, Homer? The depressin' effects of alcohol usually don't kick in 'til closing time.\n",
      "Lisa_Simpson: He's just a little nervous. (PROUDLY) He has to give a speech tomorrow on \"How To Keep Cool In A Crisis.\"\n",
      "Homer_Simpson: (SOBS) What am I gonna do? What am I gonna do?\n",
      "Barney_Gumble: Hey, I had to give a speech once. I was pretty nervous, so I used a little trick. I pictured everyone in their underwear. The judge, the jury, my lawyer, everybody.\n",
      "Homer_Simpson: Did it work?\n",
      "Barney_Gumble: I'm a free man, ain't I?\n",
      "Barney_Gumble: Whoa!\n",
      "Barney_Gumble: Huh? A pretzel? Wow, looks like I pulled a Homer!\n",
      "\n",
      "\n",
      "Patrons: (MUMBLING, NOT IN UNISON) Happy thoughts... happy thoughts... we love that boy.\n",
      "Moe_Szyslak: (INTO PHONE) Moe's Tavern. Hold on, I'll check. ... (LOUD) Hey everybody! I'm a stupid moron with an ugly face and a big butt, and my butt smells, and I like to kiss my own butt.\n",
      "Barney_Gumble: That's a new one (LAUGHING).\n",
      "Moe_Szyslak: Now wait a minute...\n",
      "\n",
      "\n",
      "Homer_Simpson: Hurry, Moe, hurry! I've only got five minutes till the music store closes.\n",
      "Moe_Szyslak: Why don't you go there first?\n",
      "Homer_Simpson: Hey, do I tell you how to do your job?\n",
      "Moe_Szyslak: Sorry, Homer.\n",
      "Homer_Simpson: You know, if you tip the glass, there won't be so much foam on top.\n",
      "Moe_Szyslak: Sorry, Homer.\n",
      "Homer_Simpson: (LOOKING AT WATCH) Ah. Finished with fifteen seconds to spare.\n",
      "Little_Man: (CONCERNED) What's the matter, buddy?\n",
      "Homer_Simpson: The moron next door closed early!\n",
      "Little_Man: (STIFFENING) I happen to be that moron.\n",
      "Homer_Simpson: Oh, me and my trenchant mouth.\n",
      "Homer_Simpson: Please, you've got to open that store.\n",
      "Little_Man: Let me think about it... Eh... No.\n",
      "Homer_Simpson: Okay, okay. But I want you to see a picture of the little girl you're disappointing. (GOES THROUGH HIS WALLET) Well I don't have one.\n",
      "Moe_Szyslak: (TO LITTLE MAN) Come on, Jer. Open up. Be a pal. Remember when I pulled you and your wife out of that burning car?\n",
      "Little_Man: (GRUDGINGLY) Okay. Okay. But now we're even. (TO HOMER) So what does your daughter need?\n",
      "Homer_Simpson: (SMOOTHLY) I'll have you know, I wrote it down.\n",
      "Homer_Simpson: Number Four and a half -- Stupid gum!\n",
      "Homer_Simpson: Number Four and a Half reed! Whoo hoo!\n",
      "Little_Man: Uh-huh. And what instrument does she play?\n",
      "Homer_Simpson: (SUNK) I dunno.\n",
      "\n",
      "\n",
      "Moe_Szyslak: (TO PATRONS) Figure of speech.\n",
      "Moe_Szyslak: Hiya, Homer. (SIGHS)\n",
      "Homer_Simpson: What's the matter, Moe?\n",
      "Moe_Szyslak: Ah, business is slow. People today are healthier and drinking less. You know, if it wasn't for the Junior High school next door no one would even use the cigarette machine.\n",
      "Homer_Simpson: (MOUTH FULL) Yeah, things are tough all over.\n",
      "Moe_Szyslak: Increased job satisfaction and family togetherness are poison for a purveyor of mind-numbing intoxicants like myself.\n",
      "Homer_Simpson: Could I get a beer?\n",
      "Moe_Szyslak: Uh, yeah, sure.\n",
      "Moe_Szyslak: Oh sorry, I forgot we're out of beer.\n",
      "Moe_Szyslak: Yeah, I know, I got behind on my beer payments. The distributor cut me off and I spent my last ten grand on the \"Love Tester\".\n",
      "Moe_Szyslak: You're too late, Homer. Barney sucked it dry. Cut his gums up pretty bad.\n",
      "Moe_Szyslak: Take it easy, Homer. I learned how to make other drinks at Bartender's School.\n",
      "Moe_Szyslak: (UNFAMILIAR) Gin and... tonic? Do they mix?\n",
      "Homer_Simpson: (BRIGHTENING) Hey, I know a good drink. Really hits the spot. I invented it myself...\n",
      "Moe_Szyslak: Sorry, Harv.\n",
      "Moe_Szyslak: Whoa, sounds like one hell of a drink. What do you call it?\n",
      "Homer_Simpson: A \"Flaming Homer\".\n",
      "Moe_Szyslak: Okay, why don't you make us up a couple of \"Flaming Homers\"?\n",
      "Homer_Simpson: Hey Moe, you got any cough syrup?\n",
      "Moe_Szyslak: Uh, let me check the lost and found.\n",
      "Moe_Szyslak: What do we got here, Bowie knife, troll doll, glass eye...\n",
      "Moe_Szyslak: Oh. Here we are.\n",
      "Moe_Szyslak: It's not without its charm.\n",
      "Homer_Simpson: Try lighting it on fire.\n",
      "Moe_Szyslak: (SMILING) Whoa! Homer, it's like there's a party in my mouth and everyone's invited.\n",
      "Larry: Hey, your Love Tester's busted. I want my nickel back. (COUGHS)\n",
      "Moe_Szyslak: Hey, buddy. Have one on the house.\n",
      "Larry: Hey, hey, this drink is delicious! And my phlegm feels looser. What do you call it?\n",
      "Homer_Simpson: Well, it's called a \"Flaming...\n",
      "Moe_Szyslak: Moe! It's called a \"Flaming Moe\"! That's right, a \"Flaming Moe\". My name is Moe, and I invented it. That's why it's called a Flaming Moe. What? What are you lookin' at, Homer? It's a Flaming Moe I'm Moe.\n",
      "Barney_Gumble: Hey, what's this?\n"
     ]
    }
   ],
   "source": [
    "view_sentence_range = (30, 100)\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL\n",
    "\"\"\"\n",
    "import numpy as np\n",
    "\n",
    "print('Dataset Stats')\n",
    "print('Roughly the number of unique words: {}'.format(len({word: None for word in text.split()})))\n",
    "scenes = text.split('\\n\\n')\n",
    "print('Number of scenes: {}'.format(len(scenes)))\n",
    "sentence_count_scene = [scene.count('\\n') for scene in scenes]\n",
    "print('Average number of sentences in each scene: {}'.format(np.average(sentence_count_scene)))\n",
    "\n",
    "sentences = [sentence for scene in scenes for sentence in scene.split('\\n')]\n",
    "print('Number of lines: {}'.format(len(sentences)))\n",
    "word_count_sentence = [len(sentence.split()) for sentence in sentences]\n",
    "print('Average number of words in each line: {}'.format(np.average(word_count_sentence)))\n",
    "\n",
    "print()\n",
    "print('The sentences {} to {}:'.format(*view_sentence_range))\n",
    "print('\\n'.join(text.split('\\n')[view_sentence_range[0]:view_sentence_range[1]]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Implement Preprocessing Functions\n",
    "The first thing to do to any dataset is preprocessing.  Implement the following preprocessing functions below:\n",
    "- Lookup Table\n",
    "- Tokenize Punctuation\n",
    "\n",
    "### Lookup Table\n",
    "To create a word embedding, you first need to transform the words to ids.  In this function, create two dictionaries:\n",
    "- Dictionary to go from the words to an id, we'll call `vocab_to_int`\n",
    "- Dictionary to go from the id to word, we'll call `int_to_vocab`\n",
    "\n",
    "Return these dictionaries in the following tuple `(vocab_to_int, int_to_vocab)`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tests Passed\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import problem_unittests as tests\n",
    "from collections import Counter\n",
    "\n",
    "def create_lookup_tables(text):\n",
    "    \"\"\"\n",
    "    Create lookup tables for vocabulary\n",
    "    :param text: The text of tv scripts split into words\n",
    "    :return: A tuple of dicts (vocab_to_int, int_to_vocab)\n",
    "    \"\"\"\n",
    "    #We use counter object like skip-gram exercise\n",
    "    word_counts = Counter(text)\n",
    "    #Words sorted by the counter number helps if you want to subsample\n",
    "    sorted_vocab = sorted(word_counts, key=word_counts.get, reverse=True)\n",
    "    int_to_vocab = {ii: word for ii, word in enumerate(sorted_vocab)}\n",
    "    vocab_to_int = {word: ii for ii, word in int_to_vocab.items()}\n",
    "    return vocab_to_int, int_to_vocab\n",
    "\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "tests.test_create_lookup_tables(create_lookup_tables)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Tokenize Punctuation\n",
    "We'll be splitting the script into a word array using spaces as delimiters.  However, punctuations like periods and exclamation marks make it hard for the neural network to distinguish between the word \"bye\" and \"bye!\".\n",
    "\n",
    "Implement the function `token_lookup` to return a dict that will be used to tokenize symbols like \"!\" into \"||Exclamation_Mark||\".  Create a dictionary for the following symbols where the symbol is the key and value is the token:\n",
    "- Period ( . )\n",
    "- Comma ( , )\n",
    "- Quotation Mark ( \" )\n",
    "- Semicolon ( ; )\n",
    "- Exclamation mark ( ! )\n",
    "- Question mark ( ? )\n",
    "- Left Parentheses ( ( )\n",
    "- Right Parentheses ( ) )\n",
    "- Dash ( -- )\n",
    "- Return ( \\n )\n",
    "\n",
    "This dictionary will be used to token the symbols and add the delimiter (space) around it.  This separates the symbols as it's own word, making it easier for the neural network to predict on the next word. Make sure you don't use a token that could be confused as a word. Instead of using the token \"dash\", try using something like \"||dash||\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tests Passed\n"
     ]
    }
   ],
   "source": [
    "def token_lookup():\n",
    "    \"\"\"\n",
    "    Generate a dict to turn punctuation into a token.\n",
    "    :return: Tokenize dictionary where the key is the punctuation and the value is the token\n",
    "    \"\"\"\n",
    "    d = {'.':'||Period||', ',': '||Comma||' , '\"': '||Quotation_Mark||', ';':'||Semicolon||', '!' : '||Exclamation_mark||' , '?': '||Question_mark||'  ,')': '||Right_Parentheses||' ,'(' :'||Left_Parentheses||' ,'--' : '||Dash||' ,\"\\n\" : '||Return||'}\n",
    "    return d\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "tests.test_tokenize(token_lookup)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preprocess all the data and save it\n",
    "Running the code cell below will preprocess all the data and save it to file."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL\n",
    "\"\"\"\n",
    "# Preprocess Training, Validation, and Testing Data\n",
    "helper.preprocess_and_save_data(data_dir, token_lookup, create_lookup_tables)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Check Point\n",
    "This is your first checkpoint. If you ever decide to come back to this notebook or have to restart the notebook, you can start from here. The preprocessed data has been saved to disk."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL\n",
    "\"\"\"\n",
    "import helper\n",
    "import numpy as np\n",
    "import problem_unittests as tests\n",
    "\n",
    "int_text, vocab_to_int, int_to_vocab, token_dict = helper.load_preprocess()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Build the Neural Network\n",
    "You'll build the components necessary to build a RNN by implementing the following functions below:\n",
    "- get_inputs\n",
    "- get_init_cell\n",
    "- get_embed\n",
    "- build_rnn\n",
    "- build_nn\n",
    "- get_batches\n",
    "\n",
    "### Check the Version of TensorFlow and Access to GPU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "TensorFlow Version: 1.1.0\n",
      "Default GPU Device: /gpu:0\n"
     ]
    }
   ],
   "source": [
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL\n",
    "\"\"\"\n",
    "from distutils.version import LooseVersion\n",
    "import warnings\n",
    "import tensorflow as tf\n",
    "\n",
    "# Check TensorFlow Version\n",
    "assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer'\n",
    "print('TensorFlow Version: {}'.format(tf.__version__))\n",
    "\n",
    "# Check for a GPU\n",
    "if not tf.test.gpu_device_name():\n",
    "    warnings.warn('No GPU found. Please use a GPU to train your neural network.')\n",
    "else:\n",
    "    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Input\n",
    "Implement the `get_inputs()` function to create TF Placeholders for the Neural Network.  It should create the following placeholders:\n",
    "- Input text placeholder named \"input\" using the [TF Placeholder](https://www.tensorflow.org/api_docs/python/tf/placeholder) `name` parameter.\n",
    "- Targets placeholder\n",
    "- Learning Rate placeholder\n",
    "\n",
    "Return the placeholders in the following tuple `(Input, Targets, LearningRate)`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tests Passed\n"
     ]
    }
   ],
   "source": [
    "def get_inputs():\n",
    "    \"\"\"\n",
    "    Create TF Placeholders for input, targets, and learning rate.\n",
    "    :return: Tuple (input, targets, learning rate)\n",
    "    \"\"\"\n",
    "    # TODO: Implement Function\n",
    "    input = tf.placeholder(tf.int32,(None,None), name='input')\n",
    "    targets = tf.placeholder(tf.int32,(None,None), name='targets')\n",
    "    learning_rate = tf.placeholder(tf.float32, name='learning_rate')\n",
    "    return input, targets, learning_rate\n",
    "\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "tests.test_get_inputs(get_inputs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Build RNN Cell and Initialize\n",
    "Stack one or more [`BasicLSTMCells`](https://www.tensorflow.org/api_docs/python/tf/contrib/rnn/BasicLSTMCell) in a [`MultiRNNCell`](https://www.tensorflow.org/api_docs/python/tf/contrib/rnn/MultiRNNCell).\n",
    "- The Rnn size should be set using `rnn_size`\n",
    "- Initalize Cell State using the MultiRNNCell's [`zero_state()`](https://www.tensorflow.org/api_docs/python/tf/contrib/rnn/MultiRNNCell#zero_state) function\n",
    "    - Apply the name \"initial_state\" to the initial state using [`tf.identity()`](https://www.tensorflow.org/api_docs/python/tf/identity)\n",
    "\n",
    "Return the cell and initial state in the following tuple `(Cell, InitialState)`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tests Passed\n"
     ]
    }
   ],
   "source": [
    "def get_init_cell(batch_size, rnn_size):\n",
    "    \"\"\"\n",
    "    Create an RNN Cell and initialize it.\n",
    "    :param batch_size: Size of batches\n",
    "    :param rnn_size: Size of RNNs\n",
    "    :return: Tuple (cell, initialize state)\n",
    "    \"\"\"\n",
    "\n",
    "    #Starting May 17' TF version 1.1 you have to individualy initialize every individual lstm cell \n",
    "    #You know what to expect from contrib! :) \n",
    "    \n",
    "    num_layers = 1\n",
    "   \n",
    "    #This line is like a tongue twister, initializes a RNN basic cells stack of num_layers size\n",
    "    cell = tf.contrib.rnn.MultiRNNCell([tf.contrib.rnn.BasicLSTMCell(rnn_size) for _ in range(num_layers)])\n",
    "    #Initial state and naming the vector according to specs\n",
    "    initial_state = tf.identity(cell.zero_state(batch_size, tf.float32),name='initial_state')\n",
    "    return cell, initial_state\n",
    "\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "tests.test_get_init_cell(get_init_cell)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Word Embedding\n",
    "Apply embedding to `input_data` using TensorFlow.  Return the embedded sequence."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tests Passed\n"
     ]
    }
   ],
   "source": [
    "def get_embed(input_data, vocab_size, embed_dim):\n",
    "    \"\"\"\n",
    "    Create embedding for <input_data>.\n",
    "    :param input_data: TF placeholder for text input.\n",
    "    :param vocab_size: Number of words in vocabulary.\n",
    "    :param embed_dim: Number of embedding dimensions\n",
    "    :return: Embedded input.\n",
    "    \"\"\"\n",
    "    \n",
    "    \"\"\"\n",
    "    tf.nn.embedding_lookup(\n",
    "    params,\n",
    "    ids,\n",
    "    partition_strategy='mod',\n",
    "    name=None,\n",
    "    validate_indices=True,\n",
    "    max_norm=None)\n",
    "    \"\"\"\n",
    "    \n",
    "    #According to TF Embeddings doc we have to create a weights matrix initialized with \"good values\" \n",
    "    #https://www.tensorflow.org/versions/master/programmers_guide/embedding\n",
    "    #I suggest you initialize it with a uniform random numbers between -1 and 1 using tf.random_uniform from skipgram \n",
    "    embedding = tf.Variable(tf.random_uniform((vocab_size, embed_dim), -1, 1))\n",
    "    \n",
    "    #Then we use the embeddings lookup\n",
    "    \n",
    "    embed = tf.nn.embedding_lookup(embedding, input_data)\n",
    "    \n",
    "    \n",
    "    return embed\n",
    "\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "tests.test_get_embed(get_embed)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Build RNN\n",
    "You created a RNN Cell in the `get_init_cell()` function.  Time to use the cell to create a RNN.\n",
    "- Build the RNN using the [`tf.nn.dynamic_rnn()`](https://www.tensorflow.org/api_docs/python/tf/nn/dynamic_rnn)\n",
    " - Apply the name \"final_state\" to the final state using [`tf.identity()`](https://www.tensorflow.org/api_docs/python/tf/identity)\n",
    "\n",
    "Return the outputs and final_state state in the following tuple `(Outputs, FinalState)` "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tests Passed\n"
     ]
    }
   ],
   "source": [
    "def build_rnn(cell, inputs):\n",
    "    \"\"\"\n",
    "    Create a RNN using a RNN Cell\n",
    "    :param cell: RNN Cell\n",
    "    :param inputs: Input text data\n",
    "    :return: Tuple (Outputs, Final State)\n",
    "    \"\"\"\n",
    "    # tf.nn.dynamic_rnn (cell,inputs,sequence_length=None,initial_state=None,dtype=None,parallel_iterations=None,\n",
    "    #swap_memory=False,time_major=False,scope=None) Returns:(outputs, state)\n",
    "    \n",
    "    Outputs, FinalState = tf.nn.dynamic_rnn(cell,inputs,dtype=tf.float32)\n",
    "    \n",
    "    #We have to add the \"tag\"\n",
    "    \n",
    "    FinalState = tf.identity(FinalState, name = \"final_state\")\n",
    "    \n",
    "    return Outputs, FinalState\n",
    "\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "tests.test_build_rnn(build_rnn)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Build the Neural Network\n",
    "Apply the functions you implemented above to:\n",
    "- Apply embedding to `input_data` using your `get_embed(input_data, vocab_size, embed_dim)` function.\n",
    "- Build RNN using `cell` and your `build_rnn(cell, inputs)` function.\n",
    "- Apply a fully connected layer with a linear activation and `vocab_size` as the number of outputs.\n",
    "\n",
    "Return the logits and final state in the following tuple (Logits, FinalState) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tests Passed\n"
     ]
    }
   ],
   "source": [
    "def build_nn(cell, rnn_size, input_data, vocab_size, embed_dim):\n",
    "    \"\"\"\n",
    "    Build part of the neural network\n",
    "    :param cell: RNN cell\n",
    "    :param rnn_size: Size of rnns\n",
    "    :param input_data: Input data\n",
    "    :param vocab_size: Vocabulary size\n",
    "    :param embed_dim: Number of embedding dimensions\n",
    "    :return: Tuple (Logits, FinalState)\n",
    "    \"\"\"\n",
    "    #Using the previous coded functions\n",
    "    \n",
    "    embed = get_embed(input_data,vocab_size,embed_dim)\n",
    "    outputs, FinalState = build_rnn(cell,embed)\n",
    "    \n",
    "    #Lets try default Xavier_Initialization for weights, zeros for bias\n",
    "    \n",
    "    Logits = tf.contrib.layers.fully_connected(outputs,vocab_size,activation_fn=None)\n",
    "    \n",
    "    return Logits, FinalState\n",
    "\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "tests.test_build_nn(build_nn)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Batches\n",
    "Implement `get_batches` to create batches of input and targets using `int_text`.  The batches should be a Numpy array with the shape `(number of batches, 2, batch size, sequence length)`. Each batch contains two elements:\n",
    "- The first element is a single batch of **input** with the shape `[batch size, sequence length]`\n",
    "- The second element is a single batch of **targets** with the shape `[batch size, sequence length]`\n",
    "\n",
    "If you can't fill the last batch with enough data, drop the last batch.\n",
    "\n",
    "For exmple, `get_batches([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], 3, 2)` would return a Numpy array of the following:\n",
    "```\n",
    "[\n",
    "  # First Batch\n",
    "  [\n",
    "    # Batch of Input\n",
    "    [[ 1  2], [ 7  8], [13 14]]\n",
    "    # Batch of targets\n",
    "    [[ 2  3], [ 8  9], [14 15]]\n",
    "  ]\n",
    "\n",
    "  # Second Batch\n",
    "  [\n",
    "    # Batch of Input\n",
    "    [[ 3  4], [ 9 10], [15 16]]\n",
    "    # Batch of targets\n",
    "    [[ 4  5], [10 11], [16 17]]\n",
    "  ]\n",
    "\n",
    "  # Third Batch\n",
    "  [\n",
    "    # Batch of Input\n",
    "    [[ 5  6], [11 12], [17 18]]\n",
    "    # Batch of targets\n",
    "    [[ 6  7], [12 13], [18  1]]\n",
    "  ]\n",
    "]\n",
    "```\n",
    "\n",
    "Notice that the last target value in the last batch is the first input value of the first batch. In this case, `1`. This is a common technique used when creating sequence batches, although it is rather unintuitive."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tests Passed\n"
     ]
    }
   ],
   "source": [
    "def get_batches(int_text, batch_size, seq_length):\n",
    "    \"\"\"\n",
    "    Return batches of input and target\n",
    "    :param int_text: Text with the words replaced by their ids\n",
    "    :param batch_size: The size of batch\n",
    "    :param seq_length: The length of sequence\n",
    "    :return: Batches as a Numpy array  \n",
    "    \"\"\"\n",
    "    n_batches = int(len(int_text) / (batch_size * seq_length))\n",
    "    #We create the \"skeleton\" np array\n",
    "    #batches = np.zeros(n_batches, 2, batch_size, seq_length)\n",
    "\n",
    "    # Drop the last few characters to make only full batches\n",
    "    xdata = np.array(int_text[: n_batches * batch_size * seq_length])\n",
    "    #Shift data magic occurs in this only line\n",
    "    ydata = np.array(int_text[1: n_batches * batch_size * seq_length + 1])\n",
    "    #The last target of the last batch should be the first input of the first batch. Being input x targets y\n",
    "    \n",
    "    ydata [-1] = xdata [0]\n",
    "    \n",
    "    x_batches = np.split(xdata.reshape(batch_size, -1), n_batches, 1)\n",
    "    y_batches = np.split(ydata.reshape(batch_size, -1), n_batches, 1)\n",
    "    \n",
    "    \n",
    "    \n",
    "    \n",
    "    #print(x_batches)\n",
    "    #print(y_batches)\n",
    "    salida = np.array(list(zip(x_batches, y_batches)))\n",
    "    #print (type(salida))\n",
    "    \n",
    "    return salida\n",
    "\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "tests.test_get_batches(get_batches)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Neural Network Training\n",
    "### Hyperparameters\n",
    "Tune the following parameters:\n",
    "\n",
    "- Set `num_epochs` to the number of epochs.\n",
    "- Set `batch_size` to the batch size.\n",
    "- Set `rnn_size` to the size of the RNNs.\n",
    "- Set `embed_dim` to the size of the embedding.\n",
    "- Set `seq_length` to the length of sequence.\n",
    "- Set `learning_rate` to the learning rate.\n",
    "- Set `show_every_n_batches` to the number of batches the neural network should print progress."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Number of Epochs\n",
    "num_epochs = 200\n",
    "# Batch Size\n",
    "batch_size = 256\n",
    "# RNN Size\n",
    "rnn_size = 512\n",
    "# Embedding Dimension Size\n",
    "embed_dim = 200\n",
    "# Sequence Length\n",
    "seq_length = 64\n",
    "# Learning Rate\n",
    "learning_rate = 0.01\n",
    "# Show stats for every n number of batches\n",
    "show_every_n_batches = 50\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "save_dir = './save'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Build the Graph\n",
    "Build the graph using the neural network you implemented."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL\n",
    "\"\"\"\n",
    "from tensorflow.contrib import seq2seq\n",
    "\n",
    "train_graph = tf.Graph()\n",
    "with train_graph.as_default():\n",
    "    vocab_size = len(int_to_vocab)\n",
    "    input_text, targets, lr = get_inputs()\n",
    "    input_data_shape = tf.shape(input_text)\n",
    "    cell, initial_state = get_init_cell(input_data_shape[0], rnn_size)\n",
    "    logits, final_state = build_nn(cell, rnn_size, input_text, vocab_size, embed_dim)\n",
    "\n",
    "    # Probabilities for generating words\n",
    "    probs = tf.nn.softmax(logits, name='probs')\n",
    "\n",
    "    # Loss function\n",
    "    cost = seq2seq.sequence_loss(\n",
    "        logits,\n",
    "        targets,\n",
    "        tf.ones([input_data_shape[0], input_data_shape[1]]))\n",
    "\n",
    "    # Optimizer\n",
    "    optimizer = tf.train.AdamOptimizer(lr)\n",
    "\n",
    "    # Gradient Clipping\n",
    "    gradients = optimizer.compute_gradients(cost)\n",
    "    capped_gradients = [(tf.clip_by_value(grad, -1., 1.), var) for grad, var in gradients if grad is not None]\n",
    "    train_op = optimizer.apply_gradients(capped_gradients)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train\n",
    "Train the neural network on the preprocessed data.  If you have a hard time getting a good loss, check the [forms](https://discussions.udacity.com/) to see if anyone is having the same problem."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch   0 Batch    0/4   train_loss = 8.825\n",
      "Epoch  12 Batch    2/4   train_loss = 3.962\n",
      "Epoch  25 Batch    0/4   train_loss = 2.592\n",
      "Epoch  37 Batch    2/4   train_loss = 1.712\n",
      "Epoch  50 Batch    0/4   train_loss = 1.112\n",
      "Epoch  62 Batch    2/4   train_loss = 0.716\n",
      "Epoch  75 Batch    0/4   train_loss = 0.364\n",
      "Epoch  87 Batch    2/4   train_loss = 0.243\n",
      "Epoch 100 Batch    0/4   train_loss = 0.104\n",
      "Epoch 112 Batch    2/4   train_loss = 0.070\n",
      "Epoch 125 Batch    0/4   train_loss = 0.054\n",
      "Epoch 137 Batch    2/4   train_loss = 0.048\n",
      "Epoch 150 Batch    0/4   train_loss = 0.042\n",
      "Epoch 162 Batch    2/4   train_loss = 0.041\n",
      "Epoch 175 Batch    0/4   train_loss = 0.038\n",
      "Epoch 187 Batch    2/4   train_loss = 0.038\n",
      "Model Trained and Saved\n"
     ]
    }
   ],
   "source": [
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL\n",
    "\"\"\"\n",
    "batches = get_batches(int_text, batch_size, seq_length)\n",
    "\n",
    "with tf.Session(graph=train_graph) as sess:\n",
    "    sess.run(tf.global_variables_initializer())\n",
    "\n",
    "    for epoch_i in range(num_epochs):\n",
    "        state = sess.run(initial_state, {input_text: batches[0][0]})\n",
    "\n",
    "        for batch_i, (x, y) in enumerate(batches):\n",
    "            feed = {\n",
    "                input_text: x,\n",
    "                targets: y,\n",
    "                initial_state: state,\n",
    "                lr: learning_rate}\n",
    "            train_loss, state, _ = sess.run([cost, final_state, train_op], feed)\n",
    "\n",
    "            # Show every <show_every_n_batches> batches\n",
    "            if (epoch_i * len(batches) + batch_i) % show_every_n_batches == 0:\n",
    "                print('Epoch {:>3} Batch {:>4}/{}   train_loss = {:.3f}'.format(\n",
    "                    epoch_i,\n",
    "                    batch_i,\n",
    "                    len(batches),\n",
    "                    train_loss))\n",
    "\n",
    "    # Save Model\n",
    "    saver = tf.train.Saver()\n",
    "    saver.save(sess, save_dir)\n",
    "    print('Model Trained and Saved')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Save Parameters\n",
    "Save `seq_length` and `save_dir` for generating a new TV script."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL\n",
    "\"\"\"\n",
    "# Save parameters for checkpoint\n",
    "helper.save_params((seq_length, save_dir))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Checkpoint"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL\n",
    "\"\"\"\n",
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "import helper\n",
    "import problem_unittests as tests\n",
    "\n",
    "_, vocab_to_int, int_to_vocab, token_dict = helper.load_preprocess()\n",
    "seq_length, load_dir = helper.load_params()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Implement Generate Functions\n",
    "### Get Tensors\n",
    "Get tensors from `loaded_graph` using the function [`get_tensor_by_name()`](https://www.tensorflow.org/api_docs/python/tf/Graph#get_tensor_by_name).  Get the tensors using the following names:\n",
    "- \"input:0\"\n",
    "- \"initial_state:0\"\n",
    "- \"final_state:0\"\n",
    "- \"probs:0\"\n",
    "\n",
    "Return the tensors in the following tuple `(InputTensor, InitialStateTensor, FinalStateTensor, ProbsTensor)` "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tests Passed\n"
     ]
    }
   ],
   "source": [
    "def get_tensors(loaded_graph):\n",
    "    \"\"\"\n",
    "    Get input, initial state, final state, and probabilities tensor from <loaded_graph>\n",
    "    :param loaded_graph: TensorFlow graph loaded from file\n",
    "    :return: Tuple (InputTensor, InitialStateTensor, FinalStateTensor, ProbsTensor)\n",
    "    \"\"\"\n",
    "    \n",
    "    \n",
    "    return loaded_graph.get_tensor_by_name(\"input:0\"), loaded_graph.get_tensor_by_name(\"initial_state:0\"), loaded_graph.get_tensor_by_name(\"final_state:0\"), loaded_graph.get_tensor_by_name(\"probs:0\")\n",
    "\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "tests.test_get_tensors(get_tensors)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Choose Word\n",
    "Implement the `pick_word()` function to select the next word using `probabilities`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tests Passed\n"
     ]
    }
   ],
   "source": [
    "def pick_word(probabilities, int_to_vocab):\n",
    "    \"\"\"\n",
    "    Pick the next word in the generated text\n",
    "    :param probabilities: Probabilites of the next word\n",
    "    :param int_to_vocab: Dictionary of word ids as the keys and words as the values\n",
    "    :return: String of the predicted word\n",
    "    \"\"\"\n",
    "    #palabra_random = np.random.choice(np.linspace(1,1,len(int_to_vocab)),p=probabilities)\n",
    "    \n",
    "    #return int_to_vocab[palabra_random]\n",
    "    return int_to_vocab[np.argmax(probabilities)]\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "tests.test_pick_word(pick_word)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Generate TV Script\n",
    "This will generate the TV script for you.  Set `gen_length` to the length of TV script you want to generate."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from ./save\n",
      "moe_szyslak:(moan) i'm ruined!\n",
      "laney_fontaine: sorry, moe. i love losers, but only after they've bounced at the bottom.(singing) 'cause i'm the brassiest piece of sass in this whole damn town!\n",
      "moe_szyslak: you guys cost me my chance with a woman of a certain age!(heaving sobs)\n",
      "carl_carlson: hey, hey!(no rhyme) this is supposed to be a children's story!\n",
      "moe_szyslak:(scared) look mister, i-i-i don't got the cash. my clientele-- they're all bums. they never pay!\n",
      "homer_simpson:(tough) just get the money.\n",
      "moe_szyslak:(tough) homer, the mob is puttin' the screws on me, see? i don't ever want to see that moolah-stealing jackpot-thief again!\n",
      "homer_simpson: wait a minute! something's reflected in the lenses of carl's sunglasses!\n",
      "homer_simpson: i just need to get me and vomit next to a guy i don't respect.\n",
      "homer_simpson:(sobs) even my best friends have left me! i'm empty inside.\n",
      "homer_simpson:(casual) oh, yeah. i'm a real expert.\n",
      "moe_szyslak: hey, what are you doin'? i gotta pay for that!\n",
      "homer_simpson: no moe. you've got it all wrong. people buy beer from you.\n",
      "moe_szyslak:(exasperated sigh) all right, look, i-i gotta go.\n",
      "homer_simpson:(matter-of-fact) i thought you said you had to go too far.\n",
      "moe_szyslak: how many people want homer banned from this place for life the(raises glass front of it.\n",
      "carl_carlson:(sadly) yeah, poor stevie. they never found his head.\n",
      "moe_szyslak:(accusing) mrs. simpson, when you took off, you left a hole in a ladies room.\n",
      "homer_simpson:(to skinner and barflies) shut up.\n",
      "comic_book_guy: my email address is edna-lover-one-seventy-two...\n",
      "moe_szyslak:(simultaneous) i/you could forget that image of her bolting for the parking lot.\n",
      "homer_simpson: okay.\n",
      "kent_brockman: so when you realized christmas was ruined, how did you feel?\n",
      "marge_simpson: how do you think i felt?\n",
      "kent_brockman: absolutely devastated.(turns to camera)\" absolutely devastated.\" the words of a heart-broken mother. for there will be forty-seven bottles of beer on the wall. yes. homer, you did not bring me here for this. what is it that you want to tell me?\n",
      "homer_simpson: okay, this isn't easy, so i'm just gonna come out and say it takes to get me fired to you.\n",
      "carl_carlson: yeah, rumor has it marge threw him out.\n",
      "barney_gumble: aw, tough break. she's\n"
     ]
    }
   ],
   "source": [
    "gen_length = 500\n",
    "# homer_simpson, moe_szyslak, or Barney_Gumble\n",
    "prime_word = 'moe_szyslak'\n",
    "\n",
    "\"\"\"\n",
    "DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE\n",
    "\"\"\"\n",
    "loaded_graph = tf.Graph()\n",
    "with tf.Session(graph=loaded_graph) as sess:\n",
    "    # Load saved model\n",
    "    loader = tf.train.import_meta_graph(load_dir + '.meta')\n",
    "    loader.restore(sess, load_dir)\n",
    "\n",
    "    # Get Tensors from loaded model\n",
    "    input_text, initial_state, final_state, probs = get_tensors(loaded_graph)\n",
    "\n",
    "    # Sentences generation setup\n",
    "    gen_sentences = [prime_word + ':']\n",
    "    prev_state = sess.run(initial_state, {input_text: np.array([[1]])})\n",
    "\n",
    "    # Generate sentences\n",
    "    for n in range(gen_length):\n",
    "        # Dynamic Input\n",
    "        dyn_input = [[vocab_to_int[word] for word in gen_sentences[-seq_length:]]]\n",
    "        dyn_seq_length = len(dyn_input[0])\n",
    "\n",
    "        # Get Prediction\n",
    "        probabilities, prev_state = sess.run(\n",
    "            [probs, final_state],\n",
    "            {input_text: dyn_input, initial_state: prev_state})\n",
    "        \n",
    "        pred_word = pick_word(probabilities[dyn_seq_length-1], int_to_vocab)\n",
    "\n",
    "        gen_sentences.append(pred_word)\n",
    "        \n",
    "    # Remove tokens\n",
    "    tv_script = ' '.join(gen_sentences)\n",
    "    for key, token in token_dict.items():\n",
    "        ending = ' ' if key in ['\\n', '(', '\"'] else ''\n",
    "        tv_script = tv_script.replace(' ' + token.lower(), key)\n",
    "    tv_script = tv_script.replace('\\n ', '\\n')\n",
    "    tv_script = tv_script.replace('( ', '(')\n",
    "        \n",
    "    print(tv_script)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# The TV Script is Nonsensical\n",
    "It's ok if the TV script doesn't make any sense.  We trained on less than a megabyte of text.  In order to get good results, you'll have to use a smaller vocabulary or get more data.  Luckly there's more data!  As we mentioned in the begging of this project, this is a subset of [another dataset](https://www.kaggle.com/wcukierski/the-simpsons-by-the-data).  We didn't have you train on all the data, because that would take too long.  However, you are free to train your neural network on all the data.  After you complete the project, of course.\n",
    "# Submitting This Project\n",
    "When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as \"dlnd_tv_script_generation.ipynb\" and save it as a HTML file under \"File\" -> \"Download as\". Include the \"helper.py\" and \"problem_unittests.py\" files in your submission."
   ]
  }
 ],
 "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": "3.5.3"
  },
  "widgets": {
   "state": {},
   "version": "1.1.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}