{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Classy words: formulas for finding category-associated terms\n",
    "## Jason S. Kessler @jasonkessler\n",
    "\n",
    "The Github repository for talk is at [https://github.com/JasonKessler/PuPPyTalk](https://github.com/JasonKessler/PuPPyTalk). \n",
    "\n",
    "Visualizations were made using [Scattertext](https://github.com/JasonKessler/scattertext).\n",
    "\n",
    "Please cite as: \n",
    "Jason S. Kessler. Scattertext: a Browser-Based Tool for Visualizing how Corpora Differ. ACL System Demonstrations. 2017."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style>.container { width:98% !important; }</style>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import scattertext as st\n",
    "import spacy\n",
    "from scipy.stats import rankdata\n",
    "from IPython.display import IFrame\n",
    "from IPython.core.display import display, HTML\n",
    "display(HTML(\"<style>.container { width:98% !important; }</style>\"))\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "ename": "TypeError",
     "evalue": "'>=' not supported between instances of 'list' and 'str'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mTypeError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-3-cf8925f6971a>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0;32massert\u001b[0m \u001b[0mst\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mversion\u001b[0m \u001b[0;34m>=\u001b[0m \u001b[0;34m'0.0.2.22'\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mTypeError\u001b[0m: '>=' not supported between instances of 'list' and 'str'"
     ]
    }
   ],
   "source": [
    "assert st.version >= '0.0.2.22'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### The data\n",
    " \n",
    "Dataset consists of reviews of movies and plot descriptions.  Plot descriptions are guaranteed to be from a movie which was reviewed. \n",
    "\n",
    "Data set is from http://www.cs.cornell.edu/people/pabo/movie-review-data/\n",
    "\n",
    "References:\n",
    "* Bo Pang, Lillian Lee, and Shivakumar Vaithyanathan, Thumbs up? Sentiment Classification using Machine Learning Techniques, Proceedings of EMNLP 2002.\n",
    "\n",
    "* Bo Pang and Lillian Lee, A Sentimental Education: Sentiment Analysis Using Subjectivity Summarization Based on Minimum Cuts, Proceedings of ACL 2004."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Positive    2455\n",
      "Negative    2411\n",
      "Plot         156\n",
      "Name: category_name, dtype: int64\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>movie_name</th>\n",
       "      <th>category_name</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>A senior at an elite college (Katie Holmes), a...</td>\n",
       "      <td>abandon</td>\n",
       "      <td>Plot</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Will Lightman is a hip Londoner who one day re...</td>\n",
       "      <td>about_a_boy</td>\n",
       "      <td>Plot</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Warren Schmidt (Nicholson) is forced to deal w...</td>\n",
       "      <td>about_schmidt</td>\n",
       "      <td>Plot</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>An account of screenwriter Charlie Kaufman's (...</td>\n",
       "      <td>adaptation</td>\n",
       "      <td>Plot</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Ali G unwittingly becomes a pawn in the evil C...</td>\n",
       "      <td>ali_g_indahouse</td>\n",
       "      <td>Plot</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                text       movie_name  \\\n",
       "0  A senior at an elite college (Katie Holmes), a...          abandon   \n",
       "1  Will Lightman is a hip Londoner who one day re...      about_a_boy   \n",
       "2  Warren Schmidt (Nicholson) is forced to deal w...    about_schmidt   \n",
       "3  An account of screenwriter Charlie Kaufman's (...       adaptation   \n",
       "4  Ali G unwittingly becomes a pawn in the evil C...  ali_g_indahouse   \n",
       "\n",
       "  category_name  \n",
       "0          Plot  \n",
       "1          Plot  \n",
       "2          Plot  \n",
       "3          Plot  \n",
       "4          Plot  "
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rdf = st.SampleCorpora.RottenTomatoes.get_data()\n",
    "rdf['category_name'] = rdf['category'].apply(lambda x: {'plot': 'Plot', 'rotten': 'Negative', 'fresh': 'Positive'}[x])\n",
    "print(rdf.category_name.value_counts())\n",
    "rdf[['text', 'movie_name', 'category_name']].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "corpus = (st.CorpusFromPandas(rdf, \n",
    "                              category_col='category_name', \n",
    "                              text_col='text',\n",
    "                              nlp = st.whitespace_nlp_with_sentences)\n",
    "          .build())\n",
    "corpus.get_term_freq_df().to_csv('term_freqs.csv')\n",
    "unigram_corpus = corpus.get_unigram_corpus()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Let's visualize the corpus using Scattertext\n",
    "\n",
    "The x-axis indicates the rank of a word or bigram in the set of positive reviews, and the y-axis negative reviews.\n",
    "\n",
    "Ranks are determined using \"dense\" ranking, meaning the most frequent terms, regardless of ties, are given rank 1, the next most frequent terms, regardless of ties, are given rank 2, etc.\n",
    "\n",
    "It appears that terms more associated with a class are a further distance from the diagonal line between the lower-left and upper-right corners.  Terms are colored according to this distance.  We'll return to this in a bit.\n",
    "\n",
    "Scattertext selectively labels points in such a way as to prevent labels from overlapping other elements of the graph. Mouse-over points and term labels for a preview, and click for a key-word in context view.\n",
    "\n",
    "References:\n",
    "* Jason S. Kessler. Scattertext: a Browser-Based Tool for Visualizing how Corpora Differ. ACL System Demonstrations. 2017."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_stdense.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x1164cc7f0>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "html = st.produce_scattertext_explorer(\n",
    "    corpus,\n",
    "    category='Positive',\n",
    "    not_categories=['Negative'],\n",
    "    sort_by_dist=False,\n",
    "    metadata=rdf['movie_name'],\n",
    "    term_scorer=st.RankDifference(),\n",
    "    transform=st.Scalers.percentile_dense\n",
    ")\n",
    "file_name = 'output/rotten_fresh_stdense.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### We view can see more terms through breaking ties in ranking alphabetically.\n",
    "Lower frequency terms are more prominent in this view, and more terms can be labeled."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_st.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x115928278>"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "html = st.produce_scattertext_explorer(\n",
    "    corpus,\n",
    "    category='Positive',\n",
    "    not_categories=['Negative'],\n",
    "    sort_by_dist=False,\n",
    "    metadata=rdf['movie_name'],\n",
    "    term_scorer=st.RankDifference(),\n",
    ")\n",
    "file_name = 'output/rotten_fresh_st.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Examining term-category association\n",
    "We'll roughly follow the explanation given in Monroe et. al (2008), culminating in a version of their Log-Odds-Ratio with an Informed Dirichlet Prior score. \n",
    "\n",
    "### Naive approach 1\n",
    "### Frequency difference (not recommended)\n",
    "$$\n",
    "\\begin{align}\n",
    "  \\mbox{freqdiff}(\\mbox{term}) & := \\mbox{% or words in catgory A that are term} &- \\mbox{% or words in catgory B that are term}\\\\\n",
    "  \\mbox{freqdiff}(\\mbox{term}) & := \\frac{\\#(\\mbox{term}\\in\\mbox{category}_a)}{\\#(\\mbox{category}_a)} &- \\frac{\\#(\\mbox{term}\\in\\mbox{category}_b)}{\\#(\\mbox{category}_b)}\\\\\n",
    "\\end{align}\n",
    "$$\n",
    "Disporportionately favors very frequent words.\n",
    "\n",
    "References\n",
    "* Burt L. Monroe, Michael P. Colaresi, and Kevin M. Quinn. 2008. Fightin’ words: Lexical feature selection and evaluation for identifying the content of political conflict. Political Analysis.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_diff.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x108aa4780>"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class FrequencyDifference:\n",
    "    def get_scores(self, a, b): \n",
    "        return a/np.sum(a) - b/np.sum(b)\n",
    "    def get_name(self): \n",
    "        return 'Frequency Difference'\n",
    "    \n",
    "html = st.produce_fightin_words_explorer(\n",
    "    corpus,\n",
    "    category='Positive',\n",
    "    not_categories=['Negative'],\n",
    "    term_scorer=FrequencyDifference(),\n",
    "    metadata = rdf['movie_name'],\n",
    "    grey_threshold=0\n",
    ")\n",
    "file_name = 'output/rotten_fresh_diff.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Naive approach 2\n",
    "### tf.idf difference (not recommended)\n",
    "$$ \\mbox{Term Frquency}(\\mbox{term}, \\mbox{category}) = \\#(\\mbox{term}\\in\\mbox{category}) $$\n",
    "\n",
    "$$ \\mbox{Inverse Document Frquency}(\\mbox{term}) = \\log \\frac{\\mbox{# of categories}}{\\mbox{# of categories containing term}} $$\n",
    "\n",
    "$$ \\mbox{tfidf}(\\mbox{term}, \\mbox{category}) = \\mbox{Term Frquency}(\\mbox{term}, \\mbox{category}) \\times \\mbox{Inverse Document Frquency}(\\mbox{term}) $$\n",
    "\n",
    "$$ \\mbox{tfidf-difference}(\\mbox{term}, \\mbox{category}) = \\mbox{tf.idf}(\\mbox{term}, \\mbox{category}_a) - \\mbox{tf.idf}(\\mbox{term}, \\mbox{category}_b) $$\n",
    "\n",
    "Tf.idf ignores terms used in each category.  Since we only consider three categories (positive, negative and plot descriptions), a large number of terms have zero (log 1) scores.  The problem is Tf.idf doesn't weight how often a term is used in another category.  This causes eccentric, brittle, low-frequency terms to be favored.\n",
    "\n",
    "This formulation does take into account data from a background corpus."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kesslej/anaconda3/envs/py36/lib/python3.6/site-packages/scattertext-0.0.2.20-py3.6.egg/scattertext/TermDocMatrixFilter.py:43: RuntimeWarning: invalid value encountered in log\n",
      "  bigram_prob[bigram] / np.product([unigram_prob[word] for word in bigram.split(' ')])\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_tfidf.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x1178e3f28>"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class TFIDF:\n",
    "    def __init__(self, corpus):\n",
    "        self.tdf = corpus.get_term_freq_df()\n",
    "    def get_scores(self, a, b): \n",
    "        n_docs = self.tdf.shape[1]\n",
    "        n_docs_with_term = (self.tdf > 0).astype(float).sum(axis=1)\n",
    "        idf = np.log(n_docs/n_docs_with_term)/np.log(2)\n",
    "        return a * idf - b * idf\n",
    "    def get_name(self): \n",
    "        return 'tf.idf difference'\n",
    "    \n",
    "html = st.produce_fightin_words_explorer(\n",
    "    corpus,\n",
    "    category='Positive',\n",
    "    not_categories=['Negative'],\n",
    "    neutral_categories=['Plot'],\n",
    "    term_scorer=TFIDF(corpus),\n",
    "    metadata = rdf['movie_name'],\n",
    "    grey_threshold=0,\n",
    "    show_neutral=True\n",
    ")\n",
    "file_name = 'output/rotten_fresh_tfidf.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Naive approach 3\n",
    "### Log-odds-ratio w/ smoothing (not preferred)\n",
    "\n",
    "$$ \\newcommand\\ddfrac[2]{\\frac{\\displaystyle #1}{\\displaystyle #2}}$$\n",
    "$$ \\mbox{Odds}(\\mbox{term}_i, \\mbox{category}) = \\frac{\\#(\\mbox{term}_i\\in\\mbox{category}_a)}{\\sum_{\\mbox{term}` \\neq \\mbox{term}} \\#(\\mbox{term}'\\in\\mbox{category})} = \\frac{y_{ai}}{n_{a} - y_{ai}} $$\n",
    "Note: this can be a very small number.\n",
    "\n",
    "The odds ratio is the ratio of the odds of a term being used in one category vs. another.\n",
    "$$ \\mbox{Odds-Ratio}(\\mbox{term}_i, \\mbox{category}_a, \\mbox{category}_b) =  \\ddfrac{y_{ai}\\mathbin{/}(n_{a} - y_{ai})}{y_{bi}\\mathbin{/}(n_{b} - y_{bi})} $$\n",
    "\n",
    "The Log-Odds-Ratio is often used, given these quantities are often very small.  For our purposes, the fact that it's normally distributed is important.\n",
    "\n",
    "$$ \\mbox{Log-Odds-Ratio}(\\mbox{term}_i, \\mbox{category}_a, \\mbox{category}_b) = \\log \\frac{y_{ai}}{n_{a} - y_{ai}} - \\log \\frac{y_{bi}}{n_{b} - y_{bi}} $$\n",
    "\n",
    "If term does not appear in $\\mbox{category}_b$ (i.e., $ y_{bi} = 0 $), the odd of term_i in category_b will be 0 resulting the odds-ratio to be undefined.  We circumvent this issue by adding a pseudo-count to each term, assuming it occurs at least $\\alpha$ times.  \n",
    "\n",
    "$$ \\mbox{Smoothed Log-Odds-Ratio}(\\mbox{term}_i, \\mbox{category}_a, \\mbox{category}_b) = \\log \\frac{y_{ai} + \\alpha}{n_{a} + \\alpha \\cdot |y| - y_{ai} - \\alpha} - \\log \\frac{y_{bi} + \\alpha}{n_{b} + \\alpha \\cdot |y| - y_{bi} - \\alpha} $$\n",
    "\n",
    "This has a slightly better profile than tf.idf, but still favors very low-frequency terms that are unique to a category.\n",
    "\n",
    "We can see that terms used exclusively in one category or another clump form extremely high or low-scoring clumps.  These terms occur infrequently.  Intuitively, we'd like to score things highly if they have a lot of evidence behind them AND they tend to be used much more in one category vs. another.\n",
    "\n",
    "Note: a common error when implementing this is to not account for other pseudocounts in the denominator.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_lorsmooth.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x119857630>"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class LogOddsRatioSmoothed:\n",
    "    def __init__(self, alpha):\n",
    "        self.alpha = alpha\n",
    "    def get_scores(self, a, b): \n",
    "        return (np.log((a + self.alpha)/(np.sum(a) + self.alpha * len(a) - a - self.alpha)) \n",
    "                - np.log((b + self.alpha)/(np.sum(b) + self.alpha * len(b) - b - self.alpha)))\n",
    "    def get_name(self): \n",
    "        return 'Smoothed Log-Odds-Ratio'\n",
    "    \n",
    "html = st.produce_fightin_words_explorer(\n",
    "    corpus,\n",
    "    category='Positive',\n",
    "    not_category_name='Negative',\n",
    "    not_categories=['Negative'],\n",
    "    term_scorer=LogOddsRatioSmoothed(0.00000001),\n",
    "    metadata = rdf['movie_name'],\n",
    "    grey_threshold=0,\n",
    ")\n",
    "file_name = 'output/rotten_fresh_lorsmooth.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Accounting for variance\n",
    "How can we account for the certainty of the estimates?  Less frequent terms will have higher variances.\n",
    "\n",
    "The standard error of the log-odds-ratio is:\n",
    "\n",
    "$$ \\mbox{SE-Smoothed Log-Odds-Ratio}(\\mbox{term}_i, \\mbox{category}_a, \\mbox{category}_b) = \\frac{1}{y_{ai} + \\alpha} + \\frac{1}{n_{a} + \\alpha \\cdot |y| - y_{ai} - \\alpha} + \\frac{1}{y_{bi} + \\alpha}+ \\frac{1}{n_{b} + \\alpha \\cdot |y| - y_{bi} - \\alpha} $$\n",
    "\n",
    "The plot below shows how terms used infrequently and especially those which only appear in one class have a very high variance."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_lorsmoothse.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x11a7996a0>"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class SELogOddsRatioSmoothed:\n",
    "    def __init__(self, alpha):\n",
    "        self.alpha = alpha\n",
    "    def get_scores(self, a, b): \n",
    "        a += self.alpha\n",
    "        b += self.alpha\n",
    "        return  1./a + 1./(np.sum(a) - a) + 1./b + 1./(np.sum(b) - b)\n",
    "    def get_name(self): \n",
    "        return 'SE Smoothed Log-Odds-Ratio'\n",
    "    \n",
    "html = st.produce_fightin_words_explorer(\n",
    "    corpus,\n",
    "    category='Positive',\n",
    "    not_category_name='Negative',\n",
    "    not_categories=['Negative'],\n",
    "    term_scorer=SELogOddsRatioSmoothed(0.00000001),\n",
    "    metadata = rdf['movie_name'],\n",
    "    grey_threshold=0\n",
    ")\n",
    "file_name = 'output/rotten_fresh_lorsmoothse.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Smoothed Log-Odds-Ratio Z-Score\n",
    "### Preferred in certain cases\n",
    "\n",
    "Finding the Z-Scores of log-odds-ratios allows us to gracefully account for variance.\n",
    "\n",
    "$$ \\mbox{LOR Smoothed Z-Score}(\\mbox{term}_i, \\mbox{category}_a, \\mbox{category}_b) = \\frac{\\mbox{LOR Smoothed}}{\\sqrt{\\mbox{SE LOR Smoothed}}} $$\n",
    "\n",
    "The chart below, given a smoothing $\\alpha$ of 0.01, dampens the score of infrequent words and allows for frequently words to rank highly.  However, the words \"and\" and \"of\" appear to be over-weighted. \n",
    "\n",
    "Eccentric words like \"winning\" are down-weighted but still significant.\n",
    "\n",
    "MCQ refers to this at the Log-Odds-Ratio with an Uninformative Dirichlet Prior"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_lorudp.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x118416e80>"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class LogOddsRatioSmoothedZScorePrior:\n",
    "    def __init__(self, prior):\n",
    "        self.prior = prior\n",
    "    def get_scores(self, a, b): \n",
    "        a += self.prior\n",
    "        b += self.prior\n",
    "        lor = (np.log(a/(np.sum(a) - a)) - np.log(b/(np.sum(b) - b)))\n",
    "        lorstd = 1./a + 1./(np.sum(a) - a) + 1./b + 1./(np.sum(b) - b)\n",
    "        return lor/np.sqrt(lorstd)\n",
    "        \n",
    "    def get_name(self): \n",
    "        return 'Smoothed Log-Odds-Ratio w/ Prior Z-Score'\n",
    "html = st.produce_fightin_words_explorer(\n",
    "    corpus,\n",
    "    category='Positive',\n",
    "    not_category_name='Negative',\n",
    "    not_categories=['Negative'],\n",
    "    term_scorer=LogOddsRatioSmoothedZScorePrior(0.01),\n",
    "    metadata = rdf['movie_name'],\n",
    "    grey_threshold = 1.96\n",
    ")\n",
    "file_name = 'output/rotten_fresh_lorudp.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Log-Odds-Ratio with Informative Dirichlet Prior Z-Score\n",
    "### Prefered, currently very popular in NLP and CSS literature, flaws exist\n",
    "### Note: this formulation is modified slightly from Monroe et al. 2008.\n",
    "\n",
    "We know, intuitively, words which which are used very frequently require much more evidence to be significantly associated with a class than less frequent words.  For example, words like \"best\" and \"entertaining\" in the above example should be scored lower than \"and\" and \"of\".  \n",
    "\n",
    "The approach developed by Monroe et. al is to more aggressively smooth terms based on their frequency in a background corpus.\n",
    "\n",
    "The downside is of this approach is that obtaining a background corpus may be difficult. Lists of background term frequencies, especially for 2+grams, can be very large and unwieldy. Moreover, term frequency lists can become outdated very quickly, especially wrt to politics, pop culture, and technology.\n",
    "\n",
    "We construct two vectors of background term counts for each category, $\\alpha_a$ and $\\alpha_b$.  The priors are based on a background set of word frequencies, referred to as $y_{c}$  These are ideally in the same domain as the documents being studied, although could come from a general list of word frequencies.  The background counts are normalized to the size of their respective categories. The hyperparameter $\\sigma$ scales the background counts to a multiple of their category size. In other words, as $\\sigma$ increases, the regularization increases, and the odds ratio tends more toward zero.\n",
    "\n",
    "The $\\sigma$ I've found most useful is 10, which I include in Scattertext.\n",
    "\n",
    "$$ \\alpha_a = \\sigma \\cdot n_a \\cdot \\frac{y_{ci}}{n_c} $$\n",
    "\n",
    "$$ \\alpha_b = \\sigma \\cdot n_b \\cdot \\frac{y_{ci}}{n_c} $$\n",
    "\n",
    "\n",
    "The $\\alpha$s can then be used analogously to find the Z-Score of the LOR.\n",
    "\n",
    "$$ \\mbox{LOR}(\\mbox{term}_i, \\mbox{category}_a, \\mbox{category}_b) = \\log \\frac{y_{ai} + \\alpha_a}{n_{a} + \\alpha_a \\cdot |y| - y_{ai} - \\alpha_a} - \\log \\frac{y_{bi} + \\alpha_b}{n_{b} + \\alpha_b \\cdot |y| - y_{bi} - \\alpha_b} $$\n",
    "\n",
    "$$ \\mbox{LOR-SE}(\\mbox{term}_i, \\mbox{category}_a, \\mbox{category}_b) = \\frac{1}{y_{ai} + \\alpha_a} + \\frac{1}{n_{a} + \\alpha_a \\cdot |y| - y_{ai} - \\alpha_a} + \\frac{1}{y_{bi} + \\alpha_b}+ \\frac{1}{n_{b} + \\alpha_b \\cdot |y| - y_{bi} - \\alpha_b} $$\n",
    "\n",
    "Monroe et al. used a different approach to finding $\\alpha$, scaling the sum of the peusdocounts to the mean number of words in a document.\n",
    "\n",
    "$$ \\alpha_a = \\alpha_b = \\overline{|d_{y}|} \\frac{y_{ci}}{\\sum n_{c}} $$\n",
    "\n",
    "Jurafsky et al. suggested using raw background corpus counts as $\\alpha$.\n",
    "\n",
    "$$ \\alpha_a = \\alpha_b = y_c $$\n",
    "\n",
    "In the following example, we use the plot description as a background corpus using the first method.\n",
    "\n",
    "* Dan Jurafsky, Victor Chahuneau, Bryan Routledge, and Noah Smith. Narrative framing of consumer sentiment in online restaurant reviews. First Monday. 2014."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_loridp.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x118118b38>"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "priors = (st.PriorFactory(unigram_corpus, \n",
    "                          category='Positive', \n",
    "                          not_categories=['Negative'],\n",
    "                          starting_count=0.01)\n",
    "          .use_neutral_categories()\n",
    "          .get_priors())\n",
    "class LogOddsRatioSmoothedZScorePrior:\n",
    "    def __init__(self, prior, prior_scale):\n",
    "        self.prior = prior\n",
    "        self.prior_scale = prior_scale\n",
    "    def get_scores(self, a, b): \n",
    "        ap = a + self.prior * self.prior_scale*sum(a)/sum(self.prior.values)\n",
    "        bp = b + self.prior * self.prior_scale*sum(b)/sum(self.prior.values)\n",
    "        lor = (np.log(ap/(np.sum(ap) - ap)) - np.log(bp/(np.sum(bp) - bp)))\n",
    "        lorstd = 1./ap + 1./(np.sum(ap) - ap) + 1./bp + 1./(np.sum(bp) - bp)\n",
    "        return lor/np.sqrt(lorstd)\n",
    "        \n",
    "    def get_name(self): \n",
    "        return 'Log-Odds-Ratio w/ Informative Dirichlet Prior Z-Score'\n",
    "html = st.produce_fightin_words_explorer(\n",
    "    unigram_corpus,\n",
    "    category='Positive',\n",
    "    not_category_name='Negative',\n",
    "    not_categories=['Negative'],\n",
    "    #term_scorer=LogOddsRatioSmoothedZScorePrior(priors, 10, ),\n",
    "    # Equivalent:\n",
    "    term_scorer=st.LogOddsRatioInformativeDirichletPrior(priors, 10, 'class-size'),    \n",
    "    metadata = rdf['movie_name']\n",
    ")\n",
    "file_name = 'output/rotten_fresh_loridp.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Effect of $\\sigma$\n",
    "In the plot below, we can see how a changing $\\sigma$ alters the z-scores of low and high-frequency terms which are associated the Positive category.\n",
    "\n",
    "A high $\\sigma$ causes the infrequent word \"winning\", which only occurs in Positive documents, to dominate more frequent, associated terms.  A very low $\\sigma$ causes rankings to revert to Smoothed LOR Z-Score."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.text.Text at 0x11959df98>"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA4EAAAH1CAYAAABIhvsyAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nUEREREREZB1RCBQREREREVlH/n/FQ2Spqu+QIQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1195a4b00>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "term_freqs = unigram_corpus.get_term_freq_df()\n",
    "data = []\n",
    "for sigma in np.linspace(0.01, 30, 40):\n",
    "    scores = LogOddsRatioSmoothedZScorePrior(priors, sigma).get_scores(term_freqs['Positive freq'], term_freqs['Negative freq'])\n",
    "    for term in ['and', 'best', 'winning', 'of', 'fun']:\n",
    "        data.append({'term': term, 'score': scores[term], 'sigma': sigma})\n",
    "ax = (pd.DataFrame(data)\n",
    " .pivot(index='sigma', columns='term', values='score')\n",
    " .plot(figsize=(15, 8), fontsize=20))\n",
    "ax.legend(prop={'size': 20})\n",
    "ax.set_xlabel('$\\sigma_{class\\ size}$', size= 20)\n",
    "ax.axvline(10, color='k', linestyle='--')\n",
    "ax.axhline(1.96, color='k', linestyle='--')\n",
    "ax.set_ylabel('Z-Score', size= 20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The method suggested in Monroe et al., on the other hand, is more restrictive when it comes to stop words, strongly minimizing the word \"and\", in particular.  The words \"fun\" and \"entertaining\" end up being below the signficance threshold."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_loridp_monroe.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x11c1c11d0>"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mean_doc_length = rdf.text.apply(lambda x: len(st.whitespace_nlp(x).sents[0])).mean()\n",
    "html = st.produce_fightin_words_explorer(\n",
    "    unigram_corpus,\n",
    "    category='Positive',\n",
    "    not_category_name='Negative',\n",
    "    not_categories=['Negative'],\n",
    "    term_scorer=st.LogOddsRatioInformativeDirichletPrior(priors, mean_doc_length, 'word'),    \n",
    "    metadata = rdf['movie_name']\n",
    ")\n",
    "file_name = 'output/rotten_fresh_loridp_monroe.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_jurfsky.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x114216e10>"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "html = st.produce_fightin_words_explorer(\n",
    "    unigram_corpus,\n",
    "    category='Positive',\n",
    "    not_category_name='Negative',\n",
    "    not_categories=['Negative'],\n",
    "    term_scorer=st.LogOddsRatioInformativeDirichletPrior(priors, 1, 'none'),    \n",
    "    metadata = rdf['movie_name']\n",
    ")\n",
    "file_name = 'output/rotten_fresh_jurfsky.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Dense rank differences\n",
    "### Prefered, but may under-score extremely frequent and infrequent terms\n",
    "\n",
    "Let $\\rho_\\mbox{category}$ be the dense rank function for a given category, with the most frequently occuring term(s) having a rank of 0.\n",
    "\n",
    "Let $d_\\mbox{category}$ be the number of distinct term frequencies in a category.  The most infrequenty terms receive this rank.\n",
    "\n",
    "This metric takes the difference of the normalized category-specific dense ranks of term frequencies. \n",
    "\n",
    "Define the rank-difference score as the difference in \n",
    "$$\\mbox{Rank-Difference} = \\frac{\\rho_{a}(y_a)}{d_a} - \\frac{\\rho_{b}(y_b)}{d_b}$$\n",
    "\n",
    "We can see that the words which occur fairly frequently are favored.  Frequent and very infrequent words are heavily regularized.\n",
    "\n",
    "No background frequencies are required!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_rankdiff.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x11847e978>"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class RankDifference:\n",
    "    def get_scores(self, a, b): \n",
    "        return rankdata(a,'dense')/np.max(rankdata(a,'dense')) - rankdata(b,'dense')/np.max(rankdata(b,'dense'))\n",
    "    def get_name(self): \n",
    "        return 'Rank Difference'\n",
    "    \n",
    "html = st.produce_fightin_words_explorer(\n",
    "    corpus,\n",
    "    category='Positive',\n",
    "    not_category_name='Negative',\n",
    "    not_categories=['Negative'],\n",
    "    term_scorer=RankDifference(),\n",
    "    metadata = rdf['movie_name'],\n",
    "    grey_threshold=0\n",
    ")\n",
    "file_name = 'output/rotten_fresh_rankdiff.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Scaled F-Score\n",
    "### Associatied terms have a *relatively* high category-specific precision and category-specific term frequency (i.e., % of terms in category are term)\n",
    "### Take the harmonic mean of precision and frequency (both have to be high)\n",
    "\n",
    "Given a word $w_i \\in W$ and a category $c_j \\in C$, define the precision of the word $w_i$ wrt to a category as:\n",
    "$$ \\mbox{prec}(w_i, c_j) = \\frac{\\#(w_i, c_j)}{\\sum_{c \\in C} \\#(w_i, c)}. $$\n",
    "\n",
    "The function $\\#(w_i, c_j)$ represents either the number of times $w_i$ occurs in a document labeled with the category $c_j$ or the number of documents labeled $c_j$ which contain $w_i$.\n",
    "\n",
    "Similarly, define the frequency a word occurs in the category as:\n",
    "\n",
    "$$ \\mbox{freq}(w_i, c_j) = \\frac{\\#(w_i, c_j)}{\\sum_{w \\in W} \\#(w, c_j)}. $$\n",
    "\n",
    "The F-Score of these two values is defined as:\n",
    "\n",
    "$$ \\mathcal{F}_\\beta(\\mbox{prec}, \\mbox{freq}) = (1 + \\beta^2) \\frac{\\mbox{prec} \\cdot \\mbox{freq}}{\\beta^2 \\cdot \\mbox{prec} + \\mbox{freq}}. $$\n",
    "\n",
    "$\\beta \\in \\mathcal{R}^+$ is a scaling factor where frequency is favored if $\\beta < 1$, precision if $\\beta > 1$, and both are equally weighted if $\\beta = 1$. F-Score is equivalent to the harmonic mean where $\\beta = 1$."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Positive freq</th>\n",
       "      <th>Negative freq</th>\n",
       "      <th>pos_precision</th>\n",
       "      <th>pos_freq_pct</th>\n",
       "      <th>pos_hmean</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>term</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>the</th>\n",
       "      <td>2346</td>\n",
       "      <td>2288</td>\n",
       "      <td>0.506258</td>\n",
       "      <td>0.024735</td>\n",
       "      <td>0.047166</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>a</th>\n",
       "      <td>1775</td>\n",
       "      <td>1613</td>\n",
       "      <td>0.523908</td>\n",
       "      <td>0.018715</td>\n",
       "      <td>0.036139</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>and</th>\n",
       "      <td>1637</td>\n",
       "      <td>1179</td>\n",
       "      <td>0.581321</td>\n",
       "      <td>0.017260</td>\n",
       "      <td>0.033524</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>of</th>\n",
       "      <td>1480</td>\n",
       "      <td>1235</td>\n",
       "      <td>0.545120</td>\n",
       "      <td>0.015604</td>\n",
       "      <td>0.030340</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>to</th>\n",
       "      <td>942</td>\n",
       "      <td>1010</td>\n",
       "      <td>0.482582</td>\n",
       "      <td>0.009932</td>\n",
       "      <td>0.019463</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>it</th>\n",
       "      <td>826</td>\n",
       "      <td>801</td>\n",
       "      <td>0.507683</td>\n",
       "      <td>0.008709</td>\n",
       "      <td>0.017124</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>is</th>\n",
       "      <td>818</td>\n",
       "      <td>726</td>\n",
       "      <td>0.529793</td>\n",
       "      <td>0.008625</td>\n",
       "      <td>0.016973</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>s</th>\n",
       "      <td>808</td>\n",
       "      <td>749</td>\n",
       "      <td>0.518947</td>\n",
       "      <td>0.008519</td>\n",
       "      <td>0.016763</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>in</th>\n",
       "      <td>676</td>\n",
       "      <td>622</td>\n",
       "      <td>0.520801</td>\n",
       "      <td>0.007127</td>\n",
       "      <td>0.014062</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>that</th>\n",
       "      <td>617</td>\n",
       "      <td>602</td>\n",
       "      <td>0.506153</td>\n",
       "      <td>0.006505</td>\n",
       "      <td>0.012846</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      Positive freq  Negative freq  pos_precision  pos_freq_pct  pos_hmean\n",
       "term                                                                      \n",
       "the            2346           2288       0.506258      0.024735   0.047166\n",
       "a              1775           1613       0.523908      0.018715   0.036139\n",
       "and            1637           1179       0.581321      0.017260   0.033524\n",
       "of             1480           1235       0.545120      0.015604   0.030340\n",
       "to              942           1010       0.482582      0.009932   0.019463\n",
       "it              826            801       0.507683      0.008709   0.017124\n",
       "is              818            726       0.529793      0.008625   0.016973\n",
       "s               808            749       0.518947      0.008519   0.016763\n",
       "in              676            622       0.520801      0.007127   0.014062\n",
       "that            617            602       0.506153      0.006505   0.012846"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from scipy.stats import hmean\n",
    "\n",
    "term_freq_df = corpus.get_term_freq_df()[['Positive freq', 'Negative freq']]\n",
    "term_freq_df = term_freq_df[term_freq_df.sum(axis=1) > 0]\n",
    "\n",
    "term_freq_df['pos_precision'] = (term_freq_df['Positive freq'] * 1./\n",
    "                                 (term_freq_df['Positive freq'] + term_freq_df['Negative freq']))\n",
    "\n",
    "term_freq_df['pos_freq_pct'] = (term_freq_df['Positive freq'] * 1.\n",
    "                                /term_freq_df['Positive freq'].sum())\n",
    "\n",
    "term_freq_df['pos_hmean'] = (term_freq_df\n",
    "                             .apply(lambda x: (hmean([x['pos_precision'], x['pos_freq_pct']])\n",
    "                                               if x['pos_precision'] > 0 and x['pos_freq_pct'] > 0 \n",
    "                                               else 0), axis=1))\n",
    "term_freq_df.sort_values(by='pos_hmean', ascending=False).iloc[:10]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Problem: harmonic means are dominated by the precision\n",
    "### Take the normal CDF of precision and frequency percentage scores, which will fall between 0 and 1, which scales and standardizes both scores.\n",
    "\n",
    "Define the the Normal CDF as:\n",
    "\n",
    "$$ \\Phi(z) = \\int_{-\\infty}^z \\mathcal{N}(x; \\mu, \\sigma^2)\\ \\mathrm{d}x.$$\n",
    "\n",
    "Where $ \\mathcal{N} $ is the PDF of the Normal distribution, $\\mu$ is the mean, and $\\sigma^2$ is the variance.\n",
    "\n",
    "$\\Phi$ is used to scale and standardize the precisions and frequencies, and place them on the same scale $[0,1]$.\n",
    "\n",
    "Now we can define Scaled F-Score as the harmonic mean of the Normal CDF transformed frequency and precision:\n",
    "\n",
    "$$ \\mbox{S-CAT}_{\\beta}(w_i, c_j) = \\mathcal{F}_{\\beta}(\\Phi(\\mbox{prec}(w_i, c_j)), \\Phi(\\mbox{freq}(w_i, c_j))).$$\n",
    "\n",
    "$\\mu$ and $\\sigma^2$ are defined separately as the mean and variance of precision and frequency.\n",
    "\n",
    "A $\\beta$ of 2 is recommended and is the default value in Scattertext.\n",
    "\n",
    "Note that any function with the range of $[0,1]$ (this includes the identity function) may be used in place of $\\Phi$.  Also, when the precision is very small (e.g., of a tiny minority class) normalization may be foregone."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Positive freq</th>\n",
       "      <th>Negative freq</th>\n",
       "      <th>pos_precision</th>\n",
       "      <th>pos_freq_pct</th>\n",
       "      <th>pos_hmean</th>\n",
       "      <th>pos_precision_normcdf</th>\n",
       "      <th>pos_freq_pct_normcdf</th>\n",
       "      <th>pos_scaled_f_score</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>term</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>the best</th>\n",
       "      <td>65</td>\n",
       "      <td>7</td>\n",
       "      <td>0.902778</td>\n",
       "      <td>0.000685</td>\n",
       "      <td>0.001370</td>\n",
       "      <td>0.800329</td>\n",
       "      <td>0.999876</td>\n",
       "      <td>0.889042</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>entertaining</th>\n",
       "      <td>58</td>\n",
       "      <td>13</td>\n",
       "      <td>0.816901</td>\n",
       "      <td>0.000612</td>\n",
       "      <td>0.001222</td>\n",
       "      <td>0.744974</td>\n",
       "      <td>0.999443</td>\n",
       "      <td>0.853648</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>heart</th>\n",
       "      <td>45</td>\n",
       "      <td>11</td>\n",
       "      <td>0.803571</td>\n",
       "      <td>0.000474</td>\n",
       "      <td>0.000948</td>\n",
       "      <td>0.735715</td>\n",
       "      <td>0.993981</td>\n",
       "      <td>0.845567</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>our</th>\n",
       "      <td>42</td>\n",
       "      <td>11</td>\n",
       "      <td>0.792453</td>\n",
       "      <td>0.000443</td>\n",
       "      <td>0.000885</td>\n",
       "      <td>0.727863</td>\n",
       "      <td>0.990311</td>\n",
       "      <td>0.839043</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ride</th>\n",
       "      <td>29</td>\n",
       "      <td>6</td>\n",
       "      <td>0.828571</td>\n",
       "      <td>0.000306</td>\n",
       "      <td>0.000611</td>\n",
       "      <td>0.752939</td>\n",
       "      <td>0.943992</td>\n",
       "      <td>0.837710</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>summer</th>\n",
       "      <td>29</td>\n",
       "      <td>6</td>\n",
       "      <td>0.828571</td>\n",
       "      <td>0.000306</td>\n",
       "      <td>0.000611</td>\n",
       "      <td>0.752939</td>\n",
       "      <td>0.943992</td>\n",
       "      <td>0.837710</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>flaws</th>\n",
       "      <td>19</td>\n",
       "      <td>1</td>\n",
       "      <td>0.950000</td>\n",
       "      <td>0.000200</td>\n",
       "      <td>0.000401</td>\n",
       "      <td>0.827414</td>\n",
       "      <td>0.844483</td>\n",
       "      <td>0.835861</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>moore</th>\n",
       "      <td>19</td>\n",
       "      <td>1</td>\n",
       "      <td>0.950000</td>\n",
       "      <td>0.000200</td>\n",
       "      <td>0.000401</td>\n",
       "      <td>0.827414</td>\n",
       "      <td>0.844483</td>\n",
       "      <td>0.835861</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>hilarious</th>\n",
       "      <td>28</td>\n",
       "      <td>6</td>\n",
       "      <td>0.823529</td>\n",
       "      <td>0.000295</td>\n",
       "      <td>0.000590</td>\n",
       "      <td>0.749514</td>\n",
       "      <td>0.937188</td>\n",
       "      <td>0.832910</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>delivers</th>\n",
       "      <td>25</td>\n",
       "      <td>5</td>\n",
       "      <td>0.833333</td>\n",
       "      <td>0.000264</td>\n",
       "      <td>0.000527</td>\n",
       "      <td>0.756150</td>\n",
       "      <td>0.912886</td>\n",
       "      <td>0.827159</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              Positive freq  Negative freq  pos_precision  pos_freq_pct  \\\n",
       "term                                                                      \n",
       "the best                 65              7       0.902778      0.000685   \n",
       "entertaining             58             13       0.816901      0.000612   \n",
       "heart                    45             11       0.803571      0.000474   \n",
       "our                      42             11       0.792453      0.000443   \n",
       "ride                     29              6       0.828571      0.000306   \n",
       "summer                   29              6       0.828571      0.000306   \n",
       "flaws                    19              1       0.950000      0.000200   \n",
       "moore                    19              1       0.950000      0.000200   \n",
       "hilarious                28              6       0.823529      0.000295   \n",
       "delivers                 25              5       0.833333      0.000264   \n",
       "\n",
       "              pos_hmean  pos_precision_normcdf  pos_freq_pct_normcdf  \\\n",
       "term                                                                   \n",
       "the best       0.001370               0.800329              0.999876   \n",
       "entertaining   0.001222               0.744974              0.999443   \n",
       "heart          0.000948               0.735715              0.993981   \n",
       "our            0.000885               0.727863              0.990311   \n",
       "ride           0.000611               0.752939              0.943992   \n",
       "summer         0.000611               0.752939              0.943992   \n",
       "flaws          0.000401               0.827414              0.844483   \n",
       "moore          0.000401               0.827414              0.844483   \n",
       "hilarious      0.000590               0.749514              0.937188   \n",
       "delivers       0.000527               0.756150              0.912886   \n",
       "\n",
       "              pos_scaled_f_score  \n",
       "term                              \n",
       "the best                0.889042  \n",
       "entertaining            0.853648  \n",
       "heart                   0.845567  \n",
       "our                     0.839043  \n",
       "ride                    0.837710  \n",
       "summer                  0.837710  \n",
       "flaws                   0.835861  \n",
       "moore                   0.835861  \n",
       "hilarious               0.832910  \n",
       "delivers                0.827159  "
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from scipy.stats import norm\n",
    "\n",
    "def normcdf(x):\n",
    "    return norm.cdf(x, x.mean(), x.std())\n",
    "\n",
    "term_freq_df['pos_precision_normcdf'] = normcdf(term_freq_df['pos_precision'])\n",
    "\n",
    "term_freq_df['pos_freq_pct_normcdf'] = normcdf(term_freq_df['pos_freq_pct'])\n",
    "\n",
    "term_freq_df['pos_scaled_f_score'] = hmean([term_freq_df['pos_precision_normcdf'], term_freq_df['pos_freq_pct_normcdf']])\n",
    "\n",
    "term_freq_df.sort_values(by='pos_scaled_f_score', ascending=False).iloc[:10]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Positive freq</th>\n",
       "      <th>Negative freq</th>\n",
       "      <th>pos_precision</th>\n",
       "      <th>pos_freq_pct</th>\n",
       "      <th>pos_hmean</th>\n",
       "      <th>pos_precision_normcdf</th>\n",
       "      <th>pos_freq_pct_normcdf</th>\n",
       "      <th>pos_scaled_f_score</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>term</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>bill perfectly</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>shoplifts</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>encyclopedia</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>homo eroticism</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>trade homo</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>rough trade</th>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>on rough</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>machismo in</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>gang machismo</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>teen gang</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                Positive freq  Negative freq  pos_precision  pos_freq_pct  \\\n",
       "term                                                                        \n",
       "bill perfectly              0              1            0.0           0.0   \n",
       "shoplifts                   0              1            0.0           0.0   \n",
       "encyclopedia                0              1            0.0           0.0   \n",
       "homo eroticism              0              1            0.0           0.0   \n",
       "trade homo                  0              1            0.0           0.0   \n",
       "rough trade                 0              2            0.0           0.0   \n",
       "on rough                    0              1            0.0           0.0   \n",
       "machismo in                 0              1            0.0           0.0   \n",
       "gang machismo               0              1            0.0           0.0   \n",
       "teen gang                   0              1            0.0           0.0   \n",
       "\n",
       "                pos_hmean  pos_precision_normcdf  pos_freq_pct_normcdf  \\\n",
       "term                                                                     \n",
       "bill perfectly        0.0               0.137436              0.467472   \n",
       "shoplifts             0.0               0.137436              0.467472   \n",
       "encyclopedia          0.0               0.137436              0.467472   \n",
       "homo eroticism        0.0               0.137436              0.467472   \n",
       "trade homo            0.0               0.137436              0.467472   \n",
       "rough trade           0.0               0.137436              0.467472   \n",
       "on rough              0.0               0.137436              0.467472   \n",
       "machismo in           0.0               0.137436              0.467472   \n",
       "gang machismo         0.0               0.137436              0.467472   \n",
       "teen gang             0.0               0.137436              0.467472   \n",
       "\n",
       "                pos_scaled_f_score  \n",
       "term                                \n",
       "bill perfectly            0.212421  \n",
       "shoplifts                 0.212421  \n",
       "encyclopedia              0.212421  \n",
       "homo eroticism            0.212421  \n",
       "trade homo                0.212421  \n",
       "rough trade               0.212421  \n",
       "on rough                  0.212421  \n",
       "machismo in               0.212421  \n",
       "gang machismo             0.212421  \n",
       "teen gang                 0.212421  "
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "term_freq_df.sort_values(by='pos_scaled_f_score', ascending=True).iloc[:10]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## A second problem: low-scores are infrequent, low precision words.\n",
    "### Solution: compute SFS of negative class. If that score has a higher magnitude than the positive SFS, keep that, but as a negative score.\n",
    "\n",
    "Define the Scaled F-Score for category $j$ as\n",
    "$$ \\mbox{S-CAT}^{j} =  \\mbox{S-CAT}_{\\beta}(w_i, c_j). $$\n",
    "\n",
    "Define a class $\\neg j$ which includes all categories other than $j$.\n",
    "\n",
    "and the Scaled F-Score for all other categories as\n",
    "$$ \\mbox{S-CAT}^{\\neg j} = \\mbox{S-CAT}_{\\beta}(w_i, c_{\\neg j}). $$\n",
    "\n",
    "Let the corrected version of Scaled F-Score be:\n",
    "\n",
    "$$\\mathcal{S}_{\\beta} = 2 \\cdot \\big(-0.5 + \\begin{cases}\n",
    "      \\mbox{S-CAT}^{j} & \\text{if}\\ \\mbox{S-CAT}^{j} > \\mbox{S-CAT}^{\\neg j}, \\\\\n",
    "      1 - \\mbox{S-CAT}^{\\neg j} & \\text{if}\\ \\mbox{S-CAT}^{j} < \\mbox{S-CAT}^{\\neg j}, \\\\\n",
    "      0 & \\text{otherwise}.\n",
    "    \\end{cases} \\big).$$\n",
    "    \n",
    "Note that the range of $\\mathcal{S}$ is now $[-1, 1]$, where $\\mathcal{S} < 0$ indicates a term less associated with the category is question than average, and a positive score being more associated."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Positive freq</th>\n",
       "      <th>Negative freq</th>\n",
       "      <th>pos_precision</th>\n",
       "      <th>pos_freq_pct</th>\n",
       "      <th>pos_hmean</th>\n",
       "      <th>pos_precision_normcdf</th>\n",
       "      <th>pos_freq_pct_normcdf</th>\n",
       "      <th>pos_scaled_f_score</th>\n",
       "      <th>neg_precision</th>\n",
       "      <th>neg_freq_pct</th>\n",
       "      <th>neg_scaled_f_score</th>\n",
       "      <th>scaled_f_score</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>term</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>the best</th>\n",
       "      <td>65</td>\n",
       "      <td>7</td>\n",
       "      <td>0.902778</td>\n",
       "      <td>0.000685</td>\n",
       "      <td>0.001370</td>\n",
       "      <td>0.800329</td>\n",
       "      <td>0.999876</td>\n",
       "      <td>0.889042</td>\n",
       "      <td>0.097222</td>\n",
       "      <td>0.000077</td>\n",
       "      <td>0.304090</td>\n",
       "      <td>0.778085</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>entertaining</th>\n",
       "      <td>58</td>\n",
       "      <td>13</td>\n",
       "      <td>0.816901</td>\n",
       "      <td>0.000612</td>\n",
       "      <td>0.001222</td>\n",
       "      <td>0.744974</td>\n",
       "      <td>0.999443</td>\n",
       "      <td>0.853648</td>\n",
       "      <td>0.183099</td>\n",
       "      <td>0.000142</td>\n",
       "      <td>0.382666</td>\n",
       "      <td>0.707296</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>heart</th>\n",
       "      <td>45</td>\n",
       "      <td>11</td>\n",
       "      <td>0.803571</td>\n",
       "      <td>0.000474</td>\n",
       "      <td>0.000948</td>\n",
       "      <td>0.735715</td>\n",
       "      <td>0.993981</td>\n",
       "      <td>0.845567</td>\n",
       "      <td>0.196429</td>\n",
       "      <td>0.000120</td>\n",
       "      <td>0.387529</td>\n",
       "      <td>0.691134</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>our</th>\n",
       "      <td>42</td>\n",
       "      <td>11</td>\n",
       "      <td>0.792453</td>\n",
       "      <td>0.000443</td>\n",
       "      <td>0.000885</td>\n",
       "      <td>0.727863</td>\n",
       "      <td>0.990311</td>\n",
       "      <td>0.839043</td>\n",
       "      <td>0.207547</td>\n",
       "      <td>0.000120</td>\n",
       "      <td>0.395904</td>\n",
       "      <td>0.678086</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>summer</th>\n",
       "      <td>29</td>\n",
       "      <td>6</td>\n",
       "      <td>0.828571</td>\n",
       "      <td>0.000306</td>\n",
       "      <td>0.000611</td>\n",
       "      <td>0.752939</td>\n",
       "      <td>0.943992</td>\n",
       "      <td>0.837710</td>\n",
       "      <td>0.171429</td>\n",
       "      <td>0.000066</td>\n",
       "      <td>0.352312</td>\n",
       "      <td>0.675421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ride</th>\n",
       "      <td>29</td>\n",
       "      <td>6</td>\n",
       "      <td>0.828571</td>\n",
       "      <td>0.000306</td>\n",
       "      <td>0.000611</td>\n",
       "      <td>0.752939</td>\n",
       "      <td>0.943992</td>\n",
       "      <td>0.837710</td>\n",
       "      <td>0.171429</td>\n",
       "      <td>0.000066</td>\n",
       "      <td>0.352312</td>\n",
       "      <td>0.675421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>flaws</th>\n",
       "      <td>19</td>\n",
       "      <td>1</td>\n",
       "      <td>0.950000</td>\n",
       "      <td>0.000200</td>\n",
       "      <td>0.000401</td>\n",
       "      <td>0.827414</td>\n",
       "      <td>0.844483</td>\n",
       "      <td>0.835861</td>\n",
       "      <td>0.050000</td>\n",
       "      <td>0.000011</td>\n",
       "      <td>0.255388</td>\n",
       "      <td>0.671722</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>moore</th>\n",
       "      <td>19</td>\n",
       "      <td>1</td>\n",
       "      <td>0.950000</td>\n",
       "      <td>0.000200</td>\n",
       "      <td>0.000401</td>\n",
       "      <td>0.827414</td>\n",
       "      <td>0.844483</td>\n",
       "      <td>0.835861</td>\n",
       "      <td>0.050000</td>\n",
       "      <td>0.000011</td>\n",
       "      <td>0.255388</td>\n",
       "      <td>0.671722</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>hilarious</th>\n",
       "      <td>28</td>\n",
       "      <td>6</td>\n",
       "      <td>0.823529</td>\n",
       "      <td>0.000295</td>\n",
       "      <td>0.000590</td>\n",
       "      <td>0.749514</td>\n",
       "      <td>0.937188</td>\n",
       "      <td>0.832910</td>\n",
       "      <td>0.176471</td>\n",
       "      <td>0.000066</td>\n",
       "      <td>0.355781</td>\n",
       "      <td>0.665820</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>delivers</th>\n",
       "      <td>25</td>\n",
       "      <td>5</td>\n",
       "      <td>0.833333</td>\n",
       "      <td>0.000264</td>\n",
       "      <td>0.000527</td>\n",
       "      <td>0.756150</td>\n",
       "      <td>0.912886</td>\n",
       "      <td>0.827159</td>\n",
       "      <td>0.166667</td>\n",
       "      <td>0.000055</td>\n",
       "      <td>0.345030</td>\n",
       "      <td>0.654317</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              Positive freq  Negative freq  pos_precision  pos_freq_pct  \\\n",
       "term                                                                      \n",
       "the best                 65              7       0.902778      0.000685   \n",
       "entertaining             58             13       0.816901      0.000612   \n",
       "heart                    45             11       0.803571      0.000474   \n",
       "our                      42             11       0.792453      0.000443   \n",
       "summer                   29              6       0.828571      0.000306   \n",
       "ride                     29              6       0.828571      0.000306   \n",
       "flaws                    19              1       0.950000      0.000200   \n",
       "moore                    19              1       0.950000      0.000200   \n",
       "hilarious                28              6       0.823529      0.000295   \n",
       "delivers                 25              5       0.833333      0.000264   \n",
       "\n",
       "              pos_hmean  pos_precision_normcdf  pos_freq_pct_normcdf  \\\n",
       "term                                                                   \n",
       "the best       0.001370               0.800329              0.999876   \n",
       "entertaining   0.001222               0.744974              0.999443   \n",
       "heart          0.000948               0.735715              0.993981   \n",
       "our            0.000885               0.727863              0.990311   \n",
       "summer         0.000611               0.752939              0.943992   \n",
       "ride           0.000611               0.752939              0.943992   \n",
       "flaws          0.000401               0.827414              0.844483   \n",
       "moore          0.000401               0.827414              0.844483   \n",
       "hilarious      0.000590               0.749514              0.937188   \n",
       "delivers       0.000527               0.756150              0.912886   \n",
       "\n",
       "              pos_scaled_f_score  neg_precision  neg_freq_pct  \\\n",
       "term                                                            \n",
       "the best                0.889042       0.097222      0.000077   \n",
       "entertaining            0.853648       0.183099      0.000142   \n",
       "heart                   0.845567       0.196429      0.000120   \n",
       "our                     0.839043       0.207547      0.000120   \n",
       "summer                  0.837710       0.171429      0.000066   \n",
       "ride                    0.837710       0.171429      0.000066   \n",
       "flaws                   0.835861       0.050000      0.000011   \n",
       "moore                   0.835861       0.050000      0.000011   \n",
       "hilarious               0.832910       0.176471      0.000066   \n",
       "delivers                0.827159       0.166667      0.000055   \n",
       "\n",
       "              neg_scaled_f_score  scaled_f_score  \n",
       "term                                              \n",
       "the best                0.304090        0.778085  \n",
       "entertaining            0.382666        0.707296  \n",
       "heart                   0.387529        0.691134  \n",
       "our                     0.395904        0.678086  \n",
       "summer                  0.352312        0.675421  \n",
       "ride                    0.352312        0.675421  \n",
       "flaws                   0.255388        0.671722  \n",
       "moore                   0.255388        0.671722  \n",
       "hilarious               0.355781        0.665820  \n",
       "delivers                0.345030        0.654317  "
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "term_freq_df['neg_precision'] = (term_freq_df['Negative freq'] * 1./\n",
    "                                 (term_freq_df['Negative freq'] + term_freq_df['Positive freq']))\n",
    "\n",
    "term_freq_df['neg_freq_pct'] = (term_freq_df['Negative freq'] * 1.\n",
    "                                /term_freq_df['Negative freq'].sum())\n",
    "\n",
    "term_freq_df['neg_scaled_f_score'] = hmean([normcdf(term_freq_df['neg_precision']),  normcdf(term_freq_df['neg_freq_pct'])])\n",
    "\n",
    "term_freq_df['scaled_f_score'] = 0\n",
    "term_freq_df.loc[term_freq_df['pos_scaled_f_score'] > term_freq_df['neg_scaled_f_score'], \n",
    "                 'scaled_f_score'] = term_freq_df['pos_scaled_f_score']\n",
    "term_freq_df.loc[term_freq_df['pos_scaled_f_score'] < term_freq_df['neg_scaled_f_score'], \n",
    "                 'scaled_f_score'] = 1-term_freq_df['neg_scaled_f_score']\n",
    "term_freq_df['scaled_f_score'] = 2 * (term_freq_df['scaled_f_score'] - 0.5)\n",
    "term_freq_df.sort_values(by='scaled_f_score', ascending=False).iloc[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Positive freq</th>\n",
       "      <th>Negative freq</th>\n",
       "      <th>pos_precision</th>\n",
       "      <th>pos_freq_pct</th>\n",
       "      <th>pos_hmean</th>\n",
       "      <th>pos_precision_normcdf</th>\n",
       "      <th>pos_freq_pct_normcdf</th>\n",
       "      <th>pos_scaled_f_score</th>\n",
       "      <th>neg_precision</th>\n",
       "      <th>neg_freq_pct</th>\n",
       "      <th>neg_scaled_f_score</th>\n",
       "      <th>scaled_f_score</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>term</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>bad</th>\n",
       "      <td>17</td>\n",
       "      <td>105</td>\n",
       "      <td>0.139344</td>\n",
       "      <td>0.000179</td>\n",
       "      <td>0.000358</td>\n",
       "      <td>0.213805</td>\n",
       "      <td>0.815360</td>\n",
       "      <td>0.338776</td>\n",
       "      <td>0.860656</td>\n",
       "      <td>0.001149</td>\n",
       "      <td>0.880301</td>\n",
       "      <td>-0.760603</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>worst</th>\n",
       "      <td>2</td>\n",
       "      <td>23</td>\n",
       "      <td>0.080000</td>\n",
       "      <td>0.000021</td>\n",
       "      <td>0.000042</td>\n",
       "      <td>0.178665</td>\n",
       "      <td>0.513404</td>\n",
       "      <td>0.265081</td>\n",
       "      <td>0.920000</td>\n",
       "      <td>0.000252</td>\n",
       "      <td>0.864072</td>\n",
       "      <td>-0.728143</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>nor</th>\n",
       "      <td>3</td>\n",
       "      <td>25</td>\n",
       "      <td>0.107143</td>\n",
       "      <td>0.000032</td>\n",
       "      <td>0.000063</td>\n",
       "      <td>0.194262</td>\n",
       "      <td>0.536341</td>\n",
       "      <td>0.285218</td>\n",
       "      <td>0.892857</td>\n",
       "      <td>0.000274</td>\n",
       "      <td>0.863366</td>\n",
       "      <td>-0.726733</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>a bad</th>\n",
       "      <td>2</td>\n",
       "      <td>22</td>\n",
       "      <td>0.083333</td>\n",
       "      <td>0.000021</td>\n",
       "      <td>0.000042</td>\n",
       "      <td>0.180537</td>\n",
       "      <td>0.513404</td>\n",
       "      <td>0.267136</td>\n",
       "      <td>0.916667</td>\n",
       "      <td>0.000241</td>\n",
       "      <td>0.858329</td>\n",
       "      <td>-0.716659</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>neither</th>\n",
       "      <td>3</td>\n",
       "      <td>24</td>\n",
       "      <td>0.111111</td>\n",
       "      <td>0.000032</td>\n",
       "      <td>0.000063</td>\n",
       "      <td>0.196610</td>\n",
       "      <td>0.536341</td>\n",
       "      <td>0.287741</td>\n",
       "      <td>0.888889</td>\n",
       "      <td>0.000263</td>\n",
       "      <td>0.858224</td>\n",
       "      <td>-0.716448</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>instead</th>\n",
       "      <td>7</td>\n",
       "      <td>32</td>\n",
       "      <td>0.179487</td>\n",
       "      <td>0.000074</td>\n",
       "      <td>0.000148</td>\n",
       "      <td>0.239703</td>\n",
       "      <td>0.626152</td>\n",
       "      <td>0.346687</td>\n",
       "      <td>0.820513</td>\n",
       "      <td>0.000350</td>\n",
       "      <td>0.853228</td>\n",
       "      <td>-0.706456</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>obvious</th>\n",
       "      <td>3</td>\n",
       "      <td>23</td>\n",
       "      <td>0.115385</td>\n",
       "      <td>0.000032</td>\n",
       "      <td>0.000063</td>\n",
       "      <td>0.199158</td>\n",
       "      <td>0.536341</td>\n",
       "      <td>0.290460</td>\n",
       "      <td>0.884615</td>\n",
       "      <td>0.000252</td>\n",
       "      <td>0.852596</td>\n",
       "      <td>-0.705191</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mess</th>\n",
       "      <td>1</td>\n",
       "      <td>19</td>\n",
       "      <td>0.050000</td>\n",
       "      <td>0.000011</td>\n",
       "      <td>0.000021</td>\n",
       "      <td>0.162372</td>\n",
       "      <td>0.490422</td>\n",
       "      <td>0.243969</td>\n",
       "      <td>0.950000</td>\n",
       "      <td>0.000208</td>\n",
       "      <td>0.850818</td>\n",
       "      <td>-0.701637</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>boring</th>\n",
       "      <td>0</td>\n",
       "      <td>17</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.137436</td>\n",
       "      <td>0.467472</td>\n",
       "      <td>0.212421</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000186</td>\n",
       "      <td>0.848754</td>\n",
       "      <td>-0.697508</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>the only</th>\n",
       "      <td>3</td>\n",
       "      <td>22</td>\n",
       "      <td>0.120000</td>\n",
       "      <td>0.000032</td>\n",
       "      <td>0.000063</td>\n",
       "      <td>0.201931</td>\n",
       "      <td>0.536341</td>\n",
       "      <td>0.293398</td>\n",
       "      <td>0.880000</td>\n",
       "      <td>0.000241</td>\n",
       "      <td>0.846446</td>\n",
       "      <td>-0.692891</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          Positive freq  Negative freq  pos_precision  pos_freq_pct  \\\n",
       "term                                                                  \n",
       "bad                  17            105       0.139344      0.000179   \n",
       "worst                 2             23       0.080000      0.000021   \n",
       "nor                   3             25       0.107143      0.000032   \n",
       "a bad                 2             22       0.083333      0.000021   \n",
       "neither               3             24       0.111111      0.000032   \n",
       "instead               7             32       0.179487      0.000074   \n",
       "obvious               3             23       0.115385      0.000032   \n",
       "mess                  1             19       0.050000      0.000011   \n",
       "boring                0             17       0.000000      0.000000   \n",
       "the only              3             22       0.120000      0.000032   \n",
       "\n",
       "          pos_hmean  pos_precision_normcdf  pos_freq_pct_normcdf  \\\n",
       "term                                                               \n",
       "bad        0.000358               0.213805              0.815360   \n",
       "worst      0.000042               0.178665              0.513404   \n",
       "nor        0.000063               0.194262              0.536341   \n",
       "a bad      0.000042               0.180537              0.513404   \n",
       "neither    0.000063               0.196610              0.536341   \n",
       "instead    0.000148               0.239703              0.626152   \n",
       "obvious    0.000063               0.199158              0.536341   \n",
       "mess       0.000021               0.162372              0.490422   \n",
       "boring     0.000000               0.137436              0.467472   \n",
       "the only   0.000063               0.201931              0.536341   \n",
       "\n",
       "          pos_scaled_f_score  neg_precision  neg_freq_pct  neg_scaled_f_score  \\\n",
       "term                                                                            \n",
       "bad                 0.338776       0.860656      0.001149            0.880301   \n",
       "worst               0.265081       0.920000      0.000252            0.864072   \n",
       "nor                 0.285218       0.892857      0.000274            0.863366   \n",
       "a bad               0.267136       0.916667      0.000241            0.858329   \n",
       "neither             0.287741       0.888889      0.000263            0.858224   \n",
       "instead             0.346687       0.820513      0.000350            0.853228   \n",
       "obvious             0.290460       0.884615      0.000252            0.852596   \n",
       "mess                0.243969       0.950000      0.000208            0.850818   \n",
       "boring              0.212421       1.000000      0.000186            0.848754   \n",
       "the only            0.293398       0.880000      0.000241            0.846446   \n",
       "\n",
       "          scaled_f_score  \n",
       "term                      \n",
       "bad            -0.760603  \n",
       "worst          -0.728143  \n",
       "nor            -0.726733  \n",
       "a bad          -0.716659  \n",
       "neither        -0.716448  \n",
       "instead        -0.706456  \n",
       "obvious        -0.705191  \n",
       "mess           -0.701637  \n",
       "boring         -0.697508  \n",
       "the only       -0.692891  "
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "term_freq_df.sort_values(by='scaled_f_score', ascending=True).iloc[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1300\"\n",
       "            height=\"700\"\n",
       "            src=\"output/rotten_fresh_scaled_f_score.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x1226d8278>"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "html = st.produce_fightin_words_explorer(\n",
    "    corpus,\n",
    "    category='Positive',\n",
    "    not_category_name='Negative',\n",
    "    not_categories=['Negative'],\n",
    "    term_scorer=st.ScaledFScorePresets(beta=1, one_to_neg_one=True),\n",
    "    metadata = rdf['movie_name'],\n",
    "    grey_threshold=0\n",
    ")\n",
    "file_name = 'output/rotten_fresh_scaled_f_score.html'\n",
    "open(file_name, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=file_name, width = 1300, height=700)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### The Semiotic Square using RankDiff and phrases from PhraseMachine"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "nlp = spacy.load('en')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "rdf['parse'] = rdf.text.apply(nlp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [],
   "source": [
    "phrase_corpus = (st.CorpusFromParsedDocuments(rdf, \n",
    "                                       category_col='category_name', \n",
    "                                       parsed_col='parse',\n",
    "                                       feats_from_spacy_doc = st.PhraseMachinePhrases())\n",
    "          .build())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [],
   "source": [
    "phrase_corpus_compact = phrase_corpus.compact(st.CompactTerms(minimum_term_count=2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1600\"\n",
       "            height=\"900\"\n",
       "            src=\"semiotic_square_rankdiff.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x12296eeb8>"
      ]
     },
     "execution_count": 98,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "semiotic_square = st.SemioticSquare(\n",
    "    phrase_corpus_compact,\n",
    "    category_a='Positive',\n",
    "    category_b='Negative',\n",
    "    neutral_categories=['Plot'],\n",
    "    scorer=st.RankDifference(),\n",
    "    labels = {'a_and_b': 'Reviews',\n",
    "              'not_a_and_not_b': 'Plot Descriptions',\n",
    "              'a_and_not_b': 'Positive/Plot',\n",
    "              'b_and_not_a': 'Negative/Plot',\n",
    "             }\n",
    ")\n",
    "\n",
    "html = st.produce_semiotic_square_explorer(semiotic_square,\n",
    "                                           category_name='Positive',\n",
    "                                           not_category_name='Negative',\n",
    "                                           x_label='Positive-Negative',\n",
    "                                           y_label='Review-Plot',\n",
    "                                           minimum_term_frequency=2,\n",
    "                                           pmi_threshold_coefficient=0,\n",
    "                                           neutral_category_name='Plot Description',\n",
    "                                           metadata=rdf['movie_name'])\n",
    "\n",
    "fn = 'semiotic_square_rankdiff.html'\n",
    "open(fn, 'wb').write(html.encode('utf-8'))\n",
    "IFrame(src=fn, width = 1600, height=900)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "unigram_corpus_compact = unigram_corpus.compact(st.ClassPercentageCompactor(term_count=2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1730993"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "unigram_semiotic_square = st.SemioticSquare(\n",
    "    unigram_corpus_compact,\n",
    "    category_a='Positive',\n",
    "    category_b='Negative',\n",
    "    neutral_categories=['Plot'],\n",
    "    scorer=st.RankDifference(),\n",
    "    labels = {'a_and_b': 'Reviews',\n",
    "              'not_a_and_not_b': 'Plot Descriptions',\n",
    "              'a_and_not_b': 'Positive/Plot',\n",
    "              'b_and_not_a': 'Negative/Plot',\n",
    "             }\n",
    ")\n",
    "\n",
    "html = st.produce_semiotic_square_explorer(unigram_semiotic_square,\n",
    "                                           category_name='Positive',\n",
    "                                           not_category_name='Negative',\n",
    "                                           x_label='Positive-Negative',\n",
    "                                           y_label='Review-Plot',\n",
    "                                           minimum_term_frequency=2,\n",
    "                                           pmi_threshold_coefficient=0,\n",
    "                                           neutral_category_name='Plot Description',\n",
    "                                           metadata=rdf['movie_name'])\n",
    "\n",
    "fn = 'semiotic_square_rankdiff_unigram.html'\n",
    "open(fn, 'wb').write(html.encode('utf-8'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"1600\"\n",
       "            height=\"900\"\n",
       "            src=\"semiotic_square_rankdiff_unigram.html\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.IFrame at 0x1135d0e10>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "IFrame(src=fn, width = 1600, height=900)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {
   "attach-environment": true,
   "environment": "Root",
   "summary": "Class Association Scores with Scattertext"
  },
  "kernelspec": {
   "display_name": "Python [py36]",
   "language": "python",
   "name": "Python [py36]"
  },
  "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
}