notebooks/Mass Evaluation Result Analysis.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "def warn(*args, **kwargs):\n",
    "    pass\n",
    "import warnings\n",
    "warnings.warn = warn\n",
    "import copy\n",
    "import json\n",
    "from pathlib import Path\n",
    "from textwrap import wrap\n",
    "  \n",
    "\n",
    "import pandas as pd\n",
    "from sklearn.metrics import accuracy_score, f1_score\n",
    "from bella.helper import read_config\n",
    "from bella.parsers import semeval_14, dong, election\n",
    "from bella import evaluation\n",
    "from bella.evaluation import distinct_sentiment_metrics\n",
    "import seaborn as sns\n",
    "from matplotlib import pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "##\n",
    "#  ADD YOUR CONFIG FILE PATH HERE \n",
    "##\n",
    "CONFIG_FP = Path('..', 'config.yaml')\n",
    "# Load all of the datasets\n",
    "youtubean_test = semeval_14(read_config('youtubean_test', CONFIG_FP))\n",
    "semeval_14_rest_test = semeval_14(read_config('semeval_2014_rest_test', CONFIG_FP))\n",
    "semeval_14_lap_test = semeval_14(read_config('semeval_2014_lap_test', CONFIG_FP))\n",
    "dong_test = dong(read_config('dong_twit_test_data', CONFIG_FP))\n",
    "_, election_test = election(read_config('election_folder_dir', CONFIG_FP))\n",
    "mitchel_test = semeval_14(read_config('mitchel_test', CONFIG_FP))\n",
    "dataset_test = {'SemEval 14 Laptop': semeval_14_lap_test,\n",
    "                'SemEval 14 Restaurant': semeval_14_rest_test,\n",
    "                'Dong Twitter': dong_test,\n",
    "                'Election Twitter': election_test,\n",
    "                'YouTuBean': youtubean_test,\n",
    "                'Mitchel': mitchel_test}\n",
    "# results folders\n",
    "results_folder = Path(read_config('results_folder', CONFIG_FP))\n",
    "lstms_folder = results_folder.joinpath('TDLstm')\n",
    "target_dep_folder = results_folder.joinpath('Target Dependent')\n",
    "tdparse_folder = results_folder.joinpath('TDParse')\n",
    "# Result files\n",
    "lstm_name_fp = ('LSTM', Path(lstms_folder, 'LSTM', 'dataset predictions.json'))\n",
    "tdlstm_name_fp = ('TDLSTM', Path(lstms_folder, 'TDLSTM', 'dataset predictions.json'))\n",
    "tclstm_name_fp = ('TCLSTM', Path(lstms_folder, 'TCLSTM', 'dataset predictions.json'))\n",
    "targ_dep_name_fp = ('Target-Dep', \n",
    "                    Path(target_dep_folder, 'Target Dependent', 'dataset predictions.json'))\n",
    "targ_plus_name_fp = ('Target-Dep+', Path(target_dep_folder, 'Target Dependent Plus', 'dataset predictions.json'))\n",
    "tdparse_name_fp = ('TDParse', Path(tdparse_folder, 'TDParse', 'dataset predictions.json'))\n",
    "tdparse_plus_name_fp = ('TDParse+', Path(tdparse_folder, 'TDParsePlus', 'dataset predictions.json'))\n",
    "all_results = [lstm_name_fp, tdlstm_name_fp, tclstm_name_fp, targ_dep_name_fp, targ_plus_name_fp,\n",
    "               tdparse_name_fp, tdparse_plus_name_fp]\n",
    "all_results = {name: json.load(results_fp.open('r')) for name, results_fp in all_results}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Mass Evaluation Result analysis\n",
    "\n",
    "## Macro F1 3-Way"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "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>LSTM</th>\n",
       "      <th>TDLSTM</th>\n",
       "      <th>TCLSTM</th>\n",
       "      <th>Target-Dep</th>\n",
       "      <th>Target-Dep+</th>\n",
       "      <th>TDParse</th>\n",
       "      <th>TDParse+</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Dong Twitter</th>\n",
       "      <td>63.60</td>\n",
       "      <td>66.09</td>\n",
       "      <td>67.14</td>\n",
       "      <td>65.66</td>\n",
       "      <td>65.67</td>\n",
       "      <td>65.74</td>\n",
       "      <td>68.06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Election Twitter</th>\n",
       "      <td>38.70</td>\n",
       "      <td>43.60</td>\n",
       "      <td>42.08</td>\n",
       "      <td>45.47</td>\n",
       "      <td>45.93</td>\n",
       "      <td>46.32</td>\n",
       "      <td>44.93</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mitchel</th>\n",
       "      <td>47.17</td>\n",
       "      <td>51.16</td>\n",
       "      <td>41.03</td>\n",
       "      <td>40.76</td>\n",
       "      <td>42.86</td>\n",
       "      <td>40.96</td>\n",
       "      <td>50.03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SemEval 14 Laptop</th>\n",
       "      <td>47.84</td>\n",
       "      <td>57.91</td>\n",
       "      <td>46.80</td>\n",
       "      <td>59.97</td>\n",
       "      <td>63.73</td>\n",
       "      <td>59.98</td>\n",
       "      <td>63.85</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SemEval 14 Restaurant</th>\n",
       "      <td>46.36</td>\n",
       "      <td>57.68</td>\n",
       "      <td>55.38</td>\n",
       "      <td>56.16</td>\n",
       "      <td>57.71</td>\n",
       "      <td>59.49</td>\n",
       "      <td>61.11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>YouTuBean</th>\n",
       "      <td>45.93</td>\n",
       "      <td>45.47</td>\n",
       "      <td>38.07</td>\n",
       "      <td>53.14</td>\n",
       "      <td>55.56</td>\n",
       "      <td>71.67</td>\n",
       "      <td>67.68</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mean</th>\n",
       "      <td>48.27</td>\n",
       "      <td>53.65</td>\n",
       "      <td>48.42</td>\n",
       "      <td>53.53</td>\n",
       "      <td>55.24</td>\n",
       "      <td>57.36</td>\n",
       "      <td>59.28</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                        LSTM  TDLSTM  TCLSTM  Target-Dep  Target-Dep+  \\\n",
       "Dong Twitter           63.60   66.09   67.14       65.66        65.67   \n",
       "Election Twitter       38.70   43.60   42.08       45.47        45.93   \n",
       "Mitchel                47.17   51.16   41.03       40.76        42.86   \n",
       "SemEval 14 Laptop      47.84   57.91   46.80       59.97        63.73   \n",
       "SemEval 14 Restaurant  46.36   57.68   55.38       56.16        57.71   \n",
       "YouTuBean              45.93   45.47   38.07       53.14        55.56   \n",
       "Mean                   48.27   53.65   48.42       53.53        55.24   \n",
       "\n",
       "                       TDParse  TDParse+  \n",
       "Dong Twitter             65.74     68.06  \n",
       "Election Twitter         46.32     44.93  \n",
       "Mitchel                  40.96     50.03  \n",
       "SemEval 14 Laptop        59.98     63.85  \n",
       "SemEval 14 Restaurant    59.49     61.11  \n",
       "YouTuBean                71.67     67.68  \n",
       "Mean                     57.36     59.28  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "f1_results = evaluation.evaluate_models(f1_score, dataset_test, all_results, \n",
    "                                        dataframe=True, average='macro')\n",
    "(f1_results * 100).round(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'Macro F1')"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x504 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "vis_f1 = copy.deepcopy(f1_results)\n",
    "vis_f1.columns = [col.replace(' 3 Class Macro F1', '') for col in vis_f1.columns]\n",
    "vis_f1.index = ['\\n'.join(wrap(index, 10)) for index in vis_f1.index]\n",
    "fig, ax = plt.subplots(figsize=(9,7))\n",
    "sns.heatmap(vis_f1, annot=True, annot_kws={'size':15},\n",
    "            cmap=sns.color_palette(\"Blues\"), ax=ax)\n",
    "ax.tick_params(axis='both', which='major', labelsize=15)\n",
    "ax.set_title('Macro F1', fontdict={'fontsize':20})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "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>LSTM</th>\n",
       "      <th>TDLSTM</th>\n",
       "      <th>TCLSTM</th>\n",
       "      <th>Target-Dep</th>\n",
       "      <th>Target-Dep+</th>\n",
       "      <th>TDParse</th>\n",
       "      <th>TDParse+</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Dong Twitter</th>\n",
       "      <td>66.47</td>\n",
       "      <td>68.79</td>\n",
       "      <td>69.51</td>\n",
       "      <td>67.34</td>\n",
       "      <td>67.77</td>\n",
       "      <td>68.35</td>\n",
       "      <td>70.09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Election Twitter</th>\n",
       "      <td>53.80</td>\n",
       "      <td>57.30</td>\n",
       "      <td>57.22</td>\n",
       "      <td>57.65</td>\n",
       "      <td>56.63</td>\n",
       "      <td>57.22</td>\n",
       "      <td>56.40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mitchel</th>\n",
       "      <td>72.14</td>\n",
       "      <td>73.45</td>\n",
       "      <td>69.81</td>\n",
       "      <td>72.64</td>\n",
       "      <td>72.85</td>\n",
       "      <td>73.25</td>\n",
       "      <td>74.47</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SemEval 14 Laptop</th>\n",
       "      <td>58.93</td>\n",
       "      <td>65.52</td>\n",
       "      <td>57.05</td>\n",
       "      <td>67.87</td>\n",
       "      <td>70.85</td>\n",
       "      <td>68.50</td>\n",
       "      <td>70.69</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SemEval 14 Restaurant</th>\n",
       "      <td>71.07</td>\n",
       "      <td>74.38</td>\n",
       "      <td>72.41</td>\n",
       "      <td>73.84</td>\n",
       "      <td>74.64</td>\n",
       "      <td>73.39</td>\n",
       "      <td>76.16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>YouTuBean</th>\n",
       "      <td>72.50</td>\n",
       "      <td>72.50</td>\n",
       "      <td>66.67</td>\n",
       "      <td>70.83</td>\n",
       "      <td>72.50</td>\n",
       "      <td>81.25</td>\n",
       "      <td>80.42</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mean</th>\n",
       "      <td>65.82</td>\n",
       "      <td>68.66</td>\n",
       "      <td>65.44</td>\n",
       "      <td>68.36</td>\n",
       "      <td>69.21</td>\n",
       "      <td>70.33</td>\n",
       "      <td>71.37</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                        LSTM  TDLSTM  TCLSTM  Target-Dep  Target-Dep+  \\\n",
       "Dong Twitter           66.47   68.79   69.51       67.34        67.77   \n",
       "Election Twitter       53.80   57.30   57.22       57.65        56.63   \n",
       "Mitchel                72.14   73.45   69.81       72.64        72.85   \n",
       "SemEval 14 Laptop      58.93   65.52   57.05       67.87        70.85   \n",
       "SemEval 14 Restaurant  71.07   74.38   72.41       73.84        74.64   \n",
       "YouTuBean              72.50   72.50   66.67       70.83        72.50   \n",
       "Mean                   65.82   68.66   65.44       68.36        69.21   \n",
       "\n",
       "                       TDParse  TDParse+  \n",
       "Dong Twitter             68.35     70.09  \n",
       "Election Twitter         57.22     56.40  \n",
       "Mitchel                  73.25     74.47  \n",
       "SemEval 14 Laptop        68.50     70.69  \n",
       "SemEval 14 Restaurant    73.39     76.16  \n",
       "YouTuBean                81.25     80.42  \n",
       "Mean                     70.33     71.37  "
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "acc_results = evaluation.evaluate_models(accuracy_score, dataset_test, \n",
    "                                         all_results, dataframe=True)\n",
    "(acc_results * 100).round(2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Macro F1 2-way\n",
    "\n",
    "We can then look at the 2 way Macro F1 score which evaluates how well the models can predict Positive or Negative sentiment while still taking into account prediciting the neutral class."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "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>LSTM</th>\n",
       "      <th>TDLSTM</th>\n",
       "      <th>TCLSTM</th>\n",
       "      <th>Target-Dep</th>\n",
       "      <th>Target-Dep+</th>\n",
       "      <th>TDParse</th>\n",
       "      <th>TDParse+</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Dong Twitter</th>\n",
       "      <td>59.19</td>\n",
       "      <td>61.59</td>\n",
       "      <td>62.94</td>\n",
       "      <td>62.39</td>\n",
       "      <td>61.92</td>\n",
       "      <td>61.33</td>\n",
       "      <td>64.44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Election Twitter</th>\n",
       "      <td>34.02</td>\n",
       "      <td>36.85</td>\n",
       "      <td>34.39</td>\n",
       "      <td>39.97</td>\n",
       "      <td>41.02</td>\n",
       "      <td>41.27</td>\n",
       "      <td>39.49</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mitchel</th>\n",
       "      <td>29.03</td>\n",
       "      <td>34.49</td>\n",
       "      <td>20.42</td>\n",
       "      <td>19.22</td>\n",
       "      <td>22.53</td>\n",
       "      <td>19.47</td>\n",
       "      <td>32.73</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SemEval 14 Laptop</th>\n",
       "      <td>64.46</td>\n",
       "      <td>65.29</td>\n",
       "      <td>57.36</td>\n",
       "      <td>68.62</td>\n",
       "      <td>71.82</td>\n",
       "      <td>69.79</td>\n",
       "      <td>71.97</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SemEval 14 Restaurant</th>\n",
       "      <td>69.54</td>\n",
       "      <td>71.08</td>\n",
       "      <td>68.35</td>\n",
       "      <td>73.61</td>\n",
       "      <td>73.78</td>\n",
       "      <td>70.90</td>\n",
       "      <td>76.14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>YouTuBean</th>\n",
       "      <td>27.05</td>\n",
       "      <td>26.27</td>\n",
       "      <td>17.71</td>\n",
       "      <td>38.99</td>\n",
       "      <td>42.34</td>\n",
       "      <td>64.02</td>\n",
       "      <td>58.22</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mean</th>\n",
       "      <td>47.22</td>\n",
       "      <td>49.26</td>\n",
       "      <td>43.53</td>\n",
       "      <td>50.47</td>\n",
       "      <td>52.23</td>\n",
       "      <td>54.46</td>\n",
       "      <td>57.17</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                        LSTM  TDLSTM  TCLSTM  Target-Dep  Target-Dep+  \\\n",
       "Dong Twitter           59.19   61.59   62.94       62.39        61.92   \n",
       "Election Twitter       34.02   36.85   34.39       39.97        41.02   \n",
       "Mitchel                29.03   34.49   20.42       19.22        22.53   \n",
       "SemEval 14 Laptop      64.46   65.29   57.36       68.62        71.82   \n",
       "SemEval 14 Restaurant  69.54   71.08   68.35       73.61        73.78   \n",
       "YouTuBean              27.05   26.27   17.71       38.99        42.34   \n",
       "Mean                   47.22   49.26   43.53       50.47        52.23   \n",
       "\n",
       "                       TDParse  TDParse+  \n",
       "Dong Twitter             61.33     64.44  \n",
       "Election Twitter         41.27     39.49  \n",
       "Mitchel                  19.47     32.73  \n",
       "SemEval 14 Laptop        69.79     71.97  \n",
       "SemEval 14 Restaurant    70.90     76.14  \n",
       "YouTuBean                64.02     58.22  \n",
       "Mean                     54.46     57.17  "
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(evaluation.evaluate_models(f1_score, dataset_test, all_results, \n",
    "                           dataframe=True, average='macro', labels=[-1, 1]) * 100).round(2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Macro F1, 1 Distinct Sentiment subset\n",
    "\n",
    "Now we look at the results of trying to predict the sentiment of all targets that are only in sentence that contain one distinct sentiment. This should be a result that is good for the LSTM model as it does not take into account any target information. However as we can see below the LSTM is one of the worse models."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "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>LSTM</th>\n",
       "      <th>TDLSTM</th>\n",
       "      <th>TCLSTM</th>\n",
       "      <th>Target-Dep</th>\n",
       "      <th>Target-Dep+</th>\n",
       "      <th>TDParse</th>\n",
       "      <th>TDParse+</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Dong Twitter</th>\n",
       "      <td>63.60</td>\n",
       "      <td>66.09</td>\n",
       "      <td>67.14</td>\n",
       "      <td>65.66</td>\n",
       "      <td>65.67</td>\n",
       "      <td>65.74</td>\n",
       "      <td>68.06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Election Twitter</th>\n",
       "      <td>45.09</td>\n",
       "      <td>47.29</td>\n",
       "      <td>42.55</td>\n",
       "      <td>50.25</td>\n",
       "      <td>50.33</td>\n",
       "      <td>48.83</td>\n",
       "      <td>48.45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mitchel</th>\n",
       "      <td>47.64</td>\n",
       "      <td>52.41</td>\n",
       "      <td>41.13</td>\n",
       "      <td>41.52</td>\n",
       "      <td>43.13</td>\n",
       "      <td>41.82</td>\n",
       "      <td>51.27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SemEval 14 Laptop</th>\n",
       "      <td>49.06</td>\n",
       "      <td>58.30</td>\n",
       "      <td>47.61</td>\n",
       "      <td>61.03</td>\n",
       "      <td>67.18</td>\n",
       "      <td>60.75</td>\n",
       "      <td>66.15</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SemEval 14 Restaurant</th>\n",
       "      <td>48.19</td>\n",
       "      <td>59.89</td>\n",
       "      <td>54.73</td>\n",
       "      <td>57.17</td>\n",
       "      <td>60.52</td>\n",
       "      <td>57.77</td>\n",
       "      <td>62.93</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>YouTuBean</th>\n",
       "      <td>46.68</td>\n",
       "      <td>46.22</td>\n",
       "      <td>38.66</td>\n",
       "      <td>54.22</td>\n",
       "      <td>57.02</td>\n",
       "      <td>73.42</td>\n",
       "      <td>69.23</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mean</th>\n",
       "      <td>50.04</td>\n",
       "      <td>55.03</td>\n",
       "      <td>48.64</td>\n",
       "      <td>54.98</td>\n",
       "      <td>57.31</td>\n",
       "      <td>58.05</td>\n",
       "      <td>61.02</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                        LSTM  TDLSTM  TCLSTM  Target-Dep  Target-Dep+  \\\n",
       "Dong Twitter           63.60   66.09   67.14       65.66        65.67   \n",
       "Election Twitter       45.09   47.29   42.55       50.25        50.33   \n",
       "Mitchel                47.64   52.41   41.13       41.52        43.13   \n",
       "SemEval 14 Laptop      49.06   58.30   47.61       61.03        67.18   \n",
       "SemEval 14 Restaurant  48.19   59.89   54.73       57.17        60.52   \n",
       "YouTuBean              46.68   46.22   38.66       54.22        57.02   \n",
       "Mean                   50.04   55.03   48.64       54.98        57.31   \n",
       "\n",
       "                       TDParse  TDParse+  \n",
       "Dong Twitter             65.74     68.06  \n",
       "Election Twitter         48.83     48.45  \n",
       "Mitchel                  41.82     51.27  \n",
       "SemEval 14 Laptop        60.75     66.15  \n",
       "SemEval 14 Restaurant    57.77     62.93  \n",
       "YouTuBean                73.42     69.23  \n",
       "Mean                     58.05     61.02  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(evaluation.evaluate_models(distinct_sentiment_metrics, dataset_test, all_results, custom_metric=True,\n",
    "                            dataframe=True, distinct_sentiments=1, \n",
    "                            metric_func=f1_score, average='macro') * 100).round(2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Macro F1, 2 Distinct Sentiment subset\n",
    "\n",
    "We do the same as above but for 2 distinct sentiments per sentence. We remove the Dong twitter dataset from this experiment due to it only containing Tweets with one distinct sentiment."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/andrew/Envs/Bella/lib/python3.6/site-packages/numpy/core/_methods.py:80: RuntimeWarning: invalid value encountered in double_scalars\n",
      "  ret = ret.dtype.type(ret / rcount)\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>LSTM</th>\n",
       "      <th>TDLSTM</th>\n",
       "      <th>TCLSTM</th>\n",
       "      <th>Target-Dep</th>\n",
       "      <th>Target-Dep+</th>\n",
       "      <th>TDParse</th>\n",
       "      <th>TDParse+</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Election Twitter</th>\n",
       "      <td>34.12</td>\n",
       "      <td>39.89</td>\n",
       "      <td>41.29</td>\n",
       "      <td>41.91</td>\n",
       "      <td>43.03</td>\n",
       "      <td>44.00</td>\n",
       "      <td>42.18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mitchel</th>\n",
       "      <td>38.00</td>\n",
       "      <td>36.24</td>\n",
       "      <td>39.21</td>\n",
       "      <td>31.60</td>\n",
       "      <td>34.83</td>\n",
       "      <td>30.65</td>\n",
       "      <td>37.92</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SemEval 14 Laptop</th>\n",
       "      <td>33.67</td>\n",
       "      <td>52.94</td>\n",
       "      <td>40.18</td>\n",
       "      <td>51.86</td>\n",
       "      <td>47.37</td>\n",
       "      <td>51.86</td>\n",
       "      <td>50.80</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SemEval 14 Restaurant</th>\n",
       "      <td>37.97</td>\n",
       "      <td>48.60</td>\n",
       "      <td>51.46</td>\n",
       "      <td>48.45</td>\n",
       "      <td>48.63</td>\n",
       "      <td>58.01</td>\n",
       "      <td>53.61</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>YouTuBean</th>\n",
       "      <td>20.51</td>\n",
       "      <td>20.51</td>\n",
       "      <td>20.51</td>\n",
       "      <td>31.52</td>\n",
       "      <td>31.11</td>\n",
       "      <td>42.22</td>\n",
       "      <td>42.22</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mean</th>\n",
       "      <td>32.86</td>\n",
       "      <td>39.64</td>\n",
       "      <td>38.53</td>\n",
       "      <td>41.07</td>\n",
       "      <td>40.99</td>\n",
       "      <td>45.35</td>\n",
       "      <td>45.34</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                        LSTM  TDLSTM  TCLSTM  Target-Dep  Target-Dep+  \\\n",
       "Election Twitter       34.12   39.89   41.29       41.91        43.03   \n",
       "Mitchel                38.00   36.24   39.21       31.60        34.83   \n",
       "SemEval 14 Laptop      33.67   52.94   40.18       51.86        47.37   \n",
       "SemEval 14 Restaurant  37.97   48.60   51.46       48.45        48.63   \n",
       "YouTuBean              20.51   20.51   20.51       31.52        31.11   \n",
       "Mean                   32.86   39.64   38.53       41.07        40.99   \n",
       "\n",
       "                       TDParse  TDParse+  \n",
       "Election Twitter         44.00     42.18  \n",
       "Mitchel                  30.65     37.92  \n",
       "SemEval 14 Laptop        51.86     50.80  \n",
       "SemEval 14 Restaurant    58.01     53.61  \n",
       "YouTuBean                42.22     42.22  \n",
       "Mean                     45.35     45.34  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "two_dist_sent = evaluation.evaluate_models(distinct_sentiment_metrics, dataset_test, \n",
    "                                           all_results, custom_metric=True,\n",
    "                                           dataframe=True, distinct_sentiments=2, \n",
    "                                           metric_func=f1_score, average='macro')\n",
    "two_dist_sent = two_dist_sent.drop('Dong Twitter')\n",
    "two_dist_sent = two_dist_sent.drop('Mean')\n",
    "two_dist_mean = pd.Series(two_dist_sent.mean(), name='Mean')\n",
    "(two_dist_sent.append(two_dist_mean) * 100).round(2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Macro F1, 3 Distinct Sentiment subset\n",
    "\n",
    "Lastly the same as above but for 3 distinct sentiments. As the majority of the datasets have very few 3 distinct sentiments per sentence we only look at the Twitter Elections datases (8.78% of the dataset contains 3 distinct sentiments)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/andrew/Envs/Bella/lib/python3.6/site-packages/numpy/core/_methods.py:80: RuntimeWarning: invalid value encountered in double_scalars\n",
      "  ret = ret.dtype.type(ret / rcount)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "TDParse        44.89\n",
       "TDParse+       42.90\n",
       "TDLSTM         42.75\n",
       "TCLSTM         42.36\n",
       "Target-Dep     41.88\n",
       "Target-Dep+    40.88\n",
       "LSTM           29.55\n",
       "Name: Election Twitter, dtype: float64"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "three_dist_sent = evaluation.evaluate_models(distinct_sentiment_metrics, dataset_test, \n",
    "                                             all_results, custom_metric=True,\n",
    "                                             dataframe=True, distinct_sentiments=3, \n",
    "                                             metric_func=f1_score, average='macro')\n",
    "(three_dist_sent.loc['Election Twitter'].sort_values(ascending=False) * 100).round(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}