notebooks/MLPipelinePreparation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# NLP & ML Pipeline Preparation\n",
    "\n",
    "In this notebook I will develop the code needed to create an NLP and ML pipeline for the project.\n",
    "\n",
    "### 1. Import libraries and load data from database.\n",
    "\n",
    "- Import Python libraries\n",
    "- Load dataset from database \n",
    "- Define feature and target variables X and Y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package punkt to /Users/pranath/nltk_data...\n",
      "[nltk_data]   Package punkt is already up-to-date!\n",
      "[nltk_data] Downloading package wordnet to /Users/pranath/nltk_data...\n",
      "[nltk_data]   Package wordnet is already up-to-date!\n",
      "[nltk_data] Downloading package averaged_perceptron_tagger to\n",
      "[nltk_data]     /Users/pranath/nltk_data...\n",
      "[nltk_data]   Package averaged_perceptron_tagger is already up-to-\n",
      "[nltk_data]       date!\n",
      "[nltk_data] Downloading package stopwords to\n",
      "[nltk_data]     /Users/pranath/nltk_data...\n",
      "[nltk_data]   Package stopwords is already up-to-date!\n"
     ]
    }
   ],
   "source": [
    "# import libraries\n",
    "import nltk\n",
    "nltk.download(['punkt', 'wordnet', 'averaged_perceptron_tagger'])\n",
    "nltk.download('stopwords')\n",
    "import re\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from nltk.tokenize import word_tokenize\n",
    "from nltk.stem import WordNetLemmatizer\n",
    "from nltk.corpus import stopwords\n",
    "from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score\n",
    "from sklearn.metrics import classification_report\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "from sklearn.ensemble import RandomForestClassifier, AdaBoostClassifier, GradientBoostingClassifier\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.svm import SVC\n",
    "from sklearn.naive_bayes import MultinomialNB\n",
    "from sklearn.multioutput import MultiOutputClassifier\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.pipeline import Pipeline, FeatureUnion\n",
    "from sklearn.base import BaseEstimator, TransformerMixin\n",
    "from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer\n",
    "from sklearn.utils import resample\n",
    "import sqlite3\n",
    "import sqlalchemy as db\n",
    "import pickle\n",
    "from sklearn.externals import joblib\n",
    "import gensim\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from collections import defaultdict\n",
    "import gensim.models as g\n",
    "import codecs\n",
    "import os\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load data from database\n",
    "engine = create_engine('sqlite:///DisasterResponse.db')\n",
    "df = pd.read_sql_table('DisasterResponse', engine)\n",
    "target_names = ['related', 'request', 'offer', 'aid_related', 'medical_help', 'medical_products', 'search_and_rescue', 'security', 'military', 'water', 'food', 'shelter', 'clothing', 'money', 'missing_people', 'refugees', 'death', 'other_aid', 'infrastructure_related', 'transport', 'buildings', 'electricity', 'tools', 'hospitals', 'shops', 'aid_centers', 'other_infrastructure', 'weather_related', 'floods', 'storm', 'fire', 'earthquake', 'cold', 'other_weather', 'direct_report']\n",
    "X = df['message']\n",
    "Y = df[target_names]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Weather update - a cold front from Cuba that could pass over Haiti\n",
      "\n",
      "Is the Hurricane over or is it not over\n",
      "\n",
      "Looking for someone but no name\n",
      "\n",
      "UN reports Leogane 80-90 destroyed. Only Hospital St. Croix functioning. Needs supplies desperately.\n",
      "\n",
      "says: west side of Haiti, rest of the country today and tonight\n",
      "\n",
      "Information about the National Palace-\n",
      "\n",
      "Storm at sacred heart of jesus\n",
      "\n",
      "Please, we need tents and water. We are in Silo, Thank you!\n",
      "\n",
      "I would like to receive the messages, thank you\n",
      "\n",
      "I am in Croix-des-Bouquets. We have health issues. They ( workers ) are in Santo 15. ( an area in Croix-des-Bouquets )\n",
      "\n",
      "There's nothing to eat and water, we starving and thirsty.\n",
      "\n",
      "I am in Petionville. I need more information regarding 4636\n",
      "\n",
      "I am in Thomassin number 32, in the area named Pyron. I would like to have some water. Thank God we are fine, but we desperately need water. Thanks\n",
      "\n",
      "Let's do it together, need food in Delma 75, in didine area\n",
      "\n",
      "More information on the 4636 number in order for me to participate. ( To see if I can use it )\n",
      "\n",
      "A Comitee in Delmas 19, Rue ( street ) Janvier, Impasse Charite #2. We have about 500 people in a temporary shelter and we are in dire need of Water, Food, Medications, Tents and Clothes. Please stop by and see us.\n",
      "\n",
      "We need food and water in Klecin 12. We are dying of hunger. Impasse Chretien Klecin 12 extended ( extension ) We are hungry and sick.\n",
      "\n",
      "are you going to call me or do you want me to call ou? let me know?\n",
      "\n",
      "I don't understand how to use this thing 4636.\n",
      "\n",
      "I would like to know if the earthquake is over. Thanks\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Show 20 example messages\n",
    "pd.options.display.max_colwidth = 110\n",
    "for i in range(20):\n",
    "    print(X.iloc[i])\n",
    "    print('')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2. Write a tokenization function to process text data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tokenize(text):\n",
    "    # Remove punctuation characters\n",
    "    text = re.sub(r\"[^a-zA-Z0-9]\", \" \", text) \n",
    "    # Tokenise\n",
    "    tokens = word_tokenize(text)\n",
    "    # Create lemmatiser\n",
    "    lemmatizer = WordNetLemmatizer()\n",
    "\n",
    "    clean_tokens = []\n",
    "    # Lemmatize, standardise (lower case) & clean tokens\n",
    "    for tok in tokens:\n",
    "        clean_tok = lemmatizer.lemmatize(tok).lower().strip()\n",
    "        clean_tokens.append(clean_tok)\n",
    "\n",
    "    # Load stop words\n",
    "    stop_words = stopwords.words('english')\n",
    "    # Remove stop words\n",
    "    clean_tokens = [word for word in clean_tokens if word not in stop_words]\n",
    "        \n",
    "    return clean_tokens"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3. Build a machine learning pipeline\n",
    "\n",
    "The machine pipeline will take in the `message` column as input and output classification results on the other 36 categories in the dataset. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define ML pipeline\n",
    "pipeline = Pipeline([\n",
    "        ('vect', CountVectorizer(tokenizer=tokenize)),\n",
    "        ('tfidf', TfidfTransformer()),\n",
    "        ('clf', MultiOutputClassifier(SVC()))\n",
    "    ])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 4. Train pipeline\n",
    "- Split data into train and test sets\n",
    "- Train pipeline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Split data into train and test\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, Y, random_state=49)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Train pipeline\n",
    "pipeline.fit(X_train, y_train)\n",
    "\n",
    "# predict on test data\n",
    "y_pred = pipeline.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred = pd.DataFrame(y_pred)\n",
    "y_pred.columns = target_names"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 5. Test the model\n",
    "\n",
    "I will report the f1 score, precision and recall for each output category of the dataset, however I will choose the f1 score to be the key score in determining the best performing model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def performance_report(actual, predicted, target_names):\n",
    "    \n",
    "    metrics = []\n",
    "    \n",
    "    # For each of the target features\n",
    "    for i in target_names:\n",
    "        # Calculate metrics\n",
    "        accuracy = accuracy_score(actual[i], predicted[i])\n",
    "        precision = precision_score(actual[i], predicted[i])\n",
    "        recall = recall_score(actual[i], predicted[i])\n",
    "        f1 = f1_score(actual[i], predicted[i])\n",
    "        # Save metrics\n",
    "        metrics.append([accuracy, precision, recall, f1])\n",
    "    \n",
    "    # Convert to numpy array\n",
    "    metrics = np.array(metrics)\n",
    "    # Convert to dataframe\n",
    "    metrics_df = pd.DataFrame(data = metrics, index = target_names, columns = ['Accuracy', 'Precision', 'Recall', 'F1'])\n",
    "    # Add mean\n",
    "    mean_acc = metrics_df['Accuracy'].mean()\n",
    "    mean_pre = metrics_df['Precision'].mean()\n",
    "    mean_rec = metrics_df['Recall'].mean()\n",
    "    mean_f1 = metrics_df['F1'].mean()\n",
    "    row = pd.Series({'Accuracy':mean_acc,'Precision':mean_pre,'Recall':mean_rec,'F1':mean_f1},name='MEAN SCORE')\n",
    "    metrics_df = metrics_df.append(row)\n",
    "    return metrics_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                        Accuracy  Precision    Recall        F1\n",
      "related                 0.763656   0.763656  1.000000  0.865992\n",
      "request                 0.832926   0.000000  0.000000  0.000000\n",
      "offer                   0.996338   0.000000  0.000000  0.000000\n",
      "aid_related             0.587428   0.000000  0.000000  0.000000\n",
      "medical_help            0.922490   0.000000  0.000000  0.000000\n",
      "medical_products        0.951938   0.000000  0.000000  0.000000\n",
      "search_and_rescue       0.973299   0.000000  0.000000  0.000000\n",
      "security                0.983369   0.000000  0.000000  0.000000\n",
      "military                0.967043   0.000000  0.000000  0.000000\n",
      "water                   0.937748   0.000000  0.000000  0.000000\n",
      "food                    0.891517   0.000000  0.000000  0.000000\n",
      "shelter                 0.911504   0.000000  0.000000  0.000000\n",
      "clothing                0.987946   0.000000  0.000000  0.000000\n",
      "money                   0.977113   0.000000  0.000000  0.000000\n",
      "missing_people          0.989472   0.000000  0.000000  0.000000\n",
      "refugees                0.965060   0.000000  0.000000  0.000000\n",
      "death                   0.953616   0.000000  0.000000  0.000000\n",
      "other_aid               0.868020   0.000000  0.000000  0.000000\n",
      "infrastructure_related  0.934696   0.000000  0.000000  0.000000\n",
      "transport               0.953311   0.000000  0.000000  0.000000\n",
      "buildings               0.950107   0.000000  0.000000  0.000000\n",
      "electricity             0.980012   0.000000  0.000000  0.000000\n",
      "tools                   0.996033   0.000000  0.000000  0.000000\n",
      "hospitals               0.989777   0.000000  0.000000  0.000000\n",
      "shops                   0.993897   0.000000  0.000000  0.000000\n",
      "aid_centers             0.988404   0.000000  0.000000  0.000000\n",
      "other_infrastructure    0.957125   0.000000  0.000000  0.000000\n",
      "weather_related         0.724596   0.000000  0.000000  0.000000\n",
      "floods                  0.916082   0.000000  0.000000  0.000000\n",
      "storm                   0.908300   0.000000  0.000000  0.000000\n",
      "fire                    0.990388   0.000000  0.000000  0.000000\n",
      "earthquake              0.913183   0.000000  0.000000  0.000000\n",
      "cold                    0.977418   0.000000  0.000000  0.000000\n",
      "other_weather           0.947818   0.000000  0.000000  0.000000\n",
      "direct_report           0.809124   0.000000  0.000000  0.000000\n",
      "MEAN SCORE              0.925450   0.021819  0.028571  0.024743\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.6/site-packages/sklearn/metrics/classification.py:1135: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n",
      "/opt/conda/lib/python3.6/site-packages/sklearn/metrics/classification.py:1135: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n"
     ]
    }
   ],
   "source": [
    "# Show metrics for each category\n",
    "print(performance_report(y_test, y_pred, target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So we can see the metric values vary considerably, this is probably due to the low number of examples for particular categories (as we can see from the warnings, some have no examples to test). In particular while accuracy is high, recall, precison and f1 scores are zero with this model."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 6. Improve the model\n",
    "\n",
    "Use grid search to find better parameters. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "#parameters = {\n",
    "        #'vect__ngram_range': ((1, 2)),\n",
    "        #'vect__max_df': (0.5, 0.75, 1.0),\n",
    "        #'vect__max_features': (None, 5000, 10000),\n",
    "        #'tfidf__use_idf': (True, False),\n",
    "        #'clf__estimator__n_estimators':[15, 50], \n",
    "        #'clf__estimator__min_samples_split':[2, 5, 10]\n",
    "        #'vect__max_df': (0.5, 0.75),\n",
    "#}\n",
    "\n",
    "#cv = GridSearchCV(pipeline, param_grid=parameters)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 7. Test the model again\n",
    "\n",
    "Show the accuracy, precision, and recall of the tuned model.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Train pipeline with grid search\n",
    "#cv.fit(X_train, y_train)\n",
    "\n",
    "# predict on test data\n",
    "#y_pred2 = cv.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "#y_pred2 = pd.DataFrame(y_pred2)\n",
    "#y_pred2.columns = target_names\n",
    "\n",
    "# Evaluate the model\n",
    "#evaluate_model(y_test, y_pred2, target_names)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 8. Trying to improve the model further\n",
    "\n",
    "Reading on various techniques in several articles including these:\n",
    "\n",
    "https://towardsdatascience.com/machine-learning-multiclass-classification-with-imbalanced-data-set-29f6a177c1a\n",
    "\n",
    "https://towardsdatascience.com/methods-for-dealing-with-imbalanced-data-5b761be45a18\n",
    "\n",
    "https://machinelearningmastery.com/tactics-to-combat-imbalanced-classes-in-your-machine-learning-dataset/\n",
    "\n",
    "I decided to try 3 approaches to try to improve results.\n",
    "\n",
    "- Oversampling minority categories in the data\n",
    "- Change model to Gradient Boosting \n",
    "- Add starting verb feature"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 8.1 Try a different model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/metrics/classification.py:1143: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/metrics/classification.py:1143: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                        Accuracy  Precision    Recall        F1\n",
      "related                 0.821941   0.837496  0.951449  0.890843\n",
      "request                 0.901282   0.815493  0.528767  0.641551\n",
      "offer                   0.996338   0.000000  0.000000  0.000000\n",
      "aid_related             0.781660   0.774472  0.664201  0.715110\n",
      "medical_help            0.928898   0.664062  0.167323  0.267296\n",
      "medical_products        0.957125   0.702381  0.187302  0.295739\n",
      "search_and_rescue       0.974672   0.909091  0.057143  0.107527\n",
      "security                0.983369   0.000000  0.000000  0.000000\n",
      "military                0.968874   0.676471  0.106481  0.184000\n",
      "water                   0.959872   0.805907  0.468137  0.592248\n",
      "food                    0.942630   0.849687  0.572433  0.684034\n",
      "shelter                 0.941868   0.813880  0.444828  0.575251\n",
      "clothing                0.989930   0.882353  0.189873  0.312500\n",
      "money                   0.978029   0.666667  0.080000  0.142857\n",
      "missing_people          0.989625   1.000000  0.014493  0.028571\n",
      "refugees                0.965975   0.625000  0.065502  0.118577\n",
      "death                   0.962313   0.890411  0.213816  0.344828\n",
      "other_aid               0.875191   0.645963  0.120231  0.202729\n",
      "infrastructure_related  0.935612   0.625000  0.035047  0.066372\n",
      "transport               0.957736   0.853659  0.114379  0.201729\n",
      "buildings               0.956668   0.759036  0.192661  0.307317\n",
      "electricity             0.982606   0.793103  0.175573  0.287500\n",
      "tools                   0.996033   0.000000  0.000000  0.000000\n",
      "hospitals               0.989625   0.000000  0.000000  0.000000\n",
      "shops                   0.993897   0.000000  0.000000  0.000000\n",
      "aid_centers             0.988404   0.000000  0.000000  0.000000\n",
      "other_infrastructure    0.957278   0.545455  0.021352  0.041096\n",
      "weather_related         0.874123   0.860294  0.648199  0.739336\n",
      "floods                  0.947666   0.915663  0.414545  0.570713\n",
      "storm                   0.934391   0.750733  0.425957  0.543524\n",
      "fire                    0.990388   0.500000  0.015873  0.030769\n",
      "earthquake              0.963076   0.888361  0.657293  0.755556\n",
      "cold                    0.980165   0.875000  0.141892  0.244186\n",
      "other_weather           0.949649   0.833333  0.043860  0.083333\n",
      "direct_report           0.868477   0.775887  0.437250  0.559305\n",
      "MEAN SCORE              0.948154   0.643853  0.233025  0.300983\n"
     ]
    }
   ],
   "source": [
    "# Logistic regression\n",
    "# Define ML pipeline\n",
    "pipeline1 = Pipeline([\n",
    "        ('vect', CountVectorizer(tokenizer=tokenize)),\n",
    "        ('tfidf', TfidfTransformer()),\n",
    "        ('clf', MultiOutputClassifier(LogisticRegression(random_state=1)))\n",
    "    ])\n",
    "\n",
    "# Train pipeline\n",
    "pipeline1.fit(X_train, y_train)\n",
    "\n",
    "# predict on test data\n",
    "y_pred1 = pipeline1.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred1 = pd.DataFrame(y_pred1)\n",
    "y_pred1.columns = target_names\n",
    "\n",
    "# Evaluate the model\n",
    "print(performance_report(y_test, y_pred1, target_names))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                        Accuracy  Precision    Recall        F1\n",
      "related                 0.789594   0.794223  0.977822  0.876511\n",
      "request                 0.900214   0.847244  0.491324  0.621965\n",
      "offer                   0.993897   0.000000  0.000000  0.000000\n",
      "aid_related             0.749466   0.780359  0.546598  0.642888\n",
      "medical_help            0.927983   0.609756  0.196850  0.297619\n",
      "medical_products        0.960940   0.700680  0.326984  0.445887\n",
      "search_and_rescue       0.969942   0.387755  0.217143  0.278388\n",
      "security                0.975282   0.079365  0.045872  0.058140\n",
      "military                0.970400   0.627907  0.250000  0.357616\n",
      "water                   0.962313   0.720548  0.644608  0.680466\n",
      "food                    0.952090   0.789781  0.760900  0.775072\n",
      "shelter                 0.949344   0.805419  0.563793  0.663286\n",
      "clothing                0.987946   0.500000  0.569620  0.532544\n",
      "money                   0.974977   0.400000  0.186667  0.254545\n",
      "missing_people          0.985810   0.239130  0.159420  0.191304\n",
      "refugees                0.962923   0.448529  0.266376  0.334247\n",
      "death                   0.966738   0.706731  0.483553  0.574219\n",
      "other_aid               0.873818   0.661017  0.090173  0.158698\n",
      "infrastructure_related  0.936680   0.685714  0.056075  0.103672\n",
      "transport               0.956363   0.596154  0.202614  0.302439\n",
      "buildings               0.959719   0.748031  0.290520  0.418502\n",
      "electricity             0.980012   0.500000  0.274809  0.354680\n",
      "tools                   0.990388   0.000000  0.000000  0.000000\n",
      "hospitals               0.986115   0.200000  0.119403  0.149533\n",
      "shops                   0.990235   0.000000  0.000000  0.000000\n",
      "aid_centers             0.983979   0.177778  0.105263  0.132231\n",
      "other_infrastructure    0.954684   0.277778  0.035587  0.063091\n",
      "weather_related         0.872597   0.875387  0.626593  0.730384\n",
      "floods                  0.958041   0.889518  0.570909  0.695460\n",
      "storm                   0.944309   0.735060  0.613977  0.669084\n",
      "fire                    0.987183   0.333333  0.333333  0.333333\n",
      "earthquake              0.971315   0.835979  0.833040  0.834507\n",
      "cold                    0.980317   0.587156  0.432432  0.498054\n",
      "other_weather           0.948886   0.544304  0.125731  0.204276\n",
      "direct_report           0.862374   0.799314  0.372502  0.508179\n",
      "MEAN SCORE              0.946196   0.539541  0.336300  0.392595\n"
     ]
    }
   ],
   "source": [
    "# Gradient boosting\n",
    "# Define ML pipeline\n",
    "pipeline2 = Pipeline([\n",
    "        ('vect', CountVectorizer(tokenizer=tokenize)),\n",
    "        ('tfidf', TfidfTransformer()),\n",
    "        ('clf', MultiOutputClassifier(GradientBoostingClassifier(random_state=1)))\n",
    "    ])\n",
    "\n",
    "# Train pipeline\n",
    "pipeline2.fit(X_train, y_train)\n",
    "\n",
    "# predict on test data\n",
    "y_pred2 = pipeline2.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred2 = pd.DataFrame(y_pred2)\n",
    "y_pred2.columns = target_names\n",
    "\n",
    "# Evaluate the model\n",
    "print(performance_report(y_test, y_pred2, target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So the gradient boosting model seems to perform the best of all models."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 8.2 Oversampling minority categories in the data\n",
    "\n",
    "We saw in the ETL phase how there was a high degree of variance on examples for each of the classes, with some classes/categories having few as 120 examples and others as high as 20,000 examples - so this is a very unbalanced dataset. Two common methods used to address this are oversampling and undersampling. Given our total dataset is around 26,000 rows, I will look at the oversampling technique to generate more samples from the minority category."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "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>0</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>shops</th>\n",
       "      <td>80</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>offer</th>\n",
       "      <td>94</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>tools</th>\n",
       "      <td>133</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>hospitals</th>\n",
       "      <td>216</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>fire</th>\n",
       "      <td>219</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing_people</th>\n",
       "      <td>229</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>aid_centers</th>\n",
       "      <td>233</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>clothing</th>\n",
       "      <td>326</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>security</th>\n",
       "      <td>362</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>cold</th>\n",
       "      <td>382</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                  0\n",
       "shops            80\n",
       "offer            94\n",
       "tools           133\n",
       "hospitals       216\n",
       "fire            219\n",
       "missing_people  229\n",
       "aid_centers     233\n",
       "clothing        326\n",
       "security        362\n",
       "cold            382"
      ]
     },
     "execution_count": 101,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Get the sum of all occurances of all categories\n",
    "categories_total_train = y_train.sum(axis = 0)\n",
    "categories_total_train_df = pd.DataFrame(categories_total_train)\n",
    "categories_total_train_df = categories_total_train_df.sort_values(by=[0])\n",
    "# List the 10 lowest categories and their total occurances\n",
    "categories_total_train_df.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "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>0</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>offer</th>\n",
       "      <td>24</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>tools</th>\n",
       "      <td>26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>shops</th>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>fire</th>\n",
       "      <td>63</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>hospitals</th>\n",
       "      <td>67</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing_people</th>\n",
       "      <td>69</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>aid_centers</th>\n",
       "      <td>76</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>clothing</th>\n",
       "      <td>79</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>security</th>\n",
       "      <td>109</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>electricity</th>\n",
       "      <td>131</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                  0\n",
       "offer            24\n",
       "tools            26\n",
       "shops            40\n",
       "fire             63\n",
       "hospitals        67\n",
       "missing_people   69\n",
       "aid_centers      76\n",
       "clothing         79\n",
       "security        109\n",
       "electricity     131"
      ]
     },
     "execution_count": 102,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Get the sum of all occurances of all categories\n",
    "categories_total_test = y_test.sum(axis = 0)\n",
    "categories_total_test_df = pd.DataFrame(categories_total_test)\n",
    "categories_total_test_df = categories_total_test_df.sort_values(by=[0])\n",
    "# List the 10 lowest categories and their total occurances\n",
    "categories_total_test_df.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "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>0</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>shelter</th>\n",
       "      <td>1734</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>storm</th>\n",
       "      <td>1842</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>earthquake</th>\n",
       "      <td>1886</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>food</th>\n",
       "      <td>2212</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>other_aid</th>\n",
       "      <td>2581</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>request</th>\n",
       "      <td>3379</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>direct_report</th>\n",
       "      <td>3824</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>weather_related</th>\n",
       "      <td>5492</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>aid_related</th>\n",
       "      <td>8156</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>related</th>\n",
       "      <td>15088</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     0\n",
       "shelter           1734\n",
       "storm             1842\n",
       "earthquake        1886\n",
       "food              2212\n",
       "other_aid         2581\n",
       "request           3379\n",
       "direct_report     3824\n",
       "weather_related   5492\n",
       "aid_related       8156\n",
       "related          15088"
      ]
     },
     "execution_count": 89,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# List the 10 highest categories and their total occurances\n",
    "categories_total_train_df.tail(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Create function to show count totals of categories as a bar chart\n",
    "def plot_target_feature_counts(features_df):\n",
    "    \"\"\"\"\"\"\n",
    "    features_totals = features_df.apply(pd.Series.value_counts)\n",
    "    ax = features_totals.iloc[1].plot.bar(figsize=(15,10))\n",
    "    ax.set_xlabel(\"Category name\")\n",
    "    ax.set_ylabel(\"Category count\")\n",
    "    ax.set_title(\"Message category usage counts\")\n",
    "\n",
    "plot_target_feature_counts(y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Rebalance training data to better represent lowest occuring categories\n",
    "def balance_training_data(training_features, training_targets):\n",
    "    \"\"\"\"\"\"\n",
    "    \n",
    "    # Get features & target names\n",
    "    training_features = pd.Series.to_frame(training_features)\n",
    "    feature_names = training_features.columns.values\n",
    "    target_names = training_targets.columns.values\n",
    "    # Get counts of training targets\n",
    "    targets_totals = training_targets.sum(axis = 0)\n",
    "    # Get mean of training targets\n",
    "    mean_targets_totals = int(targets_totals.mean())\n",
    "    # Merge features & targets\n",
    "    training_df = training_features.join(training_targets)\n",
    "    # Convert totals to df\n",
    "    targets_totals = pd.Series.to_frame(targets_totals)\n",
    "    # Get empty class rows\n",
    "    empty_rows = training_df.loc[(training_df['related'] == 0)]\n",
    "    # Get minority class rows \n",
    "    minority_rows = training_df.loc[(training_df['shops'] == 1) | (training_df['offer'] == 1) | (training_df['tools'] == 1) | (training_df['hospitals'] == 1) | (training_df['fire'] == 1) | (training_df['missing_people'] == 1) | (training_df['aid_centers'] == 1) | (training_df['clothing'] == 1) | (training_df['security'] == 1) | (training_df['cold'] == 1)]\n",
    "    minority_rows_balanced = resample(minority_rows, n_samples=training_features.shape[0], random_state=0, replace=True)\n",
    "    # Merge\n",
    "    balanced_df = pd.concat([empty_rows, minority_rows_balanced], ignore_index=True)\n",
    "    # Resplit features from targets\n",
    "    training_features_balanced = balanced_df[feature_names]\n",
    "    training_features_balanced = training_features_balanced.T.squeeze()\n",
    "    training_targets_balanced = balanced_df[target_names]\n",
    "    # Return balanced features & targets\n",
    "    return training_features_balanced, training_targets_balanced\n",
    "\n",
    "    \n",
    "X_train_bal, y_train_bal = balance_training_data(X_train, y_train)\n",
    "plot_target_feature_counts(y_train_bal)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see how the lowest occuring categories have improved a bit in proportion to the rest, but not by much. Lets try training a model with the more balanced data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                        Accuracy  Precision    Recall        F1\n",
      "related                 0.795392   0.801812  0.972428  0.878916\n",
      "request                 0.903570   0.838946  0.523288  0.644544\n",
      "offer                   0.992218   0.000000  0.000000  0.000000\n",
      "aid_related             0.758773   0.804997  0.548077  0.652145\n",
      "medical_help            0.928746   0.625767  0.200787  0.304024\n",
      "medical_products        0.961245   0.704698  0.333333  0.452586\n",
      "search_and_rescue       0.970095   0.387097  0.205714  0.268657\n",
      "security                0.975435   0.080645  0.045872  0.058480\n",
      "military                0.968874   0.571429  0.222222  0.320000\n",
      "water                   0.963381   0.738636  0.637255  0.684211\n",
      "food                    0.955294   0.800287  0.783404  0.791756\n",
      "shelter                 0.947971   0.799499  0.550000  0.651685\n",
      "clothing                0.987641   0.489130  0.569620  0.526316\n",
      "money                   0.975282   0.409091  0.180000  0.250000\n",
      "missing_people          0.985352   0.245283  0.188406  0.213115\n",
      "refugees                0.963534   0.460938  0.257642  0.330532\n",
      "death                   0.966738   0.704762  0.486842  0.575875\n",
      "other_aid               0.873360   0.647059  0.089017  0.156504\n",
      "infrastructure_related  0.935917   0.633333  0.044393  0.082969\n",
      "transport               0.954532   0.532258  0.215686  0.306977\n",
      "buildings               0.957888   0.691729  0.281346  0.400000\n",
      "electricity             0.979249   0.464789  0.251908  0.326733\n",
      "tools                   0.992218   0.000000  0.000000  0.000000\n",
      "hospitals               0.984590   0.113636  0.074627  0.090090\n",
      "shops                   0.990235   0.038462  0.025000  0.030303\n",
      "aid_centers             0.983369   0.133333  0.078947  0.099174\n",
      "other_infrastructure    0.955752   0.304348  0.024911  0.046053\n",
      "weather_related         0.871071   0.872671  0.622715  0.726802\n",
      "floods                  0.957125   0.876751  0.569091  0.690187\n",
      "storm                   0.943699   0.732000  0.608985  0.664850\n",
      "fire                    0.987489   0.369863  0.428571  0.397059\n",
      "earthquake              0.970095   0.830088  0.824253  0.827160\n",
      "cold                    0.979860   0.576923  0.405405  0.476190\n",
      "other_weather           0.949039   0.543478  0.146199  0.230415\n",
      "direct_report           0.872597   0.801449  0.442046  0.569809\n",
      "MEAN SCORE              0.946789   0.532148  0.338228  0.392118\n"
     ]
    }
   ],
   "source": [
    "# Train pipeline with balanced data\n",
    "pipeline2.fit(X_train_bal, y_train_bal)\n",
    "\n",
    "# predict on test data\n",
    "y_pred3 = pipeline2.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred3 = pd.DataFrame(y_pred2)\n",
    "y_pred3.columns = target_names\n",
    "\n",
    "# Evaluate the model\n",
    "print(performance_report(y_test, y_pred3, target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "But still not much improvement. Lets have a look at the classes and messages associated with samples in the lowest occuring classes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "40        People from Dal blocked since Wednesday in Carrefour, we having water shortage, food and medical assistance.\n",
       "18044    A double suicide bombing at a busy marketplace in Borno State's largest city, Maiduguri, on 25 November, w...\n",
       "12476    calm before the storm heading home after making election calls ready to hunker down hey we are all in this...\n",
       "21073    The Government has started to demarcate and allocate plots in some areas, but over 230,000 people are stil...\n",
       "10220    RT gsdog7 RT Huge earthquake hits Haiti Links to specific relief efforts http tinyurl.com yg2hjhu via sami...\n",
       "14681    Since then, excessive rains have mitigated the anticipated improvements in food security, especially in th...\n",
       "5926                                           Ruelle Canne-a-Sucre Prolongue (Carrefour) We need tents, water, food. \n",
       "22837    We want to reach a stage where robust yam seed production techniques will use parts of the yam plant other...\n",
       "277                                             This did not just happen in Port au Prince it's up in Pot de Paix too.\n",
       "14764    JAKARTA, Jan 24 (AFP) - A strong earthquake shook Indonesia's Sulawesi island Monday, damaging dozens of s...\n",
       "Name: message, dtype: object"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Merge features & targets\n",
    "training_features = pd.Series.to_frame(X_train)\n",
    "target_names = y_train.columns.values\n",
    "training_df = training_features.join(y_train)\n",
    "# Get shop samples\n",
    "shops_df = training_df.loc[(training_df['shops'] == 1)]\n",
    "random_10_shops = shops_df.sample(n=10)\n",
    "# Show 10 random messages for shops category\n",
    "random_10_shops['message']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Seems to be difficult to understand from these what these have to do with shops! Lets see the distrbution of other categories in the shop samples."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot category counts of all shop category messages\n",
    "shops_target_classes_df = shops_df[target_names]\n",
    "plot_target_feature_counts(shops_target_classes_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So we can see the samples with all shop categories is still very unbalanced. Lets look at the next lowest category - 'offer'."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10788                               I can cook and would be happy to try to provide meals or serve in a soup kitchen .\n",
       "24448                    The medical help involves the setting up of U.S. military mobile surgical hospitals, he said.\n",
       "11257    I have canned goods , some jars of baby food ( stage 3 ) , children 's books , children 's puzzles , cards...\n",
       "15353    Initial assessment done have not yet identified the source of the outbreak, but the national task force ha...\n",
       "11312                                              When and Where should hot meals be delivered this evening ? 11/3/12\n",
       "10804                                                            I can help packing and distributing food and supplies\n",
       "20304    They receive free transport from Iran to Herat, where they receive a $10 travel grant, a tarpaulin, blanke...\n",
       "11221    Looking to donate new ( several-year shelf life , sealed ) non-prescription medications to any organizatio...\n",
       "11106                                                                   prepare food package , help to deliver food ..\n",
       "24231    In addition to the above, 300 navy boats with 29 teams from the Navy and Airforce are being kept ready to ...\n",
       "Name: message, dtype: object"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Get shop samples\n",
    "offer_df = training_df.loc[(training_df['offer'] == 1)]\n",
    "random_10_offer = offer_df.sample(n=10)\n",
    "# Show 10 random messages for offer category\n",
    "random_10_offer['message']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Can you tell what might distinguish what an 'offer' is? lets see the category counts for all offer samples."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot category counts of all offer category messages\n",
    "offer_target_classes_df = offer_df[target_names]\n",
    "plot_target_feature_counts(offer_target_classes_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So even if we only use samples from minority classes, we can see how we are still building up an imbalanced dataset that is difficult to train on, let alone the fact that it seems very challenging for any ML model to learn what the categories have in common from the varied and often unclear message data!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 8.3 Create new feature for starting verb with new model\n",
    "\n",
    "In NLP people use a wide range of features to train text classifiers. Some use parts of speech features (such as nouns, verbs etc) as features. With this in mind we will add a new feature that detects any starting verbs.\n",
    "\n",
    "https://www.nltk.org/book/ch06.html\n",
    "\n",
    "https://www.analyticsvidhya.com/blog/2018/04/a-comprehensive-guide-to-understand-and-implement-text-classification-in-python/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "class StartingVerbExtractor(BaseEstimator, TransformerMixin):\n",
    "\n",
    "    def starting_verb(self, text):\n",
    "        sentence_list = nltk.sent_tokenize(text)\n",
    "        for sentence in sentence_list:\n",
    "            pos_tags = nltk.pos_tag(tokenize(sentence))\n",
    "            if pos_tags:\n",
    "                first_word, first_tag = pos_tags[0]\n",
    "                #if first_tag in ['VB', 'VBP'] or first_word == 'RT':\n",
    "                #if first_tag in ['NN', 'NNS', 'NNP', 'NNPS']:\n",
    "                #    return True\n",
    "                return False\n",
    "        return False\n",
    "\n",
    "    def fit(self, X, y=None):\n",
    "        return self\n",
    "\n",
    "    def transform(self, X):\n",
    "        X_tagged = pd.Series(X).apply(self.starting_verb)\n",
    "        return pd.DataFrame(X_tagged)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                        Accuracy  Precision    Recall        F1\n",
      "related                 0.788526   0.793036  0.978422  0.876029\n",
      "request                 0.900214   0.847244  0.491324  0.621965\n",
      "offer                   0.992676   0.038462  0.041667  0.040000\n",
      "aid_related             0.749924   0.781300  0.546967  0.643463\n",
      "medical_help            0.929509   0.645570  0.200787  0.306306\n",
      "medical_products        0.960330   0.692308  0.314286  0.432314\n",
      "search_and_rescue       0.970095   0.398058  0.234286  0.294964\n",
      "security                0.975893   0.084746  0.045872  0.059524\n",
      "military                0.970705   0.646341  0.245370  0.355705\n",
      "water                   0.962771   0.731638  0.634804  0.679790\n",
      "food                    0.953311   0.795620  0.766526  0.780802\n",
      "shelter                 0.948886   0.800983  0.562069  0.660588\n",
      "clothing                0.988099   0.505618  0.569620  0.535714\n",
      "money                   0.975740   0.418182  0.153333  0.224390\n",
      "missing_people          0.985200   0.220000  0.159420  0.184874\n",
      "refugees                0.963534   0.463235  0.275109  0.345205\n",
      "death                   0.967958   0.737374  0.480263  0.581673\n",
      "other_aid               0.873970   0.666667  0.090173  0.158859\n",
      "infrastructure_related  0.936222   0.638889  0.053738  0.099138\n",
      "transport               0.956515   0.601942  0.202614  0.303178\n",
      "buildings               0.959109   0.721805  0.293578  0.417391\n",
      "electricity             0.979402   0.474359  0.282443  0.354067\n",
      "tools                   0.990388   0.000000  0.000000  0.000000\n",
      "hospitals               0.986115   0.184211  0.104478  0.133333\n",
      "shops                   0.990235   0.000000  0.000000  0.000000\n",
      "aid_centers             0.984590   0.179487  0.092105  0.121739\n",
      "other_infrastructure    0.953921   0.216216  0.028470  0.050314\n",
      "weather_related         0.871987   0.875000  0.624377  0.728742\n",
      "floods                  0.957888   0.887006  0.570909  0.694690\n",
      "storm                   0.944156   0.733598  0.613977  0.668478\n",
      "fire                    0.986115   0.294118  0.317460  0.305344\n",
      "earthquake              0.971926   0.839506  0.836555  0.838028\n",
      "cold                    0.980928   0.603604  0.452703  0.517375\n",
      "other_weather           0.949496   0.575342  0.122807  0.202410\n",
      "direct_report           0.862374   0.801382  0.370903  0.507104\n",
      "MEAN SCORE              0.946249   0.539796  0.335926  0.392100\n"
     ]
    }
   ],
   "source": [
    "# Define new pipeline adding starting verb feature\n",
    "pipeline3 = Pipeline([\n",
    "    ('features', FeatureUnion([\n",
    "\n",
    "        ('text_pipeline', Pipeline([\n",
    "            ('vect', CountVectorizer(tokenizer=tokenize)),\n",
    "            ('tfidf', TfidfTransformer())\n",
    "        ])),\n",
    "\n",
    "        ('starting_verb', StartingVerbExtractor())\n",
    "    ])),\n",
    "\n",
    "    ('clf', MultiOutputClassifier(GradientBoostingClassifier(random_state=1)))\n",
    "])\n",
    "\n",
    "# Train pipeline\n",
    "pipeline3.fit(X_train, y_train)\n",
    "\n",
    "# predict on test data\n",
    "y_pred4 = pipeline3.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred4 = pd.DataFrame(y_pred4)\n",
    "y_pred4.columns = target_names\n",
    "\n",
    "# Evaluate the model\n",
    "print(performance_report(y_test, y_pred4, target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The starting verb feature has not really helped improve the model."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 8.4 Combine new model & balanced data & verb feature"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                        Accuracy  Precision    Recall        F1\n",
      "related                 0.791730   0.828876  0.916484  0.870481\n",
      "request                 0.885871   0.811490  0.412785  0.547215\n",
      "offer                   0.990388   0.024390  0.041667  0.030769\n",
      "aid_related             0.554318   0.479624  0.944527  0.636194\n",
      "medical_help            0.925694   0.554974  0.208661  0.303290\n",
      "medical_products        0.957278   0.624113  0.279365  0.385965\n",
      "search_and_rescue       0.974519   0.571429  0.182857  0.277056\n",
      "security                0.972994   0.153061  0.137615  0.144928\n",
      "military                0.969332   0.578947  0.254630  0.353698\n",
      "water                   0.952548   0.587703  0.796569  0.676379\n",
      "food                    0.949191   0.781250  0.738397  0.759219\n",
      "shelter                 0.945377   0.765550  0.551724  0.641283\n",
      "clothing                0.989777   0.568182  0.632911  0.598802\n",
      "money                   0.972078   0.305882  0.173333  0.221277\n",
      "missing_people          0.986268   0.266667  0.173913  0.210526\n",
      "refugees                0.964754   0.485714  0.148472  0.227425\n",
      "death                   0.966585   0.745665  0.424342  0.540881\n",
      "other_aid               0.865578   0.427273  0.054335  0.096410\n",
      "infrastructure_related  0.913030   0.269481  0.193925  0.225543\n",
      "transport               0.954837   0.534247  0.254902  0.345133\n",
      "buildings               0.959109   0.718519  0.296636  0.419913\n",
      "electricity             0.980623   0.521739  0.366412  0.430493\n",
      "tools                   0.992676   0.107143  0.115385  0.111111\n",
      "hospitals               0.983064   0.244186  0.313433  0.274510\n",
      "shops                   0.990235   0.071429  0.050000  0.058824\n",
      "aid_centers             0.974214   0.179310  0.342105  0.235294\n",
      "other_infrastructure    0.955447   0.280000  0.024911  0.045752\n",
      "weather_related         0.838114   0.697243  0.728532  0.712544\n",
      "floods                  0.953921   0.869048  0.530909  0.659142\n",
      "storm                   0.938511   0.723982  0.532446  0.613615\n",
      "fire                    0.988862   0.437500  0.555556  0.489510\n",
      "earthquake              0.966280   0.856557  0.734622  0.790918\n",
      "cold                    0.979402   0.536313  0.648649  0.587156\n",
      "other_weather           0.948123   0.510000  0.149123  0.230769\n",
      "direct_report           0.857644   0.750789  0.380496  0.505040\n",
      "MEAN SCORE              0.936811   0.510522  0.379732  0.407345\n"
     ]
    }
   ],
   "source": [
    "# Train pipeline\n",
    "pipeline3.fit(X_train_bal, y_train_bal)\n",
    "\n",
    "# predict on test data\n",
    "y_pred5 = pipeline3.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred5 = pd.DataFrame(y_pred5)\n",
    "y_pred5.columns = target_names\n",
    "\n",
    "# Evaluate the model\n",
    "print(performance_report(y_test, y_pred5, target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A slight improvement with this combination."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 8.5 Combine logistic model & balanced data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                        Accuracy  Precision    Recall        F1\n",
      "related                 0.786390   0.809548  0.941858  0.870706\n",
      "request                 0.883125   0.791150  0.408219  0.538554\n",
      "offer                   0.989930   0.022727  0.041667  0.029412\n",
      "aid_related             0.516173   0.459085  0.968935  0.622994\n",
      "medical_help            0.923558   0.516279  0.218504  0.307054\n",
      "medical_products        0.956057   0.585987  0.292063  0.389831\n",
      "search_and_rescue       0.974367   0.581395  0.142857  0.229358\n",
      "security                0.970552   0.171875  0.201835  0.185654\n",
      "military                0.968721   0.564706  0.222222  0.318937\n",
      "water                   0.951175   0.573090  0.845588  0.683168\n",
      "food                    0.947666   0.772997  0.732771  0.752347\n",
      "shelter                 0.948276   0.786223  0.570690  0.661339\n",
      "clothing                0.988862   0.531250  0.645570  0.582857\n",
      "money                   0.972536   0.352941  0.240000  0.285714\n",
      "missing_people          0.987489   0.324324  0.173913  0.226415\n",
      "refugees                0.966585   0.562500  0.196507  0.291262\n",
      "death                   0.965365   0.705882  0.434211  0.537678\n",
      "other_aid               0.867104   0.471154  0.056647  0.101135\n",
      "infrastructure_related  0.908148   0.251429  0.205607  0.226221\n",
      "transport               0.955752   0.560606  0.241830  0.337900\n",
      "buildings               0.957431   0.693548  0.262997  0.381375\n",
      "electricity             0.977418   0.376812  0.198473  0.260000\n",
      "tools                   0.990540   0.026316  0.038462  0.031250\n",
      "hospitals               0.983369   0.255814  0.328358  0.287582\n",
      "shops                   0.990235   0.038462  0.025000  0.030303\n",
      "aid_centers             0.973451   0.159722  0.302632  0.209091\n",
      "other_infrastructure    0.954532   0.185185  0.017794  0.032468\n",
      "weather_related         0.851999   0.732332  0.729086  0.730705\n",
      "floods                  0.954837   0.875740  0.538182  0.666667\n",
      "storm                   0.939426   0.720779  0.554077  0.626529\n",
      "fire                    0.988099   0.413793  0.571429  0.480000\n",
      "earthquake              0.966128   0.856263  0.732865  0.789773\n",
      "cold                    0.979707   0.543353  0.635135  0.585670\n",
      "other_weather           0.946597   0.456522  0.122807  0.193548\n",
      "direct_report           0.842691   0.720000  0.287770  0.411194\n",
      "MEAN SCORE              0.934980   0.498565  0.375045  0.396991\n"
     ]
    }
   ],
   "source": [
    "# Logistic regression\n",
    "# Define ML pipeline\n",
    "pipeline5 = Pipeline([\n",
    "        ('vect', CountVectorizer(tokenizer=tokenize)),\n",
    "        ('tfidf', TfidfTransformer()),\n",
    "        ('clf', MultiOutputClassifier(GradientBoostingClassifier(random_state=1)))\n",
    "    ])\n",
    "\n",
    "# Train pipeline\n",
    "pipeline5.fit(X_train_bal, y_train_bal)\n",
    "\n",
    "# predict on test data\n",
    "y_pred7 = pipeline5.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred7 = pd.DataFrame(y_pred7)\n",
    "y_pred7.columns = target_names\n",
    "\n",
    "# Evaluate the model\n",
    "print(performance_report(y_test, y_pred7, target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Again, not really much improvement with this combination."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 8.6 Create new feature for word embedding vector\n",
    "\n",
    "Word embeddings encourage a vectorisation of words into a feature space where words with a similar meaning occupy a similar part of the feature space. Could these feature help improve performance?\n",
    "\n",
    "The problem is our low occuring classes have few examples, so simply building a new word embedding model from our text data may not help us generalise so well. I will choose to use pre-trained word embedding models to see if this can help.\n",
    "\n",
    "##### Glove\n",
    "\n",
    "Glove is one word embedding model, as described here using a pre-trained model:\n",
    "\n",
    "https://nlp.stanford.edu/projects/glove/\n",
    "\n",
    "http://nadbordrozd.github.io/blog/2016/05/20/text-classification-with-word2vec/\n",
    "\n",
    "This model has been trained on 6 billion tokens, and has a vocabulary of 400,000 words - and creates 300 dimension feature vectors from text.\n",
    "\n",
    "Lets try pre-trained Glove word embeddings."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(\"glove.6B/glove.6B.300d.txt\", \"rb\") as lines:\n",
    "    w2v = {line.split()[0]: np.array(map(float, line.split()[1:]))\n",
    "           for line in lines}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define vectorisers for glove word embeddings\n",
    "class MeanEmbeddingVectorizer(object):\n",
    "    def __init__(self, word2vec):\n",
    "        self.word2vec = word2vec\n",
    "        # if a text is empty we should return a vector of zeros\n",
    "        # with the same dimensionality as all the other vectors\n",
    "        #self.dim = len(word2vec.itervalues().next())\n",
    "        self.dim = 300\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        return self\n",
    "\n",
    "    def transform(self, X):\n",
    "        return np.array([\n",
    "            np.mean([self.word2vec[w] for w in words if w in self.word2vec]\n",
    "                    or [np.zeros(self.dim)], axis=0)\n",
    "            for words in X\n",
    "        ])\n",
    "\n",
    "class TfidfEmbeddingVectorizer(object):\n",
    "    def __init__(self, word2vec):\n",
    "        self.word2vec = word2vec\n",
    "        self.word2weight = None\n",
    "        #self.dim = len(word2vec.itervalues().next())\n",
    "        #self.dim = len(next(iter(word2vec.values())))\n",
    "        self.dim = 300\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        tfidf = TfidfVectorizer(analyzer=lambda x: x)\n",
    "        tfidf.fit(X)\n",
    "        # if a word was never seen - it must be at least as infrequent\n",
    "        # as any of the known words - so the default idf is the max of \n",
    "        # known idf's\n",
    "        max_idf = max(tfidf.idf_)\n",
    "        self.word2weight = defaultdict(\n",
    "            lambda: max_idf,\n",
    "            [(w, tfidf.idf_[i]) for w, i in tfidf.vocabulary_.items()])\n",
    "\n",
    "        return self\n",
    "\n",
    "    def transform(self, X):\n",
    "        return np.array([\n",
    "                np.mean([self.word2vec[w] * self.word2weight[w]\n",
    "                         for w in words if w in self.word2vec] or\n",
    "                        [np.zeros(self.dim)], axis=0)\n",
    "                for words in X\n",
    "            ])\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/metrics/classification.py:1143: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/metrics/classification.py:1143: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                        Accuracy  Precision    Recall        F1\n",
      "related                 0.821941   0.837496  0.951449  0.890843\n",
      "request                 0.901282   0.815493  0.528767  0.641551\n",
      "offer                   0.996338   0.000000  0.000000  0.000000\n",
      "aid_related             0.781660   0.774472  0.664201  0.715110\n",
      "medical_help            0.928898   0.664062  0.167323  0.267296\n",
      "medical_products        0.957125   0.702381  0.187302  0.295739\n",
      "search_and_rescue       0.974672   0.909091  0.057143  0.107527\n",
      "security                0.983369   0.000000  0.000000  0.000000\n",
      "military                0.968874   0.676471  0.106481  0.184000\n",
      "water                   0.959872   0.805907  0.468137  0.592248\n",
      "food                    0.942630   0.849687  0.572433  0.684034\n",
      "shelter                 0.941868   0.813880  0.444828  0.575251\n",
      "clothing                0.989930   0.882353  0.189873  0.312500\n",
      "money                   0.978029   0.666667  0.080000  0.142857\n",
      "missing_people          0.989625   1.000000  0.014493  0.028571\n",
      "refugees                0.965975   0.625000  0.065502  0.118577\n",
      "death                   0.962313   0.890411  0.213816  0.344828\n",
      "other_aid               0.875191   0.645963  0.120231  0.202729\n",
      "infrastructure_related  0.935612   0.625000  0.035047  0.066372\n",
      "transport               0.957736   0.853659  0.114379  0.201729\n",
      "buildings               0.956668   0.759036  0.192661  0.307317\n",
      "electricity             0.982606   0.793103  0.175573  0.287500\n",
      "tools                   0.996033   0.000000  0.000000  0.000000\n",
      "hospitals               0.989625   0.000000  0.000000  0.000000\n",
      "shops                   0.993897   0.000000  0.000000  0.000000\n",
      "aid_centers             0.988404   0.000000  0.000000  0.000000\n",
      "other_infrastructure    0.957278   0.545455  0.021352  0.041096\n",
      "weather_related         0.874123   0.860294  0.648199  0.739336\n",
      "floods                  0.947666   0.915663  0.414545  0.570713\n",
      "storm                   0.934391   0.750733  0.425957  0.543524\n",
      "fire                    0.990388   0.500000  0.015873  0.030769\n",
      "earthquake              0.963076   0.888361  0.657293  0.755556\n",
      "cold                    0.980165   0.875000  0.141892  0.244186\n",
      "other_weather           0.949649   0.833333  0.043860  0.083333\n",
      "direct_report           0.868477   0.775887  0.437250  0.559305\n",
      "MEAN SCORE              0.948154   0.643853  0.233025  0.300983\n"
     ]
    }
   ],
   "source": [
    "# Logistic regression\n",
    "# Define ML pipeline\n",
    "\n",
    "pipeline4 = Pipeline([\n",
    "    ('features', FeatureUnion([\n",
    "\n",
    "        ('text_pipeline', Pipeline([\n",
    "            ('vect', CountVectorizer(tokenizer=tokenize)),\n",
    "            ('tfidf', TfidfTransformer())\n",
    "        ])),\n",
    "\n",
    "        ('word2vec_vectorizer', MeanEmbeddingVectorizer(w2v))\n",
    "    ])),\n",
    "\n",
    "    ('clf', MultiOutputClassifier(LogisticRegression(random_state=1)))\n",
    "])\n",
    "\n",
    "\n",
    "\n",
    "# Train pipeline\n",
    "pipeline4.fit(X_train, y_train)\n",
    "\n",
    "# predict on test data\n",
    "y_pred6 = pipeline4.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred6 = pd.DataFrame(y_pred6)\n",
    "y_pred6.columns = target_names\n",
    "\n",
    "# Evaluate the model\n",
    "print(performance_report(y_test, y_pred6, target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So the glove pre-trained embedding feature vectoriser has not helped improve performance.\n",
    "\n",
    "##### Doc2vec\n",
    "\n",
    "Lets look at Doc2Vec pre-trained word embedding models, as described here:\n",
    "\n",
    "http://ai.intelligentonlinetools.com/ml/text-clustering-doc2vec-word-embedding-machine-learning/\n",
    "\n",
    "https://github.com/jhlau/doc2vec\n",
    "\n",
    "https://markroxor.github.io/gensim/static/notebooks/doc2vec-wikipedia.html\n",
    "\n",
    "https://arxiv.org/pdf/1507.07998.pdf\n",
    "\n",
    "We will used a pre-trained model trained on 4.5 million wikipedia articles, and has a vocabulary of just under 1 million words, that also generates 300 dimension feature vectors.\n",
    "\n",
    "Note: default doc2vec from genism library does not facilitate pre-trained models, to download a forked version that can install genism fresh via 'pip install -U gensim'."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 1.05389304e-01  1.87241346e-01 -1.00067124e-01 -5.84028900e-01\n",
      "  1.07543826e-01 -1.25103831e-01  5.27499676e-01 -3.75652641e-01\n",
      "  1.17975600e-01 -2.81796735e-02 -2.52026796e-01 -5.09442866e-01\n",
      " -3.21979113e-02  5.03063910e-02 -9.26744118e-02  7.11919814e-02\n",
      " -3.74514908e-01 -1.73789307e-01 -4.19825137e-01 -1.91427618e-01\n",
      " -3.20386916e-01 -1.20774277e-01  6.60605073e-01 -2.06124559e-01\n",
      "  2.82651335e-01 -1.03867702e-01 -3.02995201e-02  2.58385479e-01\n",
      "  2.15882078e-01 -3.05927601e-02 -1.70235932e-01  2.70004213e-01\n",
      " -1.28700301e-01  4.80530083e-01  4.43913430e-01 -3.66875350e-01\n",
      "  2.82368422e-01  8.34432766e-02  2.97592998e-01  2.10374564e-01\n",
      " -4.34435606e-01 -2.89152443e-01  1.74317762e-01 -5.44270277e-01\n",
      " -2.76181340e-01  1.26449972e-01  1.77189916e-01  3.56412947e-01\n",
      " -7.53176287e-02 -7.59698525e-02 -5.08393534e-03 -1.90379962e-01\n",
      " -2.81680245e-02 -2.49372572e-01 -3.11455935e-01 -4.48736221e-01\n",
      " -9.68318954e-02 -1.48736425e-02 -1.02294259e-01 -7.75747327e-03\n",
      " -3.89026962e-02 -3.53648901e-01  2.12384481e-02 -6.82352483e-02\n",
      " -3.03487852e-02 -1.50125325e-01 -3.58133823e-01 -6.09233022e-01\n",
      " -3.81758779e-01 -4.49005067e-02 -3.47688615e-01  1.67845100e-01\n",
      " -2.41994172e-01  2.94996083e-01 -3.56902510e-01  5.75283051e-01\n",
      "  2.20480978e-01  5.77215962e-02 -1.93335771e-04  1.19412899e-01\n",
      "  2.76499063e-01  2.07106844e-01  3.19685638e-01 -1.37168050e-01\n",
      " -5.50463855e-01 -3.37839425e-02 -1.16108932e-01 -4.93645877e-01\n",
      "  5.59483051e-01 -2.56280750e-01  1.56705007e-01 -7.06306279e-01\n",
      "  4.33880061e-01 -1.59177184e-02  3.90694663e-02 -1.08630218e-01\n",
      " -1.33628622e-01 -5.64078033e-01  2.13117555e-01 -4.34925705e-01\n",
      " -3.03317815e-01 -5.24564506e-03 -2.17467532e-01  3.27868424e-02\n",
      "  9.21066254e-02 -5.56944549e-01 -3.10580522e-01  2.57777303e-01\n",
      " -3.60073298e-01  6.58024475e-02  1.20447285e-01  1.29155265e-02\n",
      " -5.24465978e-01  4.88230363e-02 -4.25387174e-01  2.87235051e-01\n",
      "  3.18777896e-02 -2.11745679e-01  1.84999883e-01  2.98808426e-01\n",
      " -3.43494326e-01 -1.80135339e-01  1.58824891e-01  5.82401529e-02\n",
      " -3.17710519e-01  4.77799252e-02 -3.02506387e-01 -4.96183634e-01\n",
      " -6.63834857e-04  7.76851326e-02  5.28123379e-01 -4.31440145e-01\n",
      " -6.12049848e-02 -2.87659407e-01 -2.06038337e-02 -2.22582862e-01\n",
      "  2.46728405e-01 -4.47185159e-01 -1.48763984e-01 -4.51815873e-02\n",
      " -3.21196113e-03  4.80970368e-02 -1.00949056e-01 -1.40826434e-01\n",
      " -4.52147834e-02  4.88156587e-01 -4.15200703e-02  4.25319709e-02\n",
      "  1.69473708e-01  6.46054804e-01  3.89645487e-01 -6.78750053e-02\n",
      "  8.71650130e-02  3.65553558e-01  1.53704941e-01  2.86747403e-02\n",
      "  1.21337667e-01 -2.84405023e-01 -1.91235822e-02 -2.05272570e-01\n",
      "  3.08977574e-01 -9.99808386e-02  4.07513976e-03  9.31824073e-02\n",
      "  2.64750570e-01 -1.31404281e-01 -1.49814601e-04  1.53074622e-01\n",
      "  6.28841758e-01  3.48244369e-01  1.74350873e-01  6.59637004e-02\n",
      "  4.02323455e-02  3.54363397e-02  4.65461403e-01 -7.28858262e-02\n",
      " -2.82445550e-01  3.21590036e-01 -4.98675555e-02  1.72930047e-01\n",
      "  1.25678182e-01  3.72142702e-01 -2.15357542e-01 -5.22496223e-01\n",
      "  2.62582004e-01  9.45821181e-02 -2.59785019e-02  1.95713826e-02\n",
      "  3.88473161e-02  5.14054596e-01 -2.16810539e-01 -9.72511828e-01\n",
      "  3.00325274e-01  2.85277188e-01 -1.03199268e-02  7.18292817e-02\n",
      "  3.76416296e-01  5.63194342e-02  2.66980886e-01 -4.77588713e-01\n",
      " -1.06341138e-01 -2.06976146e-01  2.45595410e-01 -1.67213410e-01\n",
      " -1.26754090e-01 -4.73161101e-01 -1.27872797e-02  1.69411704e-01\n",
      " -4.81302850e-02 -2.96189994e-01  4.11287546e-02  1.31624877e-01\n",
      " -1.79160595e-01 -3.64933997e-01 -4.30458874e-01 -1.42753460e-02\n",
      " -1.47867873e-01 -3.35441083e-01  4.02036868e-02 -9.18255362e-04\n",
      "  1.73511401e-01 -2.61289924e-01  1.16360940e-01  1.58215418e-01\n",
      " -4.28372651e-01 -9.63335857e-02 -6.09309673e-01  2.36605555e-01\n",
      " -1.40614927e-01  8.57911929e-02 -2.31171742e-01 -2.99862951e-01\n",
      "  1.71676800e-01  1.37325093e-01  2.99193442e-01  5.26120812e-02\n",
      " -1.26248449e-01  6.05177395e-02 -9.00050774e-02  2.21691027e-01\n",
      " -1.54082119e-01  7.62394071e-02 -2.76269257e-01  1.12100869e-01\n",
      " -3.78787220e-02  1.29795298e-01  4.99427170e-02 -1.84421450e-01\n",
      "  2.45877951e-01 -3.05840015e-01 -9.34068114e-02 -7.05072433e-02\n",
      "  4.19345140e-01 -3.67058963e-01 -1.13139702e-02  3.21381271e-01\n",
      "  4.45715874e-01 -8.68384689e-02 -3.28221500e-01 -3.74791771e-01\n",
      "  2.66350657e-01 -4.69112664e-01 -1.67671531e-01 -1.57059565e-01\n",
      "  3.18081081e-02 -2.83150673e-01  1.84176471e-02  3.75904553e-02\n",
      " -1.24146581e-01  3.09123397e-01 -3.93935084e-01 -1.85306579e-01\n",
      " -1.67175665e-01 -3.41573030e-01 -2.77022541e-01  3.10019165e-01\n",
      "  4.27478939e-01  1.90804005e-01  3.81326228e-02 -1.14155270e-01\n",
      " -1.80202514e-01  5.07765830e-01 -1.98834732e-01  7.46126706e-03\n",
      "  3.71668428e-01  2.85068333e-01 -1.07950699e-02 -1.65655941e-01\n",
      "  3.18403214e-01 -4.09968853e-01 -1.96621358e-01  2.01632261e-01\n",
      " -2.01181427e-01 -6.81699812e-02 -9.97488387e-03 -1.21440060e-01\n",
      " -2.03540072e-01  4.44254816e-01 -1.67612266e-02 -4.03775275e-01]\n"
     ]
    }
   ],
   "source": [
    "# Load pre-trained model from file \n",
    "model=\"doc2vec/doc2vec.bin\"\n",
    " \n",
    "#inference hyper-parameters\n",
    "start_alpha=0.01\n",
    "infer_epoch=1000\n",
    "\n",
    "# Load pre-trained model\n",
    "doc2vec = g.Doc2Vec.load(model)\n",
    "\n",
    "# Lets show an example 300 feature vector for the first message of our training set\n",
    "document = tokenize(X_train[0])\n",
    "vector = doc2vec.infer_vector(document, alpha=start_alpha, steps=infer_epoch)\n",
    "print((vector))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/logistic.py:433: FutureWarning: Default solver will be changed to 'lbfgs' in 0.22. Specify a solver to silence this warning.\n",
      "  FutureWarning)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/metrics/classification.py:1143: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n",
      "/anaconda3/lib/python3.6/site-packages/sklearn/metrics/classification.py:1143: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                        Accuracy  Precision    Recall        F1\n",
      "related                 0.824382   0.856984  0.924276  0.889359\n",
      "request                 0.887702   0.773212  0.463927  0.579909\n",
      "offer                   0.996338   0.000000  0.000000  0.000000\n",
      "aid_related             0.767013   0.755315  0.643861  0.695149\n",
      "medical_help            0.928898   0.600962  0.246063  0.349162\n",
      "medical_products        0.956057   0.613445  0.231746  0.336406\n",
      "search_and_rescue       0.974367   0.684211  0.074286  0.134021\n",
      "security                0.982911   0.200000  0.009174  0.017544\n",
      "military                0.972383   0.633588  0.384259  0.478386\n",
      "water                   0.950870   0.691964  0.379902  0.490506\n",
      "food                    0.935459   0.749135  0.609001  0.671839\n",
      "shelter                 0.942478   0.773585  0.494828  0.603575\n",
      "clothing                0.990235   0.674419  0.367089  0.475410\n",
      "money                   0.978334   0.564516  0.233333  0.330189\n",
      "missing_people          0.989319   0.333333  0.014493  0.027778\n",
      "refugees                0.967196   0.594595  0.192140  0.290429\n",
      "death                   0.959414   0.663793  0.253289  0.366667\n",
      "other_aid               0.868477   0.509677  0.091329  0.154902\n",
      "infrastructure_related  0.935154   0.526316  0.070093  0.123711\n",
      "transport               0.958499   0.757576  0.163399  0.268817\n",
      "buildings               0.955752   0.641221  0.256881  0.366812\n",
      "electricity             0.983522   0.734694  0.274809  0.400000\n",
      "tools                   0.995575   0.200000  0.038462  0.064516\n",
      "hospitals               0.989625   0.333333  0.014925  0.028571\n",
      "shops                   0.993744   0.000000  0.000000  0.000000\n",
      "aid_centers             0.987946   0.000000  0.000000  0.000000\n",
      "other_infrastructure    0.957278   0.513514  0.067616  0.119497\n",
      "weather_related         0.881446   0.816893  0.734072  0.773271\n",
      "floods                  0.949496   0.819242  0.510909  0.629339\n",
      "storm                   0.937138   0.706783  0.537438  0.610586\n",
      "fire                    0.991150   0.727273  0.126984  0.216216\n",
      "earthquake              0.969332   0.833333  0.808436  0.820696\n",
      "cold                    0.982148   0.724638  0.337838  0.460829\n",
      "other_weather           0.947360   0.469388  0.067251  0.117647\n",
      "direct_report           0.855050   0.738510  0.372502  0.495218\n",
      "MEAN SCORE              0.946916   0.577584  0.285560  0.353913\n"
     ]
    }
   ],
   "source": [
    "# Create a transformer class for our pipeline that uses the pre-trained doc2vec model\n",
    "# Logistic regression\n",
    "class Doc2vecVectorizer(object):\n",
    "    def __init__(self, doc2vec):\n",
    "        self.doc2vec = doc2vec\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        return self\n",
    "\n",
    "    def transform(self, X):\n",
    "        return np.asmatrix(np.array([self.doc2vec.infer_vector(tokenize(document)) for document in X]))\n",
    "\n",
    "    \n",
    "    \n",
    "# Define new pipeline\n",
    "pipeline6 = Pipeline([\n",
    "        ('doc2vec_vectorizer', Doc2vecVectorizer(doc2vec)),\n",
    "        ('clf', MultiOutputClassifier(LogisticRegression(random_state=1)))\n",
    "    ])\n",
    "\n",
    "# Train pipeline\n",
    "pipeline6.fit(X_train, y_train)\n",
    "\n",
    "# predict on test data\n",
    "y_pred7 = pipeline6.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred7 = pd.DataFrame(y_pred7)\n",
    "y_pred7.columns = target_names\n",
    "\n",
    "# Evaluate the model\n",
    "print(performance_report(y_test, y_pred7, target_names))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                        Accuracy  Precision    Recall        F1\n",
      "related                 0.826060   0.847509  0.941658  0.892107\n",
      "request                 0.884956   0.787521  0.426484  0.553318\n",
      "offer                   0.992524   0.000000  0.000000  0.000000\n",
      "aid_related             0.756485   0.738587  0.634246  0.682451\n",
      "medical_help            0.931492   0.680982  0.218504  0.330849\n",
      "medical_products        0.953158   0.540000  0.171429  0.260241\n",
      "search_and_rescue       0.970857   0.192308  0.028571  0.049751\n",
      "security                0.977571   0.047619  0.018349  0.026490\n",
      "military                0.970095   0.590909  0.300926  0.398773\n",
      "water                   0.942936   0.639344  0.191176  0.294340\n",
      "food                    0.917608   0.713217  0.402250  0.514388\n",
      "shelter                 0.934239   0.792157  0.348276  0.483832\n",
      "clothing                0.986421   0.391304  0.227848  0.288000\n",
      "money                   0.975587   0.407407  0.146667  0.215686\n",
      "missing_people          0.982911   0.042553  0.028986  0.034483\n",
      "refugees                0.962618   0.362069  0.091703  0.146341\n",
      "death                   0.957431   0.634409  0.194079  0.297229\n",
      "other_aid               0.867409   0.464286  0.030058  0.056460\n",
      "infrastructure_related  0.933933   0.424242  0.032710  0.060738\n",
      "transport               0.954837   0.580645  0.117647  0.195652\n",
      "buildings               0.956363   0.741176  0.192661  0.305825\n",
      "electricity             0.978181   0.380000  0.145038  0.209945\n",
      "tools                   0.991913   0.034483  0.038462  0.036364\n",
      "hospitals               0.986115   0.071429  0.029851  0.042105\n",
      "shops                   0.991150   0.000000  0.000000  0.000000\n",
      "aid_centers             0.983064   0.073171  0.039474  0.051282\n",
      "other_infrastructure    0.956210   0.411765  0.049822  0.088889\n",
      "weather_related         0.867867   0.807466  0.683102  0.740096\n",
      "floods                  0.936527   0.765873  0.350909  0.481297\n",
      "storm                   0.929204   0.684636  0.422629  0.522634\n",
      "fire                    0.985200   0.075000  0.047619  0.058252\n",
      "earthquake              0.956973   0.809935  0.659051  0.726744\n",
      "cold                    0.976655   0.460317  0.195946  0.274882\n",
      "other_weather           0.946597   0.346154  0.026316  0.048913\n",
      "direct_report           0.848184   0.725352  0.329337  0.452996\n",
      "MEAN SCORE              0.941981   0.464681  0.221765  0.280610\n"
     ]
    }
   ],
   "source": [
    "# Define new pipeline\n",
    "# Gradient boosting\n",
    "pipeline7 = Pipeline([\n",
    "        ('doc2vec_vectorizer', Doc2vecVectorizer(doc2vec)),\n",
    "        ('clf', MultiOutputClassifier(GradientBoostingClassifier(random_state=1)))\n",
    "    ])\n",
    "\n",
    "# Train pipeline\n",
    "pipeline7.fit(X_train, y_train)\n",
    "\n",
    "# predict on test data\n",
    "y_pred8 = pipeline7.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred8 = pd.DataFrame(y_pred8)\n",
    "y_pred8.columns = target_names\n",
    "\n",
    "# Evaluate the model\n",
    "print(performance_report(y_test, y_pred8, target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 9. Summary of best improvements\n",
    "\n",
    "So the model with the best f1 score was the gradient boosting model, with balanced data & starting verb feature - which gives us a mean F1 score of 0.407.\n",
    "\n",
    "We also know that while this is the mean, there is high variance in the results for each category - some target categories have very good scores, while others have very poor scores. The possible reasons for this highlighted earlier include:\n",
    "\n",
    "- Some categories have very few examples (around 100)\n",
    "- The content of the messages for many of these low value categories seem particularly difficult to correlate (to humans anyway!)\n",
    "\n",
    "Therefore, with such variable results it is difficult to see how we could deploy even this best model. However, if we use a 'triage' approach from medicine - if we only trained a model on those categories that have an F1 score of more than 0.5, while it could not be used to predict certain categories, those that it did it could do so more reliably - making the whole model more reliable.\n",
    "\n",
    "For our best model, the categories (excluding the questionable 'related' category) that have a F1 score of at least 0.5 are:\n",
    "\n",
    "- earthquake - 0.790918\n",
    "- food - 0.759219\n",
    "- weather_related - 0.712544\n",
    "- water - 0.676379\n",
    "- floods - 0.659142\n",
    "- shelter - 0.641283\n",
    "- aid_related - 0.636194\n",
    "- storm - 0.613615\n",
    "- clothing - 0.598802\n",
    "- cold - 0.587156\n",
    "- request - 0.547215\n",
    "- death - 0.540881\n",
    "- direct_report - 0.505040\n",
    "\n",
    "We will therefore re-train our best model, to target only these categories and will then run a grid search on these in our command line application."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                 Accuracy  Precision    Recall        F1\n",
      "aid_related      0.749466   0.780359  0.546598  0.642888\n",
      "water            0.962313   0.720548  0.644608  0.680466\n",
      "food             0.952090   0.789781  0.760900  0.775072\n",
      "shelter          0.949344   0.805419  0.563793  0.663286\n",
      "clothing         0.987946   0.500000  0.569620  0.532544\n",
      "death            0.966738   0.706731  0.483553  0.574219\n",
      "weather_related  0.872597   0.875387  0.626593  0.730384\n",
      "floods           0.958041   0.889518  0.570909  0.695460\n",
      "storm            0.944309   0.735060  0.613977  0.669084\n",
      "earthquake       0.971315   0.835979  0.833040  0.834507\n",
      "cold             0.980317   0.587156  0.432432  0.498054\n",
      "request          0.900214   0.847244  0.491324  0.621965\n",
      "direct_report    0.862374   0.799314  0.372502  0.508179\n",
      "MEAN SCORE       0.927466   0.759423  0.577681  0.648162\n"
     ]
    }
   ],
   "source": [
    "# Subset data to predict only top performing categories\n",
    "target_names_best = ['aid_related', 'water', 'food', 'shelter', 'clothing', 'death', 'weather_related', 'floods', 'storm', 'earthquake', 'cold', 'request', 'direct_report']\n",
    "Y_best = df[target_names_best]\n",
    "# Split data into train and test\n",
    "X_train, X_test, Y_train_best, Y_test_best = train_test_split(X, Y_best, random_state=49)\n",
    "\n",
    "# Run best model on data\n",
    "pipeline_best = Pipeline([\n",
    "        ('vect', CountVectorizer(tokenizer=tokenize)),\n",
    "        ('tfidf', TfidfTransformer()),\n",
    "        ('clf', MultiOutputClassifier(GradientBoostingClassifier(random_state=1)))\n",
    "    ])\n",
    "\n",
    "# Train pipeline\n",
    "pipeline_best.fit(X_train, Y_train_best)\n",
    "\n",
    "# predict on test data\n",
    "y_pred8 = pipeline_best.predict(X_test)\n",
    "\n",
    "# Re-structure returned numpy array as dataframe with same column names\n",
    "y_pred8 = pd.DataFrame(y_pred8)\n",
    "y_pred8.columns = target_names_best\n",
    "\n",
    "# Evaluate the model\n",
    "print(performance_report(y_test, y_pred8, target_names_best))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This model now has a mean F1 score of 0.65 for the top 13 target categories.\n",
    "\n",
    "Let us see if we can understand the most important tfidf features....\n",
    "\n",
    "https://buhrmann.github.io/tfidf-analysis.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Xtr = pipeline_best.fit_transform(X_train)\n",
    "#vec = pipeline_best.named_steps['vect']\n",
    "#features = vec.get_feature_names()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [],
   "source": [
    "# we create a function that takes a single row of the tf-idf matrix (corresponding to a particular document), \n",
    "# and return the n highest scoring words (or more generally tokens or features):\n",
    "def top_tfidf_feats(row, features, top_n=25):\n",
    "    ''' Get top n tfidf values in row and return them with their corresponding feature names.'''\n",
    "    topn_ids = np.argsort(row)[::-1][:top_n]\n",
    "    top_feats = [(features[i], row[i]) for i in topn_ids]\n",
    "    df = pd.DataFrame(top_feats)\n",
    "    df.columns = ['feature', 'tfidf']\n",
    "    return df\n",
    "\n",
    "# The result of a tf-idf, however, is typically a sparse matrix, which doesn’t support all the usual matrix or array \n",
    "# operations. So in order to apply the above function to inspect a particular document, we convert a single row into \n",
    "# dense format first:\n",
    "def top_feats_in_doc(Xtr, features, row_id, top_n=25):\n",
    "    ''' Top tfidf features in specific document (matrix row) '''\n",
    "    row = np.squeeze(Xtr[row_id].toarray())\n",
    "    return top_tfidf_feats(row, features, top_n)\n",
    "\n",
    "#top_feats_in_doc(Xtr, features, 3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 10. Export the model and subset data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['final_model.sav']"
      ]
     },
     "execution_count": 90,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# save the model as pickle file to disk\n",
    "filename = 'final_model.sav'\n",
    "joblib.dump(pipeline_best, 'final_model.sav')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Merge optimised features & targets\n",
    "features_opt = pd.Series.to_frame(X)\n",
    "training__opt_df = features_opt.join(Y_best)\n",
    "# Delete previous db\n",
    "#os.remove(\"DisasterResponseOpt.db\")\n",
    "# Save current version of database\n",
    "engine = db.create_engine('sqlite:///DisasterResponseOpt.db')\n",
    "training__opt_df.to_sql('DisasterResponseOpt', engine, index=False)"
   ]
  },
  {
   "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}