misspellings_approaches.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "949ea94b-fe31-4e36-a450-2d9d22a247d5",
   "metadata": {},
   "source": [
    "# Sentence encoders: Misspellings approaches"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "33ee072e-9578-4af2-8b13-4fc56a1a5474",
   "metadata": {},
   "source": [
    "This notebook investigates the different misspellings approaches that can be taken when using tensorflow sentence encoders on a [Kaggle essay scores](https://www.kaggle.com/competitions/learning-agency-lab-automated-essay-scoring-2) dataset. Varying the approaches between doing nothing, dropping misspellings and correcting misspellings shows us that dropping is likely the most effective strategy.   \n",
    "\n",
    "Due to how long it takes the spellchecker library to correct misspellings (~ 10 seconds per essay), the dataset is limited to 900 random essays with equal score distributions. The model is then fit and evaluated on the embeddings for each approach 3 times to obtain more stable accuracy averages. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "5a9104aa-d3ee-4030-b29f-a3b63c59df01",
   "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>essay_id</th>\n",
       "      <th>full_text</th>\n",
       "      <th>score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>000d118</td>\n",
       "      <td>Many people have car where they live. The thin...</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>000fe60</td>\n",
       "      <td>I am a scientist at NASA that is discussing th...</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>001ab80</td>\n",
       "      <td>People always wish they had the same technolog...</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>001bdc0</td>\n",
       "      <td>We all heard about Venus, the planet without a...</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>002ba53</td>\n",
       "      <td>Dear, State Senator\\n\\nThis is a letter to arg...</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  essay_id                                          full_text  score\n",
       "0  000d118  Many people have car where they live. The thin...      3\n",
       "1  000fe60  I am a scientist at NASA that is discussing th...      3\n",
       "2  001ab80  People always wish they had the same technolog...      4\n",
       "3  001bdc0  We all heard about Venus, the planet without a...      4\n",
       "4  002ba53  Dear, State Senator\\n\\nThis is a letter to arg...      3"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "\n",
    "train = pd.read_csv('data/train.csv')\n",
    "train.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5f832c02-3f34-4907-9dff-6ab664c02334",
   "metadata": {},
   "source": [
    "## Limit dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "900ad3ad-931a-4dca-bc60-b3917c170546",
   "metadata": {},
   "source": [
    "The dataset is limited to a random selection of 900 essays with an equal score distribution. The value counts for each score are first listed. A random selection no larger than the minimum value count is then retrieved for each score."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "094201ba-12ce-4325-866d-df330c5d7fce",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "score\n",
       "3    6280\n",
       "2    4723\n",
       "4    3926\n",
       "1    1252\n",
       "5     970\n",
       "6     156\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# lists values counts\n",
    "train.score.value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "6548e9b7-18f5-4976-82e2-52f16ce6cb30",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "900"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import random\n",
    "\n",
    "# randomly selects 150 essays for each score (1-6)\n",
    "random_index = []\n",
    "for i in range(6):\n",
    "    random_index.extend( random.sample(list(train.loc[train.score==i+1].index), 150) )\n",
    "\n",
    "# reassigns the train dataframe to 900 essays\n",
    "train = train.loc[random_index]\n",
    "len(train)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c80bbd42-77ea-4d23-a052-69c482c412a2",
   "metadata": {},
   "source": [
    "## Preprocess text"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "30950f63-ef50-4cc8-9422-55715bb36b1b",
   "metadata": {},
   "source": [
    "Common preprocessing steps are applied to simplify the texts and reduce potential noise."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "c506b32b-643d-4fcd-8dff-da1af1c16889",
   "metadata": {},
   "outputs": [],
   "source": [
    "import re\n",
    "import string\n",
    "import nltk\n",
    "from nltk.tokenize import word_tokenize\n",
    "from nltk.corpus import stopwords\n",
    "from nltk.stem import WordNetLemmatizer\n",
    "\n",
    "def preprocess_text(text):\n",
    "\n",
    "    # removes unprintable characters\n",
    "    unprintable = r\"[^\\w\\s]\"\n",
    "    text = re.sub(unprintable, '', text)\n",
    "\n",
    "    # removes non-breaking space characters\n",
    "    non_breaking = '\\xa0'\n",
    "    text = re.sub(non_breaking, ' ', text)\n",
    "\n",
    "    # removes newline characters\n",
    "    newline = '\\n'\n",
    "    text = re.sub(newline, ' ', text)\n",
    "\n",
    "    # removes punctuation\n",
    "    text = text.translate(str.maketrans('', '', string.punctuation))\n",
    "    \n",
    "    # lowercase the text\n",
    "    text = text.lower()\n",
    "    \n",
    "    # tokenization\n",
    "    tokens = word_tokenize(text)\n",
    "    \n",
    "    # remove stopwords\n",
    "    stop_words = set(stopwords.words('english'))\n",
    "    tokens = [token for token in tokens if token not in stop_words]\n",
    "    \n",
    "    # lemmatization\n",
    "    lemmatizer = WordNetLemmatizer()\n",
    "    tokens = [lemmatizer.lemmatize(token) for token in tokens]\n",
    "    \n",
    "    # join tokens back into a string\n",
    "    preprocessed_text = ' '.join(tokens)\n",
    "    \n",
    "    return preprocessed_text"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c3ef8230-75d4-45c4-9275-0ff1204e0bcf",
   "metadata": {},
   "source": [
    "## Misspelling function"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5fcbe45d-0f28-4000-a91b-1443d5859beb",
   "metadata": {},
   "source": [
    "A misspelling function splits the essay into words and either drops or corrects unknown words."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "e712edc1-6cb0-43b2-a1d5-0ac19999fc32",
   "metadata": {},
   "outputs": [],
   "source": [
    "from spellchecker import SpellChecker\n",
    "\n",
    "# creates a spellckecker object\n",
    "spell = SpellChecker()\n",
    "\n",
    "# returns a count of misspelled words\n",
    "def misspellings(essay, drop=True):\n",
    "\n",
    "    # the essay is split into words\n",
    "    words = essay.split()\n",
    "\n",
    "    # either drops or corrects unknown words\n",
    "    final_words = []\n",
    "    if drop == True:\n",
    "        # loops through each word\n",
    "        for word in words:\n",
    "            \n",
    "            # checks if the word is known\n",
    "            if spell.known([word]):\n",
    "                \n",
    "                # appends to a list of words to return\n",
    "                final_words.append(word)\n",
    "    \n",
    "    else:\n",
    "        # loops through each word\n",
    "        for word in words:\n",
    "            \n",
    "            # checks if the word is known\n",
    "            if spell.known([word]):\n",
    "                \n",
    "                # appends to a list of words to return\n",
    "                final_words.append(word)\n",
    "\n",
    "            else:\n",
    "                # attempts to correct the word\n",
    "                correction = spell.correction(word)\n",
    "\n",
    "                # appends the correction if one was found\n",
    "                if correction != None:\n",
    "                    # appends to a list of words to return\n",
    "                    final_words.append(correction)\n",
    "\n",
    "    # joins the words back together with a space \n",
    "    return ' '.join(final_words)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "74f912ef-9844-4697-8bdd-3979685a378a",
   "metadata": {},
   "source": [
    "## Model functions"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "008ce65e-50bf-4235-aa41-e0e3adf0577c",
   "metadata": {},
   "source": [
    "The sentence encoder model is first loaded-in. A number of functions are then defined to split the essay texts in into a variable number of words, obtain the embeddings, fit a model and then record the time and accuracy results."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "340d34ba-8b24-4d10-a950-984d2e7eeb11",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\Users\\Jack-\\AppData\\Local\\Programs\\Python\\Python311\\Lib\\site-packages\\tf_keras\\src\\losses.py:2976: The name tf.losses.sparse_softmax_cross_entropy is deprecated. Please use tf.compat.v1.losses.sparse_softmax_cross_entropy instead.\n",
      "\n",
      "WARNING:tensorflow:From C:\\Users\\Jack-\\AppData\\Local\\Programs\\Python\\Python311\\Lib\\site-packages\\tensorflow_hub\\resolver.py:120: The name tf.gfile.MakeDirs is deprecated. Please use tf.io.gfile.makedirs instead.\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\Users\\Jack-\\AppData\\Local\\Programs\\Python\\Python311\\Lib\\site-packages\\tensorflow_hub\\resolver.py:120: The name tf.gfile.MakeDirs is deprecated. Please use tf.io.gfile.makedirs instead.\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\Users\\Jack-\\AppData\\Local\\Programs\\Python\\Python311\\Lib\\site-packages\\tensorflow_hub\\module_v2.py:126: The name tf.saved_model.load_v2 is deprecated. Please use tf.compat.v2.saved_model.load instead.\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\Users\\Jack-\\AppData\\Local\\Programs\\Python\\Python311\\Lib\\site-packages\\tensorflow_hub\\module_v2.py:126: The name tf.saved_model.load_v2 is deprecated. Please use tf.compat.v2.saved_model.load instead.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "import tensorflow_hub as hub\n",
    "from sklearn.model_selection import train_test_split\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Input, Embedding\n",
    "from sklearn.metrics import cohen_kappa_score\n",
    "\n",
    "# downloads the model to obtain embeddings \n",
    "model_url = \"https://tfhub.dev/google/universal-sentence-encoder/4\"\n",
    "encoder_model = hub.load(model_url)\n",
    "\n",
    "# returns the essay embeddings\n",
    "def encode_sentences(essays):\n",
    "    embeddings = encoder_model(essays)\n",
    "    return embeddings.numpy()\n",
    "\n",
    "# creates the model\n",
    "def create_model(embedding_dim):\n",
    "    model = Sequential([\n",
    "          Input(shape=(embedding_dim,)),\n",
    "          Dense(64, activation='relu'),\n",
    "          Dense(32, activation='relu'),\n",
    "          Dense(7, activation='softmax')\n",
    "      ])\n",
    "    model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "    return model\n",
    "\n",
    "# fits the model on the training data and evaluates accuracy\n",
    "def fit_model(embeddings, scores, accuracy_df, approach):\n",
    "\n",
    "    # defines how many times the model should be fit and evaluated to obain a more stable accuracy average\n",
    "    turns = 3\n",
    "\n",
    "    # stores each accuracy result in order to obtain an average\n",
    "    accuracy_scores = []\n",
    "\n",
    "    # loops through each turn\n",
    "    for i in range(turns):\n",
    "\n",
    "        # splits the data into train and validation sets\n",
    "        x_train, x_val, y_train, y_val = train_test_split(embeddings, scores, test_size=0.3)\n",
    "    \n",
    "        # creates and embedding dimension and uses it to create a model\n",
    "        embedding_dim = embeddings[0].shape[0]\n",
    "        model = create_model(embedding_dim)\n",
    "    \n",
    "        # converts essay scores to one hot encodings\n",
    "        y_train_one_hot = to_categorical(y_train)\n",
    "        y_val_one_hot = to_categorical(y_val)\n",
    "    \n",
    "        # fits the model\n",
    "        model.fit(x_train, y_train_one_hot, epochs=12, batch_size=32, validation_data=(x_val, y_val_one_hot)) \n",
    "    \n",
    "        # retrieves predictions \n",
    "        predictions = model.predict(x_val)\n",
    "    \n",
    "        # converts raw predictions (decimals) into final predictions (1 - 6 integers)\n",
    "        converted_preds = [np.argmax(pred) for pred in predictions]\n",
    "        converted_y_val = [np.argmax(y) for y in y_val_one_hot]\n",
    "    \n",
    "        # appends rounded Cohen Kappa scores \n",
    "        accuracy_scores.append(round(cohen_kappa_score(converted_preds, converted_y_val, weights='quadratic'), 2))\n",
    "    \n",
    "    # adds the average accuracy score to the accuracy dataframe\n",
    "    accuracy_df.loc[0, approach] = round(np.mean(accuracy_scores), 2)\n",
    "    \n",
    "    return accuracy_df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "145bb180-fab5-4d70-8cb8-ea6f3f907309",
   "metadata": {},
   "source": [
    "## Misspellings approaches"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "26cb8762-7256-40f9-b4ef-e206c41d27c6",
   "metadata": {},
   "source": [
    "The data is prepared and each approach is carried out as part of a loop. First, the relavant misspellings approach is applied to the text. Second, the embeddings are retrieved. Third, a model is fit and evaluated 3 times to obtain a more stable average. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "174dbfd7-7cf5-4756-b3fd-1b2443337e66",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Preparing embeddings: nothing\n",
      "Fitting model: nothing\n",
      "Epoch 1/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 26ms/step - accuracy: 0.2036 - loss: 1.9284 - val_accuracy: 0.3370 - val_loss: 1.8652\n",
      "Epoch 2/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.3782 - loss: 1.8192 - val_accuracy: 0.3111 - val_loss: 1.7225\n",
      "Epoch 3/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.3797 - loss: 1.6461 - val_accuracy: 0.3407 - val_loss: 1.5604\n",
      "Epoch 4/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.4141 - loss: 1.4581 - val_accuracy: 0.3185 - val_loss: 1.4371\n",
      "Epoch 5/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.4956 - loss: 1.3342 - val_accuracy: 0.3704 - val_loss: 1.3709\n",
      "Epoch 6/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.5177 - loss: 1.2488 - val_accuracy: 0.4037 - val_loss: 1.3297\n",
      "Epoch 7/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.5059 - loss: 1.2111 - val_accuracy: 0.4111 - val_loss: 1.3127\n",
      "Epoch 8/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.5485 - loss: 1.1410 - val_accuracy: 0.4074 - val_loss: 1.2832\n",
      "Epoch 9/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.5908 - loss: 1.1111 - val_accuracy: 0.4111 - val_loss: 1.2786\n",
      "Epoch 10/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 11ms/step - accuracy: 0.6187 - loss: 1.0576 - val_accuracy: 0.4481 - val_loss: 1.2544\n",
      "Epoch 11/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.6004 - loss: 1.0421 - val_accuracy: 0.4259 - val_loss: 1.2380\n",
      "Epoch 12/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.6338 - loss: 1.0180 - val_accuracy: 0.4407 - val_loss: 1.2492\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 13ms/step\n",
      "Epoch 1/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 34ms/step - accuracy: 0.2221 - loss: 1.8893 - val_accuracy: 0.2481 - val_loss: 1.7884\n",
      "Epoch 2/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.2985 - loss: 1.7181 - val_accuracy: 0.3259 - val_loss: 1.6316\n",
      "Epoch 3/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 10ms/step - accuracy: 0.3993 - loss: 1.5460 - val_accuracy: 0.3593 - val_loss: 1.4927\n",
      "Epoch 4/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.4400 - loss: 1.4173 - val_accuracy: 0.4185 - val_loss: 1.4052\n",
      "Epoch 5/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.4640 - loss: 1.3217 - val_accuracy: 0.4333 - val_loss: 1.3468\n",
      "Epoch 6/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 11ms/step - accuracy: 0.4730 - loss: 1.2977 - val_accuracy: 0.4444 - val_loss: 1.3130\n",
      "Epoch 7/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 10ms/step - accuracy: 0.5476 - loss: 1.1962 - val_accuracy: 0.4630 - val_loss: 1.2862\n",
      "Epoch 8/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.5403 - loss: 1.1843 - val_accuracy: 0.4593 - val_loss: 1.2687\n",
      "Epoch 9/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.5658 - loss: 1.1453 - val_accuracy: 0.4519 - val_loss: 1.2486\n",
      "Epoch 10/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.5731 - loss: 1.1395 - val_accuracy: 0.4630 - val_loss: 1.2349\n",
      "Epoch 11/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.5613 - loss: 1.0757 - val_accuracy: 0.4630 - val_loss: 1.2262\n",
      "Epoch 12/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.6190 - loss: 1.0689 - val_accuracy: 0.4778 - val_loss: 1.2138\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 16ms/step\n",
      "Epoch 1/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 32ms/step - accuracy: 0.2120 - loss: 1.9290 - val_accuracy: 0.3519 - val_loss: 1.8451\n",
      "Epoch 2/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.3613 - loss: 1.8040 - val_accuracy: 0.3444 - val_loss: 1.6776\n",
      "Epoch 3/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.3867 - loss: 1.6195 - val_accuracy: 0.3481 - val_loss: 1.5129\n",
      "Epoch 4/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.3996 - loss: 1.4853 - val_accuracy: 0.3963 - val_loss: 1.4062\n",
      "Epoch 5/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.4330 - loss: 1.3692 - val_accuracy: 0.3889 - val_loss: 1.3479\n",
      "Epoch 6/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.4600 - loss: 1.2733 - val_accuracy: 0.4296 - val_loss: 1.3026\n",
      "Epoch 7/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.4883 - loss: 1.2568 - val_accuracy: 0.4185 - val_loss: 1.2953\n",
      "Epoch 8/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.5229 - loss: 1.2082 - val_accuracy: 0.4333 - val_loss: 1.2678\n",
      "Epoch 9/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.5192 - loss: 1.1737 - val_accuracy: 0.4296 - val_loss: 1.2484\n",
      "Epoch 10/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.5688 - loss: 1.1083 - val_accuracy: 0.4370 - val_loss: 1.2418\n",
      "Epoch 11/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.6079 - loss: 1.0596 - val_accuracy: 0.4370 - val_loss: 1.2365\n",
      "Epoch 12/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.6054 - loss: 1.0411 - val_accuracy: 0.4481 - val_loss: 1.2350\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 11ms/step\n",
      "Dropping misspellings\n",
      "Preparing embeddings: drop\n",
      "Fitting model: drop\n",
      "Epoch 1/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 44ms/step - accuracy: 0.1995 - loss: 1.9377 - val_accuracy: 0.2963 - val_loss: 1.8924\n",
      "Epoch 2/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 11ms/step - accuracy: 0.4242 - loss: 1.8410 - val_accuracy: 0.3481 - val_loss: 1.7686\n",
      "Epoch 3/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 16ms/step - accuracy: 0.3638 - loss: 1.6877 - val_accuracy: 0.3296 - val_loss: 1.6222\n",
      "Epoch 4/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 13ms/step - accuracy: 0.4294 - loss: 1.5051 - val_accuracy: 0.3778 - val_loss: 1.4996\n",
      "Epoch 5/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 12ms/step - accuracy: 0.5001 - loss: 1.3346 - val_accuracy: 0.4074 - val_loss: 1.4281\n",
      "Epoch 6/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 16ms/step - accuracy: 0.5190 - loss: 1.2678 - val_accuracy: 0.4000 - val_loss: 1.3821\n",
      "Epoch 7/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 11ms/step - accuracy: 0.5761 - loss: 1.1664 - val_accuracy: 0.4667 - val_loss: 1.3508\n",
      "Epoch 8/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 13ms/step - accuracy: 0.5749 - loss: 1.1216 - val_accuracy: 0.3926 - val_loss: 1.3550\n",
      "Epoch 9/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 9ms/step - accuracy: 0.5892 - loss: 1.0438 - val_accuracy: 0.4222 - val_loss: 1.3283\n",
      "Epoch 10/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 13ms/step - accuracy: 0.6210 - loss: 1.0015 - val_accuracy: 0.4556 - val_loss: 1.3265\n",
      "Epoch 11/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 10ms/step - accuracy: 0.6179 - loss: 1.0132 - val_accuracy: 0.4556 - val_loss: 1.3013\n",
      "Epoch 12/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 11ms/step - accuracy: 0.6398 - loss: 0.9680 - val_accuracy: 0.4370 - val_loss: 1.3024\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 19ms/step\n",
      "Epoch 1/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 37ms/step - accuracy: 0.2467 - loss: 1.9302 - val_accuracy: 0.3741 - val_loss: 1.8526\n",
      "Epoch 2/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.3525 - loss: 1.8044 - val_accuracy: 0.3556 - val_loss: 1.6507\n",
      "Epoch 3/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3889 - loss: 1.5721 - val_accuracy: 0.4370 - val_loss: 1.4150\n",
      "Epoch 4/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.4729 - loss: 1.3786 - val_accuracy: 0.4259 - val_loss: 1.3143\n",
      "Epoch 5/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.4199 - loss: 1.3065 - val_accuracy: 0.4407 - val_loss: 1.2588\n",
      "Epoch 6/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 11ms/step - accuracy: 0.4867 - loss: 1.2660 - val_accuracy: 0.4444 - val_loss: 1.2635\n",
      "Epoch 7/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.4973 - loss: 1.2065 - val_accuracy: 0.4963 - val_loss: 1.2069\n",
      "Epoch 8/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 10ms/step - accuracy: 0.5775 - loss: 1.1256 - val_accuracy: 0.4889 - val_loss: 1.2050\n",
      "Epoch 9/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 10ms/step - accuracy: 0.5906 - loss: 1.0573 - val_accuracy: 0.4963 - val_loss: 1.1898\n",
      "Epoch 10/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.6247 - loss: 1.0305 - val_accuracy: 0.4741 - val_loss: 1.1990\n",
      "Epoch 11/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.6113 - loss: 1.0358 - val_accuracy: 0.4815 - val_loss: 1.1846\n",
      "Epoch 12/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.6392 - loss: 0.9753 - val_accuracy: 0.5037 - val_loss: 1.1616\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 13ms/step\n",
      "Epoch 1/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 37ms/step - accuracy: 0.1848 - loss: 1.9210 - val_accuracy: 0.1444 - val_loss: 1.8683\n",
      "Epoch 2/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.2068 - loss: 1.8159 - val_accuracy: 0.3037 - val_loss: 1.7462\n",
      "Epoch 3/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.3699 - loss: 1.6545 - val_accuracy: 0.3481 - val_loss: 1.5859\n",
      "Epoch 4/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.4648 - loss: 1.4497 - val_accuracy: 0.3963 - val_loss: 1.4394\n",
      "Epoch 5/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.4903 - loss: 1.3350 - val_accuracy: 0.3852 - val_loss: 1.3684\n",
      "Epoch 6/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.5351 - loss: 1.2567 - val_accuracy: 0.3889 - val_loss: 1.3339\n",
      "Epoch 7/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 14ms/step - accuracy: 0.5439 - loss: 1.1997 - val_accuracy: 0.4148 - val_loss: 1.3109\n",
      "Epoch 8/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.5729 - loss: 1.1304 - val_accuracy: 0.4370 - val_loss: 1.2855\n",
      "Epoch 9/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.5939 - loss: 1.0567 - val_accuracy: 0.4481 - val_loss: 1.2622\n",
      "Epoch 10/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 9ms/step - accuracy: 0.5966 - loss: 1.1258 - val_accuracy: 0.4519 - val_loss: 1.2520\n",
      "Epoch 11/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 10ms/step - accuracy: 0.6437 - loss: 1.0150 - val_accuracy: 0.4519 - val_loss: 1.2499\n",
      "Epoch 12/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.6160 - loss: 1.0206 - val_accuracy: 0.4630 - val_loss: 1.2386\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 12ms/step\n",
      "Correcting misspellings\n",
      "Preparing embeddings: correct\n",
      "Fitting model: correct\n",
      "Epoch 1/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 13ms/step - accuracy: 0.2016 - loss: 1.9108 - val_accuracy: 0.2926 - val_loss: 1.7807\n",
      "Epoch 2/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.3310 - loss: 1.7275 - val_accuracy: 0.3519 - val_loss: 1.5916\n",
      "Epoch 3/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 5ms/step - accuracy: 0.3775 - loss: 1.5353 - val_accuracy: 0.3889 - val_loss: 1.4498\n",
      "Epoch 4/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 5ms/step - accuracy: 0.4411 - loss: 1.3823 - val_accuracy: 0.4111 - val_loss: 1.3760\n",
      "Epoch 5/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.4920 - loss: 1.2922 - val_accuracy: 0.4296 - val_loss: 1.3375\n",
      "Epoch 6/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.4906 - loss: 1.2426 - val_accuracy: 0.4148 - val_loss: 1.3068\n",
      "Epoch 7/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5244 - loss: 1.2072 - val_accuracy: 0.4222 - val_loss: 1.2851\n",
      "Epoch 8/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 3ms/step - accuracy: 0.5442 - loss: 1.1393 - val_accuracy: 0.4333 - val_loss: 1.2711\n",
      "Epoch 9/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5541 - loss: 1.1279 - val_accuracy: 0.4593 - val_loss: 1.2606\n",
      "Epoch 10/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5353 - loss: 1.1036 - val_accuracy: 0.4778 - val_loss: 1.2501\n",
      "Epoch 11/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 3ms/step - accuracy: 0.6006 - loss: 1.0460 - val_accuracy: 0.4593 - val_loss: 1.2354\n",
      "Epoch 12/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5818 - loss: 1.0068 - val_accuracy: 0.4630 - val_loss: 1.2335\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 6ms/step \n",
      "Epoch 1/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 18ms/step - accuracy: 0.1962 - loss: 1.9264 - val_accuracy: 0.2630 - val_loss: 1.8447\n",
      "Epoch 2/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 5ms/step - accuracy: 0.3132 - loss: 1.8038 - val_accuracy: 0.3481 - val_loss: 1.7119\n",
      "Epoch 3/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.4316 - loss: 1.6263 - val_accuracy: 0.3481 - val_loss: 1.5610\n",
      "Epoch 4/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.4349 - loss: 1.4483 - val_accuracy: 0.3963 - val_loss: 1.4435\n",
      "Epoch 5/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.4799 - loss: 1.3241 - val_accuracy: 0.4000 - val_loss: 1.3900\n",
      "Epoch 6/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5234 - loss: 1.2529 - val_accuracy: 0.4037 - val_loss: 1.3766\n",
      "Epoch 7/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.5332 - loss: 1.1570 - val_accuracy: 0.4185 - val_loss: 1.3438\n",
      "Epoch 8/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 3ms/step - accuracy: 0.5579 - loss: 1.1295 - val_accuracy: 0.3963 - val_loss: 1.3505\n",
      "Epoch 9/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5647 - loss: 1.1093 - val_accuracy: 0.4444 - val_loss: 1.3174\n",
      "Epoch 10/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.6089 - loss: 1.0720 - val_accuracy: 0.4296 - val_loss: 1.3314\n",
      "Epoch 11/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.6266 - loss: 1.0183 - val_accuracy: 0.4370 - val_loss: 1.3136\n",
      "Epoch 12/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.6280 - loss: 0.9991 - val_accuracy: 0.4481 - val_loss: 1.3036\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 6ms/step \n",
      "Epoch 1/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 13ms/step - accuracy: 0.1732 - loss: 1.9247 - val_accuracy: 0.3296 - val_loss: 1.8531\n",
      "Epoch 2/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.3547 - loss: 1.8043 - val_accuracy: 0.3370 - val_loss: 1.7121\n",
      "Epoch 3/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.3857 - loss: 1.6493 - val_accuracy: 0.4185 - val_loss: 1.5494\n",
      "Epoch 4/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.4398 - loss: 1.4796 - val_accuracy: 0.4519 - val_loss: 1.4253\n",
      "Epoch 5/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.4665 - loss: 1.3476 - val_accuracy: 0.4222 - val_loss: 1.3558\n",
      "Epoch 6/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5249 - loss: 1.2603 - val_accuracy: 0.4074 - val_loss: 1.3127\n",
      "Epoch 7/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5250 - loss: 1.2099 - val_accuracy: 0.4407 - val_loss: 1.2862\n",
      "Epoch 8/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5167 - loss: 1.2057 - val_accuracy: 0.4370 - val_loss: 1.2653\n",
      "Epoch 9/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5712 - loss: 1.1072 - val_accuracy: 0.4111 - val_loss: 1.2534\n",
      "Epoch 10/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 3ms/step - accuracy: 0.5609 - loss: 1.1061 - val_accuracy: 0.4556 - val_loss: 1.2367\n",
      "Epoch 11/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 4ms/step - accuracy: 0.5772 - loss: 1.0438 - val_accuracy: 0.4556 - val_loss: 1.2315\n",
      "Epoch 12/12\n",
      "\u001b[1m20/20\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 3ms/step - accuracy: 0.5974 - loss: 1.0120 - val_accuracy: 0.4370 - val_loss: 1.2283\n",
      "\u001b[1m9/9\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 5ms/step \n"
     ]
    }
   ],
   "source": [
    "import time as t\n",
    "\n",
    "# creates an accuracy and time dataframe to store results\n",
    "accuracy_df = pd.DataFrame()\n",
    "time_df = pd.DataFrame()\n",
    "\n",
    "# preprocesses the texts\n",
    "train['full_text'] = [preprocess_text(x) for x in train['full_text']]\n",
    "\n",
    "# converts scores to a numpy array for use in the model\n",
    "scores = np.array(train['score'])\n",
    "\n",
    "# limits each essay to the first 300 words\n",
    "texts = [' '.join(x.split()[0:300]) for x in train['full_text']]\n",
    "\n",
    "# defines the various misspellings approaches\n",
    "misspelling_approaches = ['nothing', 'drop', 'correct']\n",
    "\n",
    "# loops through each approach\n",
    "for approach in misspelling_approaches:\n",
    "\n",
    "    # records a start time for later analysis\n",
    "    start_time = t.time()\n",
    "\n",
    "    # simply obtains the embeddings\n",
    "    if approach == 'nothing':\n",
    "        print('Preparing embeddings: ' + approach)\n",
    "        embeddings = encode_sentences(texts)\n",
    "\n",
    "    # drops misspellings and obtains the embeddings\n",
    "    elif approach == 'drop':\n",
    "        print('Dropping misspellings')\n",
    "        dropped_misspellings = [misspellings(x) for x in texts]\n",
    "        print('Preparing embeddings: ' + approach)\n",
    "        embeddings = encode_sentences(dropped_misspellings)\n",
    "\n",
    "    # corrects misspellings and obtains the embeddings\n",
    "    elif approach == 'correct':\n",
    "        print('Correcting misspellings')\n",
    "        corrected_misspellings = [misspellings(x, drop=False) for x in texts]\n",
    "        print('Preparing embeddings: ' + approach)\n",
    "        embeddings = encode_sentences(corrected_misspellings)\n",
    "\n",
    "    # records an end time for analysis\n",
    "    end_time = t.time()\n",
    "\n",
    "    # adds the total time taken to the time_df for analysis\n",
    "    time_df.loc[0, approach] = round(end_time-start_time, 2)\n",
    "\n",
    "    # fits the model 3 times and adds the average accuracy to the accuracy_df\n",
    "    print('Fitting model: ' + approach)\n",
    "    accuracy_df = fit_model(embeddings, scores, accuracy_df, approach)\n",
    "\n",
    "    # reassigns the embeddings object to nother to conserve memory\n",
    "    embeddings = ''"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7789901b-f06a-41fa-919b-a26abfa5dcab",
   "metadata": {},
   "source": [
    "## Visualization"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "34b68593-2dd6-45ab-8f8b-98193a79119e",
   "metadata": {},
   "source": [
    "We can see that the time it takes to correct the essay misspellings is significantly longer and does not contribute to greater accuracy. Dropping the misspellings appears to yield slightly better results and so this should be the preffered method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "02bcc28b-e04b-469d-89c0-4acd59d397a4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1200x600 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "# a single results dataframe is created \n",
    "results_df = pd.DataFrame()\n",
    "results_df['time'] = time_df.transpose()\n",
    "results_df['accuracy'] = accuracy_df.transpose()\n",
    "\n",
    "# a subplots object is defined\n",
    "fig, ax = plt.subplots(1, 2, figsize=(12, 6))\n",
    "\n",
    "# different ylabels are defined \n",
    "ylabels = ['Seconds', 'Cohen Kappa Score']\n",
    "\n",
    "# a lineplpot is created for each result type\n",
    "for i, col in enumerate(results_df.columns):\n",
    "    sns.barplot(x=results_df.index, y=results_df.iloc[:,i], data=results_df, ax=ax[i]);\n",
    "    ax[i].set_title(results_df.iloc[:,i].name.title(), size=16);\n",
    "    ax[i].set_ylabel(ylabels[i]);\n",
    "    ax[i].set_xlabel('');\n",
    "    \n",
    "fig.suptitle('Misspellings Approaches', size=30, y=1.025)\n",
    "plt.tight_layout()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.11.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}