ML.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "e07f0e79",
   "metadata": {},
   "source": [
    "# ML Modeling\n",
    "\n",
    "This Jupyter Notebook demonstrates the process of training three different machine learning classifiers - logistic regression, random forest, and naive Bayes classifier. The aim is to compare the performance of these classifiers based on their ROC curves and confusion matrices. The dataset used for this task is clean_data, which was previously cleaned and processed in the EDA.ipynb notebook.\n",
    "\n",
    "## Data Loading\n",
    "\n",
    "The first step in this process is to load the clean_data from the EDA.ipynb notebook. This dataset should have been saved in a file format such as CSV, so it can be easily loaded into the current notebook using a data loading function or library.\n",
    "\n",
    "## Data Preparation\n",
    "\n",
    "Before training the models, the data needs to be prepared by creating a VectorAssembler object. This object will take in all the features and merge them into a single vector. This is required by some of the machine learning models, including logistic regression and naive Bayes classifier.\n",
    "\n",
    "After creating the VectorAssembler object, the data is split into training and test sets. The training data will be used to train the models, while the test data will be used to evaluate their performance.\n",
    "\n",
    "## Model Training\n",
    "\n",
    "With the data prepared, the next step is to train the three machine learning models - logistic regression, random forest, and naive Bayes classifier. Each of these models will be trained using the training data set.\n",
    "\n",
    "## Model Comparison\n",
    "\n",
    "Once the models have been trained, their ROC curves are compared. The ROC curve is a graphical representation of the performance of a classifier. A good classifier will have a curve that is close to the upper left corner of the plot. This indicates that the classifier has a high true positive rate and a low false positive rate.\n",
    "\n",
    "After comparing the ROC curves, the confusion matrix is calculated for each model. The confusion matrix is a table that summarizes the performance of a classifier. It shows the number of true positives, false positives, true negatives, and false negatives.\n",
    "\n",
    "## Conclusion\n",
    "\n",
    "In this Jupyter Notebook, we have demonstrated the process of training three different machine learning classifiers and comparing their performance based on their ROC curves and confusion matrices. The logistic regression, random forest, and naive Bayes classifier models were trained using the clean_data dataset that was loaded from the EDA.ipynb notebook. The performance of these models was compared using their ROC curves and confusion matrices, and the results were evaluated to determine which model performed best for the given dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "16fbcca2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Starter code\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from pyspark.sql import SparkSession\n",
    "from pyspark.sql.functions import col, count, udf\n",
    "from pyspark.sql.functions import max as fmax, min as fmin\n",
    "from pyspark.sql.types import IntegerType, FloatType"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "04b67abd",
   "metadata": {},
   "outputs": [],
   "source": [
    "from pyspark.ml.feature import VectorAssembler\n",
    "from pyspark.ml.classification import LogisticRegression, RandomForestClassifier, NaiveBayes\n",
    "from pyspark.mllib.evaluation import MulticlassMetrics\n",
    "\n",
    "from pyspark.ml.evaluation import BinaryClassificationEvaluator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "9f970488",
   "metadata": {},
   "outputs": [],
   "source": [
    "from CurveMetrics import CurveMetrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "cfd8477a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "23/03/11 16:22:03 WARN Utils: Your hostname, pop-os resolves to a loopback address: 127.0.1.1; using 192.168.248.232 instead (on interface wlp110s0)\n",
      "23/03/11 16:22:03 WARN Utils: Set SPARK_LOCAL_IP if you need to bind to another address\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Setting default log level to \"WARN\".\n",
      "To adjust logging level use sc.setLogLevel(newLevel). For SparkR, use setLogLevel(newLevel).\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "23/03/11 16:22:03 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable\n",
      "23/03/11 16:22:04 WARN Utils: Service 'SparkUI' could not bind on port 4040. Attempting port 4041.\n",
      "23/03/11 16:22:04 WARN Utils: Service 'SparkUI' could not bind on port 4041. Attempting port 4042.\n",
      "23/03/11 16:22:04 WARN Utils: Service 'SparkUI' could not bind on port 4042. Attempting port 4043.\n"
     ]
    }
   ],
   "source": [
    "# Create spark session\n",
    "spark = SparkSession \\\n",
    "    .builder \\\n",
    "    .appName(\"Sparkify\") \\\n",
    "    .getOrCreate()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "3236272d",
   "metadata": {},
   "outputs": [],
   "source": [
    "spark.conf.set(\"spark.sql.repl.eagerEval.enabled\", True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "51d48b4b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Read in full sparkify dataset\n",
    "# Full dataset\n",
    "# event_data = \"s3n://udacity-dsnd/sparkify/sparkify_event_data.json\"\n",
    "event_data = \"clean_data/part-00000-641e06ab-b8b2-4a90-b032-f1f5a8577ef8-c000.csv\"\n",
    "# Mini dataset\n",
    "#event_data = \"mini_sparkify_event_data.json\"\n",
    "df = spark.read.csv(event_data, header=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "14138d28",
   "metadata": {},
   "outputs": [],
   "source": [
    "features = ['n_pages', \n",
    "            'thumbs_down', \n",
    "            'home', \n",
    "            'downgrade', \n",
    "            'roll_advert', \n",
    "            'about', \n",
    "            'add_playlist', \n",
    "            'nextsong', \n",
    "            'thumbs_up', \n",
    "            'error', \n",
    "            'submit_upgrade', \n",
    "            'total_length']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "2baa11f9",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Cast data to integers\n",
    "for i in range(len(features)):\n",
    "    df = df.withColumn(features[i] , df[features[i]].cast(IntegerType()))\n",
    "    \n",
    "df = df.withColumn('label' , df['label'].cast(IntegerType()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "efe80c7f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "root\n",
      " |-- userId: string (nullable = true)\n",
      " |-- n_pages: integer (nullable = true)\n",
      " |-- thumbs_down: integer (nullable = true)\n",
      " |-- home: integer (nullable = true)\n",
      " |-- downgrade: integer (nullable = true)\n",
      " |-- roll_advert: integer (nullable = true)\n",
      " |-- about: integer (nullable = true)\n",
      " |-- add_playlist: integer (nullable = true)\n",
      " |-- nextsong: integer (nullable = true)\n",
      " |-- thumbs_up: integer (nullable = true)\n",
      " |-- error: integer (nullable = true)\n",
      " |-- submit_upgrade: integer (nullable = true)\n",
      " |-- total_length: integer (nullable = true)\n",
      " |-- label: integer (nullable = true)\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df.printSchema()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "d688193c",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Vector Assembler\n",
    "assembler = VectorAssembler(inputCols=features, outputCol='features')\n",
    "df = assembler.transform(df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "7fe2420c",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = df.select(col('label'), col('features'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "b1427b51",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Split data into training and test data\n",
    "train, test = data.randomSplit([0.9, 0.1], seed=42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "d94b754c",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "                                                                                \r"
     ]
    }
   ],
   "source": [
    "# Logistic regression\n",
    "lr = LogisticRegression()\n",
    "lrModel = lr.fit(train)\n",
    "\n",
    "# Random Forest\n",
    "rf = RandomForestClassifier()\n",
    "rfModel = rf.fit(train)\n",
    "\n",
    "# Naive Bayes\n",
    "nb = NaiveBayes()\n",
    "nbModel = nb.fit(train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "92240899",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border='1'>\n",
       "<tr><th>label</th><th>features</th><th>rawPrediction</th><th>probability</th><th>prediction</th></tr>\n",
       "<tr><td>0</td><td>(12,[0,1,2,4,7,11...</td><td>[-221.59303798145...</td><td>[0.75843309110217...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,1,2,4,7,11...</td><td>[-272.97534601314...</td><td>[0.76124500185073...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,1,2,7,8,11...</td><td>[-386.94495601567...</td><td>[0.76645236484021...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,1,2,7,11],...</td><td>[-108.00059919369...</td><td>[0.76280601383043...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,1,4,7,8,11...</td><td>[-631.32961601553...</td><td>[0.75451882498922...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2],[1.0,1.0])</td><td>[-14.219797442970...</td><td>[0.77002187679423...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,3,6,7,11...</td><td>[-536.70583081301...</td><td>[0.6955309713495,...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,3,6,7,11...</td><td>[-603.46088708032...</td><td>[0.69272123989684...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,3,7,8,11...</td><td>[-807.22550490496...</td><td>[0.75959600754192...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,6,7,11...</td><td>[-276.36912178957...</td><td>[0.78086188922126...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,6,7,11...</td><td>[-368.90597052055...</td><td>[0.76594084501109...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,6,7,11...</td><td>[-523.18213175364...</td><td>[0.76196970117589...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,6,7,11...</td><td>[-608.63783830705...</td><td>[0.75697606756871...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,6,7,11...</td><td>[-837.79485826460...</td><td>[0.75992950826814...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,7,8,11...</td><td>[-147.85691951063...</td><td>[0.78185858610241...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,7,8,11...</td><td>[-160.21245471383...</td><td>[0.78191693797300...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,7,8,11...</td><td>[-275.32552834513...</td><td>[0.80063248191031...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,7,8,11...</td><td>[-312.70636591469...</td><td>[0.77705790567020...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,7,8,11...</td><td>[-334.23296958763...</td><td>[0.77595673385604...</td><td>0.0</td></tr>\n",
       "<tr><td>0</td><td>(12,[0,2,4,7,8,11...</td><td>[-389.03547371851...</td><td>[0.79848284475103...</td><td>0.0</td></tr>\n",
       "</table>\n",
       "only showing top 20 rows\n"
      ],
      "text/plain": [
       "+-----+--------------------+--------------------+--------------------+----------+\n",
       "|label|            features|       rawPrediction|         probability|prediction|\n",
       "+-----+--------------------+--------------------+--------------------+----------+\n",
       "|    0|(12,[0,1,2,4,7,11...|[-221.59303798145...|[0.75843309110217...|       0.0|\n",
       "|    0|(12,[0,1,2,4,7,11...|[-272.97534601314...|[0.76124500185073...|       0.0|\n",
       "|    0|(12,[0,1,2,7,8,11...|[-386.94495601567...|[0.76645236484021...|       0.0|\n",
       "|    0|(12,[0,1,2,7,11],...|[-108.00059919369...|[0.76280601383043...|       0.0|\n",
       "|    0|(12,[0,1,4,7,8,11...|[-631.32961601553...|[0.75451882498922...|       0.0|\n",
       "|    0|(12,[0,2],[1.0,1.0])|[-14.219797442970...|[0.77002187679423...|       0.0|\n",
       "|    0|(12,[0,2,3,6,7,11...|[-536.70583081301...|[0.6955309713495,...|       0.0|\n",
       "|    0|(12,[0,2,3,6,7,11...|[-603.46088708032...|[0.69272123989684...|       0.0|\n",
       "|    0|(12,[0,2,3,7,8,11...|[-807.22550490496...|[0.75959600754192...|       0.0|\n",
       "|    0|(12,[0,2,4,6,7,11...|[-276.36912178957...|[0.78086188922126...|       0.0|\n",
       "|    0|(12,[0,2,4,6,7,11...|[-368.90597052055...|[0.76594084501109...|       0.0|\n",
       "|    0|(12,[0,2,4,6,7,11...|[-523.18213175364...|[0.76196970117589...|       0.0|\n",
       "|    0|(12,[0,2,4,6,7,11...|[-608.63783830705...|[0.75697606756871...|       0.0|\n",
       "|    0|(12,[0,2,4,6,7,11...|[-837.79485826460...|[0.75992950826814...|       0.0|\n",
       "|    0|(12,[0,2,4,7,8,11...|[-147.85691951063...|[0.78185858610241...|       0.0|\n",
       "|    0|(12,[0,2,4,7,8,11...|[-160.21245471383...|[0.78191693797300...|       0.0|\n",
       "|    0|(12,[0,2,4,7,8,11...|[-275.32552834513...|[0.80063248191031...|       0.0|\n",
       "|    0|(12,[0,2,4,7,8,11...|[-312.70636591469...|[0.77705790567020...|       0.0|\n",
       "|    0|(12,[0,2,4,7,8,11...|[-334.23296958763...|[0.77595673385604...|       0.0|\n",
       "|    0|(12,[0,2,4,7,8,11...|[-389.03547371851...|[0.79848284475103...|       0.0|\n",
       "+-----+--------------------+--------------------+--------------------+----------+\n",
       "only showing top 20 rows"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "nbModel.transform(test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "de71b919",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/manuel/anaconda3/envs/data-analysis/lib/python3.8/site-packages/pyspark/sql/context.py:157: FutureWarning: Deprecated in 3.0.0. Use SparkSession.builder.getOrCreate() instead.\n",
      "  warnings.warn(\n",
      "                                                                                \r"
     ]
    }
   ],
   "source": [
    "# ROC Curve\n",
    "def get_roc(model, data):\n",
    "    \"\"\"\n",
    "    Auxiliary function to calculate roc values of a model.\n",
    "    \"\"\"\n",
    "    preds = model.transform(data).select('label','probability').rdd.map(lambda row: (float(row['probability'][1]), float(row['label'])))\n",
    "    points = CurveMetrics(preds).get_curve('roc')\n",
    "\n",
    "    fpr = [x[0] for x in points]\n",
    "    tpr = [x[1] for x in points]\n",
    "\n",
    "    return fpr, tpr\n",
    "\n",
    "roc_lr = lrModel.summary.roc.toPandas()\n",
    "roc_rf = rfModel.summary.roc.toPandas()\n",
    "fpr_nb, tpr_nb = get_roc(nbModel, train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "648d471d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(roc_lr['FPR'].values, roc_lr['TPR'].values, label='logistic regression')\n",
    "plt.plot(roc_rf['FPR'].values, roc_rf['TPR'].values, label='random forest')\n",
    "plt.plot(fpr_nb, tpr_nb, label='naive bayes')\n",
    "plt.title('ROC Curve')\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.grid('on')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "e727d7e0",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Prediction\n",
    "pred_lr = lrModel.transform(test)\n",
    "pred_rf = rfModel.transform(test)\n",
    "pred_nb = nbModel.transform(test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "5e4aa08b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Confusion Matrix\n",
    "def get_confusion_matrix(pred):\n",
    "    #important: need to cast to float type, and order by prediction, else it won't work\n",
    "    preds_and_labels = pred.select(['prediction','label'])\\\n",
    "                        .withColumn('label', col('label').cast(FloatType()))\\\n",
    "                        .orderBy('prediction')\n",
    "\n",
    "    #select only prediction and label columns\n",
    "    preds_and_labels = preds_and_labels.select(['prediction','label'])\n",
    "\n",
    "    metrics = MulticlassMetrics(preds_and_labels.rdd.map(tuple))\n",
    "    \n",
    "    return metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "c5dea971",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/manuel/anaconda3/envs/data-analysis/lib/python3.8/site-packages/pyspark/sql/context.py:157: FutureWarning: Deprecated in 3.0.0. Use SparkSession.builder.getOrCreate() instead.\n",
      "  warnings.warn(\n"
     ]
    }
   ],
   "source": [
    "metrics_lr = get_confusion_matrix(pred_lr)\n",
    "metrics_rf = get_confusion_matrix(pred_rf)\n",
    "metrics_nb = get_confusion_matrix(pred_nb)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "ae5f6163",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Logistic Regression\n",
      "[[1674   47]\n",
      " [ 218  282]]\n",
      "\n",
      " Random Forest\n",
      "[[1721    0]\n",
      " [ 261  239]]\n",
      "\n",
      " Naive Bayes\n",
      "[[1533  188]\n",
      " [ 351  149]]\n"
     ]
    }
   ],
   "source": [
    "print('Logistic Regression')\n",
    "print(metrics_lr.confusionMatrix().toArray().astype('int'))\n",
    "\n",
    "print('\\n Random Forest')\n",
    "print(metrics_rf.confusionMatrix().toArray().astype('int'))\n",
    "\n",
    "print('\\n Naive Bayes')\n",
    "print(metrics_nb.confusionMatrix().toArray().astype('int'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "087f0451",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.8.15"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}