{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Manual Mapping of Ordinal Features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "from itertools import chain\n",
    "from pyspark.sql import functions as F\n",
    "from pyspark.sql import SparkSession"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "spark = SparkSession \\\n",
    "    .builder \\\n",
    "    .appName(\"Pysparkexample\") \\\n",
    "    .config(\"spark.some.config.option\", \"some-value\") \\\n",
    "    .getOrCreate()\n",
    "sc = spark.sparkContext"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Assume having columns in a dataframe with ordinal values which you want to map to integers (or doubles)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+------+\n",
      "|feat1| feat2|\n",
      "+-----+------+\n",
      "|   HI| LARGE|\n",
      "|  MID|MEDIUM|\n",
      "|   LO| SMALL|\n",
      "+-----+------+\n",
      "only showing top 3 rows\n",
      "\n"
     ]
    }
   ],
   "source": [
    "def fresh_df(N=100000, seed=None):\n",
    "    np.random.seed(seed)\n",
    "    feat1 = np.random.choice([\"HI\", \"LO\", \"MID\"], size=N)\n",
    "    feat2 = np.random.choice([\"SMALL\", \"MEDIUM\", \"LARGE\"], size=N)\n",
    "\n",
    "    pdf = pd.DataFrame({\n",
    "        \"feat1\": feat1,\n",
    "        \"feat2\": feat2\n",
    "    })\n",
    "    return spark.createDataFrame(pdf)\n",
    "\n",
    "fresh_df(N=100).show(3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So, you build manually the mappings:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "feat1_dict = {\"HI\": 1, \"MID\": 2, \"LO\": 3}\n",
    "feat2_dict = {\"SMALL\": 0, \"MEDIUM\": 1, \"LARGE\": 2}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And turn each of them into a dataframe of its own:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "feat1_map_df = spark.createDataFrame(pd.DataFrame([(k,v) for k, v in feat1_dict.items()], columns=[\"feat\", \"label\"]))\n",
    "feat2_map_df = spark.createDataFrame(pd.DataFrame([(k,v) for k, v in feat2_dict.items()], columns=[\"feat\", \"label\"]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Collect the mapping dataframes in a dictionary:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "mappings = {\n",
    "    \"feat1\": feat1_map_df,\n",
    "    \"feat2\": feat2_map_df\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+-----+------+\n",
      "|feat1| feat2|\n",
      "+-----+------+\n",
      "|  MID| LARGE|\n",
      "|   HI| LARGE|\n",
      "|  MID| SMALL|\n",
      "|  MID| LARGE|\n",
      "|   HI|MEDIUM|\n",
      "|   HI| SMALL|\n",
      "|  MID|MEDIUM|\n",
      "|   LO|MEDIUM|\n",
      "|  MID|MEDIUM|\n",
      "|  MID|MEDIUM|\n",
      "+-----+------+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "df = fresh_df(N=10, seed=42)\n",
    "df.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Using `join`s you can now replace the values using the mappings:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+----+-----+\n",
      "|feat|label|\n",
      "+----+-----+\n",
      "|  HI|    1|\n",
      "| MID|    2|\n",
      "|  LO|    3|\n",
      "+----+-----+\n",
      "\n"
     ]
    }
   ],
   "source": [
    "mappings[\"feat1\"].show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def label_column(df, mapping, feat_name):\n",
    "    return df.join(F.broadcast(mapping), df[feat_name] == mapping.feat)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here's an example:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = fresh_df(N=1000000, seed=42)\n",
    "cols = df.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "250 ms ± 88.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit\n",
    "for col in cols:\n",
    "    df2 = label_column(df, mappings[col], col).drop('feat')\n",
    "    df2 = df2.withColumnRenamed(\"label\", col + \"_mapped\")\n",
    "df2.cache().count()"
   ]
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}