Testing windowing as a substitute for straight aggregation

windowing-test-a.ipynb

windowing-test-b.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Windowing as a substitute for straight aggregation\n",
    "\n",
    "This notebook will simulate the execution of the following query to the [aggregate operation](https://pathling.csiro.au/docs/aggregate.html) in [Pathling](https://pathling.csiro.au):\n",
    "\n",
    "```\n",
    "subjectResource: Patient\n",
    "aggregation: count()\n",
    "grouping: name.prefix contains 'Mrs.'\n",
    "grouping: name.given contains 'Karina848'\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pyspark.sql import SparkSession\n",
    "from pyspark.sql.functions import *\n",
    "from pyspark.sql import Window\n",
    "import time\n",
    "\n",
    "spark = SparkSession.builder \\\n",
    "  .master('local[*]') \\\n",
    "  .config(\"spark.driver.memory\", \"2g\") \\\n",
    "  .config(\"spark.executor.memory\", \"6g\") \\\n",
    "  .config(\"spark.sql.autoBroadcastJoinThreshold\", \"-1\") \\\n",
    "  .config(\"spark.sql.adaptive.enabled\", \"true\") \\\n",
    "  .config(\"spark.sql.shuffle.service.enabled\", \"true\") \\\n",
    "  .config(\"spark.sql.shuffle.partitions\", \"2\") \\\n",
    "  .config(\"spark.scheduler.mode\", \"FAIR\") \\\n",
    "  .config(\"spark.dynamicAllocation.enabled\", \"true\") \\\n",
    "  .getOrCreate()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "start_time = time.time()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Grouping 1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First we get the `Patient` dataset from file."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "patient = spark.read.parquet(\"file:///usr/share/data/synthea/1k_20190829/parquet/Patient.parquet\")\n",
    "patient_id = patient[\"id\"]\n",
    "patient_columns = [ patient[col] for col in patient.columns ]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Traverse to the `name` element within `Patient`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "grouping1 = patient.withColumn(\"a\", patient[\"id\"])\n",
    "grouping1 = grouping1.withColumn(\"b\", explode_outer(patient[\"name\"]))\n",
    "grouping1_id = grouping1[\"a\"]\n",
    "grouping1_value = grouping1[\"b\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Traverse to the `prefix` element within `name`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "grouping1 = grouping1.withColumn(\"c\", explode_outer(grouping1_value.getField(\"prefix\")))\n",
    "grouping1_value = grouping1[\"c\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create a dataset to represent the literal value `Mrs.`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "mrs_literal = patient.select(patient_id.alias(\"d\"), lit(\"Mrs.\").alias(\"e\"))\n",
    "mrs_literal_id = mrs_literal[\"d\"]\n",
    "mrs_literal_value = mrs_literal[\"e\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create a column which calculates the result of a `contains` operation, with `prefix` on the left and `Mrs.` on the right."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pyspark.sql.functions import max as spark_max\n",
    "\n",
    "grouping1 = grouping1.join(mrs_literal, grouping1_id == mrs_literal_id, \"left_outer\")\n",
    "equality_with_null_checks = when(grouping1_value.isNull(), lit(None)) \\\n",
    "    .when(mrs_literal_value.isNull(), lit(None)) \\\n",
    "    .otherwise(grouping1_value == mrs_literal_value)\n",
    "agg_column = spark_max(equality_with_null_checks)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Use __aggregation__ to get the result of the `contains` operation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "grouping1 = grouping1.groupBy(patient_columns).agg(agg_column)\n",
    "grouping1_id = grouping1[grouping1.columns[0]]\n",
    "grouping1_value = grouping1[grouping1.columns[-1]]\n",
    "patient_id = grouping1_id\n",
    "patient_columns = [ grouping1[col] for col in grouping1.columns[0:(len(patient_columns) - 1)] ]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Grouping 2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Traverse to the `name` element within `Patient`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "grouping2 = grouping1.withColumn(\"g\", grouping1_id)\n",
    "grouping2 = grouping2.withColumn(\"h\", explode_outer(patient[\"name\"]))\n",
    "grouping2_id = grouping2[\"g\"]\n",
    "grouping2_value = grouping2[\"h\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Traverse to the `given` element within `name`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "grouping2 = grouping2.withColumn(\"i\", explode_outer(grouping2_value.getField(\"given\")))\n",
    "grouping2_value = grouping2[\"i\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create a dataset to represent the literal value `Karina848`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "karina_literal = patient.select(patient_id.alias(\"j\"), lit(\"Karina848\").alias(\"k\"))\n",
    "karina_literal_id = karina_literal[\"j\"]\n",
    "karina_literal_value = karina_literal[\"k\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create a column which calculates the result of a `contains` operation, with `given` on the left and `Karina848` on the right."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pyspark.sql.functions import max as spark_max\n",
    "\n",
    "grouping2 = grouping2.join(karina_literal, grouping2_id == karina_literal_id, \"left_outer\")\n",
    "equality_with_null_checks = when(grouping2_value.isNull(), lit(None)) \\\n",
    "    .when(karina_literal_value.isNull(), lit(None)) \\\n",
    "    .otherwise(grouping2_value == karina_literal_value)\n",
    "agg_column = spark_max(equality_with_null_checks)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Use __aggregation__ to get the result of the `contains` operation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "group_by = patient_columns.append(grouping1_value)\n",
    "grouping2 = grouping2.groupBy(patient_columns).agg(agg_column)\n",
    "grouping2_id = grouping2[grouping2.columns[0]]\n",
    "grouping2_value = grouping2[grouping2.columns[-1]]\n",
    "patient_id = grouping2_id\n",
    "patient_columns = [ grouping2[col] for col in grouping2.columns[0:(len(patient_columns) - 1)] ]\n",
    "grouping1_value = grouping2[grouping2.columns[len(patient_columns)]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Aggregation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create a column which counts the number of patients within each grouping."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "grouping_columns = [grouping1_value, grouping2_value]\n",
    "count_column = count(patient_id)\n",
    "null_aware_count = when(count_column.isNull(), 0) \\\n",
    "    .otherwise(count_column)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create a dataset to represent the aggregation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "aggregation = grouping2.groupBy(grouping_columns).agg(null_aware_count)\n",
    "aggregation_value = aggregation[aggregation.columns[-1]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Final result"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Take the distinct values from the selection of each of the grouping columns, and the aggregation column."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "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>grouping1</th>\n",
       "      <th>grouping2</th>\n",
       "      <th>count</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>897</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>474</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>None</td>\n",
       "      <td>False</td>\n",
       "      <td>234</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  grouping1  grouping2  count\n",
       "0     False      False    897\n",
       "1      True      False    474\n",
       "2      None      False    234\n",
       "3      True       True      1"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "result = aggregation.select(grouping1_value.alias(\"grouping1\"),\n",
    "                            grouping2_value.alias(\"grouping2\"),\n",
    "                            aggregation_value.alias(\"count\")).distinct()\n",
    "result.toPandas()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "13.59262752532959"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "time.time() - start_time"
   ]
  }
 ],
 "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}