arrow_performance_comparison_notebook.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Apache Arrow"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1 Compare performance of csv, Parquet and Arrow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pyarrow.parquet as pq\n",
    "import pyarrow as pa\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import os\n",
    "import psutil"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1.1 Load and prepare data "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Read Palmer Station Penguin dataset from GitHub\n",
    "df = pd.read_csv(\"https://raw.githubusercontent.com/allisonhorst/\"\n",
    "                 \"palmerpenguins/47a3476d2147080e7ceccef4cf70105c808f2cbf/\"\n",
    "                 \"data-raw/penguins_raw.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Increase dataset to 1m rows and reset index\n",
    "df = df.sample(1_000_000, replace=True).reset_index(drop=True)\n",
    "\n",
    "\n",
    "# Update sample number (0 to 999'999)\n",
    "df[\"Sample Number\"] = df.index\n",
    "# Add some random variation to numeric columns\n",
    "df[[\"Culmen Length (mm)\", \"Culmen Depth (mm)\", \n",
    "    \"Flipper Length (mm)\", \"Body Mass (g)\"]] = df[[\"Culmen Length (mm)\", \"Culmen Depth (mm)\", \n",
    "                                                   \"Flipper Length (mm)\", \"Body Mass (g)\"]] \\\n",
    "                                               + np.random.rand(df.shape[0], 4)\n",
    "\n",
    "# Create dataframe where missing numeric values are filled with zero\n",
    "df_nonan = df.copy()\n",
    "df_nonan[[\"Culmen Length (mm)\", \"Culmen Depth (mm)\", \n",
    "          \"Flipper Length (mm)\", \"Body Mass (g)\"]] = df[[\"Culmen Length (mm)\", \"Culmen Depth (mm)\", \n",
    "                                                         \"Flipper Length (mm)\", \"Body Mass (g)\"]].fillna(0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1.2 Write to disk "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\pyarrow\\pandas_compat.py:383: FutureWarning: RangeIndex._start is deprecated and will be removed in a future version. Use RangeIndex.start instead\n",
      "  'start': level._start,\n",
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\pyarrow\\pandas_compat.py:384: FutureWarning: RangeIndex._stop is deprecated and will be removed in a future version. Use RangeIndex.stop instead\n",
      "  'stop': level._stop,\n",
      "C:\\ProgramData\\Anaconda3\\lib\\site-packages\\pyarrow\\pandas_compat.py:385: FutureWarning: RangeIndex._step is deprecated and will be removed in a future version. Use RangeIndex.step instead\n",
      "  'step': level._step\n"
     ]
    }
   ],
   "source": [
    "# Write to csv\n",
    "df.to_csv(\"penguin-dataset.csv\")\n",
    "\n",
    "# Write to parquet\n",
    "df.to_parquet(\"penguin-dataset.parquet\")\n",
    "\n",
    "# Write to Arrow\n",
    "# Convert from pandas to Arrow\n",
    "table = pa.Table.from_pandas(df)\n",
    "# Write out to file\n",
    "with pa.OSFile('penguin-dataset.arrow', 'wb') as sink:\n",
    "    with pa.RecordBatchFileWriter(sink, table.schema) as writer:\n",
    "        writer.write_table(table)\n",
    "\n",
    "# Convert from no-NaN pandas to Arrow\n",
    "table_nonan = pa.Table.from_pandas(df_nonan)\n",
    "# Write out to file\n",
    "with pa.OSFile('penguin-dataset-nonan.arrow', 'wb') as sink:\n",
    "    with pa.RecordBatchFileWriter(sink, table_nonan.schema) as writer:\n",
    "        writer.write_table(table_nonan)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1.3 Reading time - calculate average of numeric column"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.3.1 Read csv and calculate mean"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3.63 s ± 159 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit\n",
    "pd.read_csv(\"penguin-dataset.csv\")[\"Flipper Length (mm)\"].mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.3.2 Read parquet and calculate mean"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "56.0 ms ± 8.72 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit\n",
    "pd.read_parquet(\"penguin-dataset.parquet\", columns=[\"Flipper Length (mm)\"]).mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.3.3 Read Arrow using file API"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "152 ms ± 17.2 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit\n",
    "with pa.OSFile('penguin-dataset.arrow', 'rb') as source:\n",
    "    table = pa.ipc.open_file(source).read_all().column(\"Flipper Length (mm)\")\n",
    "result = table.to_pandas().mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.3.4 Read Arrow with memory-mapped API with missing values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "17.9 ms ± 2.4 ms per loop (mean ± std. dev. of 7 runs, 100 loops each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit\n",
    "source = pa.memory_map('penguin-dataset.arrow', 'r')\n",
    "table = pa.ipc.RecordBatchFileReader(source).read_all().column(\"Flipper Length (mm)\")\n",
    "result = table.to_pandas().mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.3.5 Read Arrow with memory-mapped API without missing values (zero-copy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "6.15 ms ± 491 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit\n",
    "source = pa.memory_map('penguin-dataset-nonan.arrow', 'r')\n",
    "table = pa.ipc.RecordBatchFileReader(source).read_all().column(\"Flipper Length (mm)\")\n",
    "result = table.to_pandas().mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1.4 Memory consumption - read column"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Measure initial memory consumption\n",
    "memory_init = psutil.Process(os.getpid()).memory_info().rss >> 20"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.4.1 Read csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "col_csv = pd.read_csv(\"penguin-dataset.csv\")[\"Flipper Length (mm)\"]\n",
    "memory_post_csv = psutil.Process(os.getpid()).memory_info().rss >> 20"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.4.2 Read parquet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "col_parquet = pd.read_parquet(\"penguin-dataset.parquet\", columns=[\"Flipper Length (mm)\"])\n",
    "memory_post_parquet = psutil.Process(os.getpid()).memory_info().rss >> 20"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.4.3 Read Arrow using file API"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "with pa.OSFile('penguin-dataset.arrow', 'rb') as source:\n",
    "    col_arrow_file = pa.ipc.open_file(source).read_all().column(\"Flipper Length (mm)\").to_pandas()\n",
    "memory_post_arrowos = psutil.Process(os.getpid()).memory_info().rss >> 20"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.4.4 Read Arrow with memory-mapped API with missing values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "source = pa.memory_map('penguin-dataset.arrow', 'r')\n",
    "table_mmap = pa.ipc.RecordBatchFileReader(source).read_all().column(\"Flipper Length (mm)\")\n",
    "col_arrow_mapped = table_mmap.to_pandas()\n",
    "memory_post_arrowmmap = psutil.Process(os.getpid()).memory_info().rss >> 20"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.4.5 Read Arrow with memory-mapped API without missing values (zero-copy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "source = pa.memory_map('penguin-dataset-nonan.arrow', 'r')\n",
    "table_mmap_zc = pa.ipc.RecordBatchFileReader(source).read_all().column(\"Flipper Length (mm)\")\n",
    "col_arrow_mapped_zc = table_mmap_zc.to_pandas()\n",
    "memory_post_arrowmmap_zc = psutil.Process(os.getpid()).memory_info().rss >> 20"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 1.4.6 Display memory consupmtion"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "csv: 48\n",
      "Parquet: 9\n",
      "Arrow file API: 189\n",
      "Arrow memory-mapped API with NaNs: 15\n",
      "Arrow memory-mapped API (zero-copy): 0\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Print memory consumption\n",
    "print(f\"csv: {memory_post_csv - memory_init}\\n\"\n",
    "      f\"Parquet: {memory_post_parquet - memory_post_csv}\\n\"\n",
    "      f\"Arrow file API: {memory_post_arrowos - memory_post_parquet}\\n\"\n",
    "      f\"Arrow memory-mapped API with NaNs: {memory_post_arrowmmap - memory_post_arrowos}\\n\"\n",
    "      f\"Arrow memory-mapped API (zero-copy): {memory_post_arrowmmap_zc - memory_post_arrowmmap}\\n\")"
   ]
  }
 ],
 "metadata": {
  "hide_input": false,
  "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": "3.7.1"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": true,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}