L-24-Optimized_TF_ PipleLine_Performance_Prefetch_Cache.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "46feecec",
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "a0f31303",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'2.10.1'"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c4018fb5",
   "metadata": {},
   "source": [
    "# Prefetch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "ad158106",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Making demy class for better understanding\n",
    "class FileDataset(tf.data.Dataset):\n",
    "    def read_files_in_batches(num_samples):\n",
    "        # open file\n",
    "        time.sleep(0.03)\n",
    "        for sample_idx in range(num_samples):\n",
    "            time.sleep(0.015)\n",
    "            yield (sample_idx,) # yield is generator function\n",
    "            \n",
    "    def __new__(cls, num_samples=3): #over riding new method\n",
    "        return tf.data.Dataset.from_generator(\n",
    "            cls.read_files_in_batches,\n",
    "            output_signature =tf.TensorSpec(shape =(1,), dtype = tf.int64),\n",
    "            args=(num_samples,)\n",
    "        )\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "1ef4e515",
   "metadata": {},
   "outputs": [],
   "source": [
    "def benchmark(dataset, num_epochs=2):\n",
    "    for epoch_num in range(num_epochs):\n",
    "        for sample in dataset:\n",
    "            time.sleep(0.01)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "0704e03f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "398 ms ± 21.2 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit\n",
    "benchmark(FileDataset())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "a269f409",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "345 ms ± 21.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit\n",
    "benchmark(FileDataset().prefetch(1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "cf20cd4f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "338 ms ± 17.9 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit\n",
    "benchmark(FileDataset().prefetch(tf.data.AUTOTUNE))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "09523f2d",
   "metadata": {},
   "source": [
    "# Cache API"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "53017d24",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n",
      "1\n",
      "2\n",
      "3\n",
      "4\n"
     ]
    }
   ],
   "source": [
    "# Creating new dataset and dataset is nothing but a bunch of numbers\n",
    "\n",
    "dataset = tf.data.Dataset.range(5)\n",
    "\n",
    "for d in dataset:\n",
    "    print(d.numpy())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "3f0cd5f1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0, 1, 4, 9, 16]"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Squaring of each number\n",
    "\n",
    "dataset = tf.data.Dataset.range(5)\n",
    "dataset = dataset.map(lambda x: x**2)\n",
    "dataset = dataset.cache(\"mycache.txt\")\n",
    "# The first time reading through the data will generate the data using\n",
    "# `range` and `map`.\n",
    "list(dataset.as_numpy_iterator())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "6bdb70ec",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0, 1, 4, 9, 16]"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Reading data from cache\n",
    "\n",
    "dataset = dataset.cache()\n",
    "\n",
    "list(dataset.as_numpy_iterator())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "84aa9aba",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0, 1, 4, 9, 16]"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "list(dataset.as_numpy_iterator())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "3a139de1",
   "metadata": {},
   "outputs": [],
   "source": [
    "def mapped_function(s):\n",
    "    # Do some hard pre-processing\n",
    "    tf.py_function(lambda: time.sleep(0.03), [], ())\n",
    "    return s"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "c6c6e6c8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.57 s ± 0 ns per loop (mean ± std. dev. of 1 run, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit -r1 -n1\n",
    "benchmark(FileDataset().map(mapped_function), 5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "9389ab0d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "540 ms ± 0 ns per loop (mean ± std. dev. of 1 run, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%%timeit -r1 -n1\n",
    "benchmark(FileDataset().map(mapped_function).cache(), 5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e26199b0",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}