Data Leakage Findings from Validation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "source": [
    "I find some data leakage from the validation set. Though I am not sure whether it is intentional for it being called 'validation' set. I am actually not sure about the validation set purpose."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
    "_kg_hide-input": true,
    "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "        <script type=\"text/javascript\">\n",
       "        window.PlotlyConfig = {MathJaxConfig: 'local'};\n",
       "        if (window.MathJax) {MathJax.Hub.Config({SVG: {font: \"STIX-Web\"}});}\n",
       "        if (typeof require !== 'undefined') {\n",
       "        require.undef(\"plotly\");\n",
       "        requirejs.config({\n",
       "            paths: {\n",
       "                'plotly': ['https://cdn.plot.ly/plotly-latest.min']\n",
       "            }\n",
       "        });\n",
       "        require(['plotly'], function(Plotly) {\n",
       "            window._Plotly = Plotly;\n",
       "        });\n",
       "        }\n",
       "        </script>\n",
       "        "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import time\n",
    "script_start_time = time.time()\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import json\n",
    "import gc\n",
    "\n",
    "# Visualization\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "%matplotlib inline\n",
    "\n",
    "import plotly.plotly as py\n",
    "import cufflinks as cf\n",
    "cf.set_config_file(offline=True, world_readable=True, theme='ggplot')\n",
    "plt.rcParams[\"figure.figsize\"] = 12,8\n",
    "sns.set(rc={'figure.figsize':(20,12)})\n",
    "plt.style.use('fivethirtyeight')\n",
    "\n",
    "pd.set_option('display.max_rows', 600)\n",
    "pd.set_option('display.max_columns', 50)\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')\n",
    "\n",
    "# Data path\n",
    "data_path = './data'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "9b71624737b094f61d08a636414e77de6ea073c0"
   },
   "source": [
    "### 1. Load Data\n",
    "Reference: [](https://www.kaggle.com/badalgupta/simple-data-exploration)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "_uuid": "d3ce804f50e9450a193650e617541631da67abe7"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.04 min: Start loading data\n",
      "Train No. of images: 1014544\n",
      "Test No. of images: 39706\n",
      "Validation No. of images: 9897\n",
      "0.16 min: Finish loading data\n",
      "==================================================\n"
     ]
    }
   ],
   "source": [
    "# 1. Load data =================================================================\n",
    "print('%0.2f min: Start loading data'%((time.time() - script_start_time)/60))\n",
    "train={}\n",
    "test={}\n",
    "validation={}\n",
    "with open('%s/train.json'%(data_path)) as json_data:\n",
    "    train= json.load(json_data)\n",
    "with open('%s/test.json'%(data_path)) as json_data:\n",
    "    test= json.load(json_data)\n",
    "with open('%s/validation.json'%(data_path)) as json_data:\n",
    "    validation = json.load(json_data)\n",
    "\n",
    "print('Train No. of images: %d'%(len(train['images'])))\n",
    "print('Test No. of images: %d'%(len(test['images'])))\n",
    "print('Validation No. of images: %d'%(len(validation['images'])))\n",
    "\n",
    "# JSON TO PANDAS DATAFRAME\n",
    "# train data\n",
    "train_img_url=train['images']\n",
    "train_img_url=pd.DataFrame(train_img_url)\n",
    "train_ann=train['annotations']\n",
    "train_ann=pd.DataFrame(train_ann)\n",
    "train=pd.merge(train_img_url, train_ann, on='imageId', how='inner')\n",
    "\n",
    "# test data\n",
    "test=pd.DataFrame(test['images'])\n",
    "\n",
    "# Validation Data\n",
    "val_img_url=validation['images']\n",
    "val_img_url=pd.DataFrame(val_img_url)\n",
    "val_ann=validation['annotations']\n",
    "val_ann=pd.DataFrame(val_ann)\n",
    "validation=pd.merge(val_img_url, val_ann, on='imageId', how='inner')\n",
    "\n",
    "del (train_img_url, train_ann, val_img_url, val_ann)\n",
    "gc.collect()\n",
    "\n",
    "print('%0.2f min: Finish loading data'%((time.time() - script_start_time)/60))\n",
    "print('='*50)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "9f7a8ca33d5d584a99ea623a3df5c59ed65f17ee"
   },
   "source": [
    "### 2. Check Data for Missing or Duplicated Values\n",
    "Findings:\n",
    "a. No missing values\n",
    "b. No duplicates in each data set, but val urls are from test.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "_uuid": "84885b31648270bace3eeab62f8d1deb0fe755e9"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train shape: (1014544, 3)\n",
      "Train Unique imageId: 1014544\n",
      "Train Unique url: 1014544\n",
      "Train NA: \n",
      "imageId    0\n",
      "url        0\n",
      "labelId    0\n",
      "dtype: int64\n",
      "Train total unique url: 1014544\n",
      "--------------------------------------------------\n",
      "Test shape: (39706, 2)\n",
      "Test Unique imageId: 39706\n",
      "Test Unique url: 39706\n",
      "Test NA: \n",
      "imageId    0\n",
      "url        0\n",
      "dtype: int64\n",
      "Test total unique url: 39706\n",
      "--------------------------------------------------\n",
      "Validation shape: (9897, 3)\n",
      "Validation Unique imageId: 9897\n",
      "Validation Unique url: 9897\n",
      "Validation NA: \n",
      "imageId    0\n",
      "url        0\n",
      "labelId    0\n",
      "dtype: int64\n",
      "Validation total unique url: 9897\n",
      "--------------------------------------------------\n",
      "Total images: 1064147\n",
      "Total unique images: 1054250\n",
      "Duplicated url: 9897\n"
     ]
    }
   ],
   "source": [
    "datas = {'Train': train, 'Test': test, 'Validation': validation}\n",
    "\n",
    "total_url = []\n",
    "dataset_url = {}\n",
    "for data_name, data in datas.items():\n",
    "    print('%s shape: %s'%(data_name, str(data.shape)))\n",
    "    print('%s Unique imageId: %s'%(data_name, len(data['imageId'].unique())))\n",
    "    print('%s Unique url: %s'%(data_name, len(data['url'].unique())))\n",
    "    print('%s NA: '%(data_name)) # No missing values\n",
    "    print(data.isnull().sum()) # No missing values\n",
    "    print('%s total unique url: %d'%(data_name, len(set(data['url'].tolist()))))\n",
    "    total_url = total_url + data['url'].tolist()\n",
    "    dataset_url[data_name] = data['url'].tolist()\n",
    "\n",
    "    print('-'*50)\n",
    "\n",
    "\n",
    "print('Total images: %d'%(len(total_url)))\n",
    "print('Total unique images: %d'%(len(set(total_url))))\n",
    "print('Duplicated url: %d'%(len(total_url) - len(set(total_url))))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "2fddf812105b69112fef7833b09b596a675684dd"
   },
   "source": [
    "9897? A bit familiar? Yes, same as the number of val rows. \n",
    "Still, let's create a grid to search for the duplicated the url"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "_uuid": "3915a5f028261a310e21b5be5169de8e5d0f0215"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train inter Train: 1014544 | 1014544\n",
      "Train inter Test: 1014544 | 0\n",
      "Train inter Validation: 1014544 | 0\n",
      "Test inter Train: 39706 | 0\n",
      "Test inter Test: 39706 | 39706\n",
      "Test inter Validation: 39706 | 9897\n",
      "Validation inter Train: 9897 | 0\n",
      "Validation inter Test: 9897 | 9897\n",
      "Validation inter Validation: 9897 | 9897\n"
     ]
    }
   ],
   "source": [
    "# Find the duplicated url\n",
    "from itertools import product\n",
    "combinations = list(product(*[datas.keys(), datas.keys()]))\n",
    "for comb in combinations:\n",
    "    print('%s inter %s: %d | %d'%(comb[0], comb[1], len(set(dataset_url[comb[0]])), len(set(dataset_url[comb[0]]).intersection(set(dataset_url[comb[1]])))))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "53e07125e5945ff754b1e2de42c5fb7ca6151f78"
   },
   "source": [
    "'Test inter Validation: 39706 | 9897'\n",
    "What? Val is part of test?\n",
    "Let's confirm the duplicated url by merging test and val on url"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "_uuid": "2c1a5a03494cfb71de18c276be9c731f83c719bc"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test NA: \n",
      "imageId    0.000000\n",
      "url        0.000000\n",
      "labelId    0.750743\n",
      "dtype: float64\n"
     ]
    }
   ],
   "source": [
    "# Confirm the duplicated url\n",
    "test[['url']].merge(validation[['url']], how = 'inner').shape\n",
    "test_ = test.merge(validation[['url', 'labelId']], on = 'url',how = 'left')\n",
    "print('%s NA: '%('Test')) # No missing values\n",
    "print(test_.isnull().sum() / test_.isnull().count())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "295e4283f0c076cca2dac5f8d148d91a338d05ae"
   },
   "source": [
    "Let's which part of test are leaked"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "_uuid": "e4ce0e6344b82999940ae4a2c2e86145bd4f9845"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x122151f60>]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x864 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot imageId of of leaked lable\n",
    "test_leaked_lableId = test_[~pd.isna(test_['labelId'])]\n",
    "test_unleaked_lableId = test_[pd.isna(test_['labelId'])]\n",
    "number_of_known_lableId = validation.shape[0]\n",
    "plt.plot(test_leaked_lableId['imageId'], '.')\n",
    "plt.plot(test_unleaked_lableId['imageId'], '.')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "8fb473aa20f4e8bb603a31a43667f93b97008b2a"
   },
   "source": [
    "Display images to confirm again the leakage"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "_uuid": "da32f54e00b5c27e426db6aca81971342a72b52b"
   },
   "outputs": [],
   "source": [
    "from IPython.display import Image\n",
    "from IPython.core.display import HTML\n",
    "\n",
    "def display_category(urls, category_name):\n",
    "    img_style = \"width: 180px; margin: 0px; float: left; border: 1px solid black;\"\n",
    "    images_list = ''.join([f\"<img style='{img_style}' src='{u}' />\" for _, u in urls.head(12).iteritems()])\n",
    "    display(HTML(images_list))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "_uuid": "1ed6e017a26fb7b5e715a3bf1fc52fd99eb7f397"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/5452f9925f313f502bf119ff-large' /><img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/540584051d2d435c5a300a82-large' /><img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/540c6760f4bdc40bcc10296d-large' /><img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/5447777d4ad3ab71267befe2-large' />"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#test data Images\n",
    "urls = test['url'][1:5]\n",
    "display_category(urls, \"\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "_uuid": "bd26ce25803a6ffc960214c4ef13214806749c5e"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/5452f9925f313f502bf119ff-large' /><img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/540584051d2d435c5a300a82-large' /><img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/540c6760f4bdc40bcc10296d-large' /><img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/5447777d4ad3ab71267befe2-large' />"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#validation Images\n",
    "urls = validation['url'][1:5]\n",
    "display_category(urls, \"\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "_uuid": "a0c0b42f9f8e427836f1c214024c515617752def"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/5468f1c0d96b290ff8e5c805-large' /><img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/546410237d57f323e72ca414-large' /><img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/550b955fdd699c1a0351f84e-large' /><img style='width: 180px; margin: 0px; float: left; border: 1px solid black;' src='https://contestimg.wish.com/api/webimage/54451f33355b4e0fd3028a30-large' />"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#train_data Images\n",
    "urls = train['url'][1:5]\n",
    "display_category(urls, \"\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "36d9ea74ef5ed33ac644a0932dff09958397dd64"
   },
   "source": [
    "Now, we are 100% sure that the first 9897 of test set has their label id in val."
   ]
  }
 ],
 "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": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}