prepare-dollar-street-data.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "17b0c3c9-f03b-4378-9713-87cc6522729f",
   "metadata": {},
   "source": [
    "# Prepare dollar street dataset for deep learning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "06be0473-4bc4-467c-8365-fe8879bf9664",
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL import Image\n",
    "import pandas as pd\n",
    "import pathlib\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e790950b-b28f-4d02-b5f6-f5bde1dddce0",
   "metadata": {},
   "source": [
    "## Load data\n",
    "Download the data from https://www.kaggle.com/datasets/mlcommons/the-dollar-street-dataset. Note that it is 101GB!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "7c14ddc6-9187-492e-8901-053658a9c618",
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_FOLDER = pathlib.Path('../data/dataset_dollarstreet/') # change to location where you stored the data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0ddc6d1b-10cc-4159-b04f-106f229aeca1",
   "metadata": {},
   "source": [
    "### Show example image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "341ee5ff-7683-4d95-94dc-c0222fc4caba",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/jpeg": "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",
      "image/png": "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",
      "text/plain": [
       "<PIL.Image.Image image mode=RGB size=64x64>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "example_filepath = DATA_FOLDER / 'assets/5d4bde20cf0b3a0f3f3359fb/5d4bde20cf0b3a0f3f3359fb.jpg'\n",
    "image = Image.open(example_filepath)\n",
    "image = image.resize((64, 64))\n",
    "display(image)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fdaff052-e330-476a-8b07-1606a921f2d3",
   "metadata": {},
   "source": [
    "## Prepare labels\n",
    "1. Only take examples with one imagenet_synonym label\n",
    "2. Use only examples with the 10 most frequently occuring labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "8a41d87c-830c-4358-838f-1e7f37952638",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_csv(DATA_FOLDER / 'images_v2_imagenet_test.csv', converters={'imagenet_synonyms': eval})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "72fbbda4-7638-424a-810d-8cd3f6d8f365",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Take only rows where we have one label (to simplify the problem to multiclass not multilabel classification)\n",
    "df_selection = df[df['imagenet_synonyms'].apply(len) == 1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "65d3adae-9189-4bd0-8b10-de5311d80cf8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "4308"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "3579ff89-02ef-4177-9d01-a1c06cefe426",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3772"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(df_selection)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "7cb520fe-a93b-4445-a526-35d6d75695fa",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/ff/lzvrl_f53bl3fwlbvnsw4f840000gn/T/ipykernel_58901/1875909903.py:1: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  df_selection['label'] = df_selection['imagenet_synonyms'].apply(lambda x: x[0])\n"
     ]
    }
   ],
   "source": [
    "df_selection['label'] = df_selection['imagenet_synonyms'].apply(lambda x: x[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "5a3adbdc-a9fa-40b8-bbf8-e00dcba994d0",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_10 = df_selection[df_selection['label'].isin(df_selection['label'].value_counts()[:10].index)]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ed8c28fa-fe82-4f17-8565-89cf3676fa01",
   "metadata": {},
   "source": [
    "## Prepare images\n",
    "1. Downscale to 64 x 64 pixels\n",
    "2. Store as numpy array"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "75c394cd-1276-4ce0-96b4-aabed5518d51",
   "metadata": {},
   "outputs": [],
   "source": [
    "images_matrix = np.zeros((len(df_10), 64, 64, 3), dtype='int32')\n",
    "for i, (_, row) in enumerate(df_10[['label', 'imageRelPath']].iterrows()):\n",
    "    filepath = DATA_FOLDER / row['imageRelPath']\n",
    "    image = Image.open(filepath)\n",
    "    image = image.resize((64, 64))\n",
    "    data = np.array(image, dtype='int32')\n",
    "    images_matrix[i, :, :, :] = data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3d038632-ff1d-444a-977e-247f9b6f4f4a",
   "metadata": {},
   "source": [
    "## Prepare data for training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "a1da8602-3820-4550-9ad3-0bfa5d887d79",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "train_images, val_images, train_labels, val_labels = train_test_split(\n",
    "    images_matrix, df_10['label'].reset_index(drop=True), stratify=df_10['label'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "02ff5278-0d21-4ef7-b423-cf53915b94cb",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import LabelEncoder\n",
    "encoder = LabelEncoder().fit(train_labels)\n",
    "train_labels = encoder.transform(train_labels)\n",
    "val_labels = encoder.transform(val_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "465fe7e9-1026-414d-acb9-174fd2718fbc",
   "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>label</th>\n",
       "      <th>index</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>day bed</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>dishrag</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>plate</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>running shoe</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>soap dispenser</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>street sign</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>table lamp</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>tile roof</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>toilet seat</td>\n",
       "      <td>8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>washing machine</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             label  index\n",
       "0          day bed      0\n",
       "1          dishrag      1\n",
       "2            plate      2\n",
       "3     running shoe      3\n",
       "4   soap dispenser      4\n",
       "5      street sign      5\n",
       "6       table lamp      6\n",
       "7        tile roof      7\n",
       "8      toilet seat      8\n",
       "9  washing machine      9"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.DataFrame({'label': encoder.classes_, 'index': encoder.transform(encoder.classes_)})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "464ab03b-9546-4396-850a-df311a0c44d7",
   "metadata": {},
   "source": [
    "## Save data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "7713a0ab-9a34-42f7-a5db-a369e78eb98f",
   "metadata": {},
   "outputs": [],
   "source": [
    "np.save(DATA_FOLDER / 'train_images.npy', train_images)\n",
    "np.save(DATA_FOLDER / 'test_images.npy', val_images)\n",
    "np.save(DATA_FOLDER / 'train_labels.npy', train_labels)\n",
    "np.save(DATA_FOLDER / 'test_labels.npy', val_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7dffbd36-1071-4a1a-99fa-8e9fc6d76ac6",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.11.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}