utf-8''AI4M_C1_W1_lecture_ex_02.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## AI for Medicine Course 1 Week 1 lecture exercises"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"counting-labels\"></a>\n",
    "# Counting labels\n",
    "\n",
    "As you saw in the lecture videos, one way to avoid having class imbalance impact the loss function is to weight the losses differently.  To choose the weights, you first need to calculate the class frequencies.\n",
    "\n",
    "For this exercise, you'll just get the count of each label.  Later on, you'll use the concepts practiced here to calculate frequencies in the assignment!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Import the necessary packages\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Read csv file containing training datadata\n",
    "train_df = pd.read_csv(\"nih/train-small.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Count up the number of instances of each class (drop non-class columns from the counts)\n",
    "class_counts = train_df.sum().drop(['Image','PatientId'])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The class Atelectasis has 106 samples\n",
      "The class Cardiomegaly has 20 samples\n",
      "The class Consolidation has 33 samples\n",
      "The class Edema has 16 samples\n",
      "The class Effusion has 128 samples\n",
      "The class Emphysema has 13 samples\n",
      "The class Fibrosis has 14 samples\n",
      "The class Hernia has 2 samples\n",
      "The class Infiltration has 175 samples\n",
      "The class Mass has 45 samples\n",
      "The class Nodule has 54 samples\n",
      "The class Pleural_Thickening has 21 samples\n",
      "The class Pneumonia has 10 samples\n",
      "The class Pneumothorax has 38 samples\n"
     ]
    }
   ],
   "source": [
    "for column in class_counts.keys():\n",
    "    print(f\"The class {column} has {train_df[column].sum()} samples\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot up the distribution of counts\n",
    "sns.barplot(class_counts.values, class_counts.index, color='b')\n",
    "plt.title('Distribution of Classes for Training Dataset', fontsize=15)\n",
    "plt.xlabel('Number of Patients', fontsize=15)\n",
    "plt.ylabel('Diseases', fontsize=15)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a name=\"weighted-loss\"></a>\n",
    "# Weighted Loss function\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Below is an example of calculating weighted loss. For this example, you'll first define a hypothetical set of true labels and then a set of random predictions. You'll use the made up examples to practice with the weighted loss function.\n",
    "\n",
    "Run the next two cells to create simple examples of labels and predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[1]\n",
      " [1]\n",
      " [1]\n",
      " [1]\n",
      " [1]\n",
      " [1]\n",
      " [1]\n",
      " [0]\n",
      " [0]\n",
      " [0]] (10, 1)\n"
     ]
    }
   ],
   "source": [
    "# Generate an array of 10 binary label values, 7 positive and 3 negative, then reshape to a column\n",
    "y_true = np.array([1, 1, 1, 1, 1, 1, 1, 0, 0, 0]).reshape(10, 1)\n",
    "print(y_true, y_true.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0.9]\n",
      " [0. ]\n",
      " [0. ]\n",
      " [0. ]\n",
      " [0. ]\n",
      " [0. ]\n",
      " [0.9]\n",
      " [0.9]\n",
      " [0. ]\n",
      " [0.9]] (10, 1)\n"
     ]
    }
   ],
   "source": [
    "# Generate an array of random predictions (either 0 or 0.9 for each) and reshape to a column\n",
    "y_predict = np.random.randint(0, 2, 10).reshape(10, 1) * 0.9\n",
    "print(y_predict, y_predict.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Run the next two cells to define positive and negative weights and a value for epsilon to be used in the loss function."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define positive and negative weights to be used in the loss function\n",
    "# The positive weight is determined by the fraction of labels that are negative (3/10 in this case)\n",
    "positive_weight = 0.3\n",
    "# The negative weight is determined by the fraction of labels that are positive (7/10 in this case)\n",
    "negative_weight = 0.7"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define a value \"epsilon\" to be used in calculating the loss\n",
    "# This value is just used to avoid an error due to taking the log of zero.\n",
    "epsilon = 1e-7"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Weighted Loss Equation\n",
    "Calculate the loss for the zero-th label (column at index 0)\n",
    "\n",
    "- The loss is made up of two terms:\n",
    "    - $loss_{pos}$: we'll use this to refer to the loss where the actual label is positive (the positive examples).\n",
    "    - $loss_{neg}$: we'll use this to refer to the loss where the actual label is negative (the negative examples).  \n",
    "- Note that within the $log()$ function, we'll add a tiny positive value, to avoid an error if taking the log of zero.\n",
    "\n",
    "$$ loss^{(i)} = loss_{pos}^{(i)} + los_{neg}^{(i)} $$\n",
    "\n",
    "$$loss_{pos}^{(i)} = -1 \\times weight_{pos}^{(i)} \\times y^{(i)} \\times log(\\hat{y}^{(i)} + \\epsilon)$$\n",
    "\n",
    "$$loss_{neg}^{(i)} = -1 \\times weight_{neg}^{(i)} \\times (1- y^{(i)}) \\times log(1 - \\hat{y}^{(i)} + \\epsilon)$$\n",
    "\n",
    "$$\\epsilon = \\text{a tiny positive number}$$\n",
    "\n",
    "Run the next three cells to calculate the positive, negative and total loss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "24.24035971916551"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Calculate and print out the positive loss\n",
    "positive_loss = -1 * np.sum(positive_weight * \n",
    "                y_true * \n",
    "                np.log(y_predict + epsilon)\n",
    "              )\n",
    "positive_loss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3.2236176601923674"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Calculate and print out the negative loss\n",
    "negative_loss = -1 * np.sum( \n",
    "                negative_weight * \n",
    "                (1 - y_true) * \n",
    "                np.log(1 - y_predict + epsilon)\n",
    "              )\n",
    "negative_loss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "27.46397737935788\n"
     ]
    }
   ],
   "source": [
    "# Sum positive and negative losses to calculate total loss\n",
    "total_loss = positive_loss + negative_loss\n",
    "print(total_loss)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### That's all for this lab. You now have a couple more tools you'll need for this week's assignment!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}