Data_label_stats.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "#import tensorflow as tf\n",
    "import sys\n",
    "sys.path.append('/nas/longleaf/home/siyangj/NiftyNet/')\n",
    "#import niftynet as nn\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "import SimpleITK as sitk\n",
    "import h5py\n",
    "\n",
    "import os\n",
    "\n",
    "data_path = '/proj/NIRAL/users/siyangj/myData/min_normal/seg_inf_1'\n",
    "pretrain_path = '/proj/NIRAL/users/siyangj/myPretrainData/seg'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [],
   "source": [
    "def summarize_label_stats(path,log=False):\n",
    "    ret = np.zeros([0,4])\n",
    "    for f in os.listdir(path):\n",
    "        if f[-5:]!='.nrrd':\n",
    "            continue\n",
    "        arr = sitk.GetArrayFromImage(sitk.ReadImage(os.path.join(path,f)))\n",
    "        labels,counts = np.unique(arr,return_counts=True)\n",
    "        if log:\n",
    "            print(labels)\n",
    "        ret = np.vstack((ret,counts / np.size(arr)))\n",
    "    return ret"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_counts = summarize_label_stats(data_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.55854725, 0.1931317 , 0.18283226, 0.06548879],\n",
       "       [0.52026658, 0.17538694, 0.23078265, 0.07356383],\n",
       "       [0.56095184, 0.12740168, 0.23441763, 0.07722885],\n",
       "       [0.45246233, 0.13773697, 0.29905289, 0.11074781],\n",
       "       [0.56499663, 0.13780866, 0.22290233, 0.07429238],\n",
       "       [0.57158552, 0.16580152, 0.20074366, 0.0618693 ],\n",
       "       [0.58673483, 0.13184272, 0.20958601, 0.07183644],\n",
       "       [0.47517613, 0.17499167, 0.25043306, 0.09939914]])"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_counts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.25846599, 0.74749751, 0.78960609, 2.2044304 ],\n",
       "       [0.30090878, 0.89261366, 0.67835596, 2.1281216 ],\n",
       "       [0.26561346, 1.16950068, 0.63560219, 1.92928367],\n",
       "       [0.40471584, 1.32948092, 0.61232873, 1.65347451],\n",
       "       [0.26247353, 1.07610551, 0.66529885, 1.99612211],\n",
       "       [0.24193531, 0.8340498 , 0.68887218, 2.23514271],\n",
       "       [0.24364384, 1.08427927, 0.68207954, 1.98999735],\n",
       "       [0.38486242, 1.04506368, 0.73024479, 1.83982911]])"
      ]
     },
     "execution_count": 113,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "raw_d_weights_all = 1 / data_counts\n",
    "d_weights_all = raw_d_weights_all / raw_d_weights_all.sum(axis=1)[:,None] * 4\n",
    "d_weights_all"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [],
   "source": [
    "dw_m = d_weights_all.mean(axis=0)\n",
    "dw_s = d_weights_all.std(axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [],
   "source": [
    "raw_p_weights_all = 1 / pretrain_count\n",
    "p_weights_all = raw_p_weights_all / raw_p_weights_all.sum(axis=1)[:,None] * 4\n",
    "pw_m = p_weights_all.mean(axis=0)\n",
    "pw_s = p_weights_all.std(axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.16094833, 0.80775838, 0.52695764, 2.50433564],\n",
       "       [0.29603318, 0.99596716, 0.64195656, 2.0660431 ],\n",
       "       [0.23008542, 0.83228339, 0.62007698, 2.31755421],\n",
       "       ...,\n",
       "       [0.15409721, 0.66224479, 0.48871488, 2.69494312],\n",
       "       [0.15617935, 0.63264697, 0.49975089, 2.7114228 ],\n",
       "       [0.10438997, 0.56499196, 0.40455541, 2.92606266]])"
      ]
     },
     "execution_count": 116,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "p_weights_all"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "pretrain_count = summarize_label_stats(pretrain_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "p_m = pretrain_count.mean(axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "p_s = pretrain_count.std(axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "d_m = data_counts.mean(axis=0)\n",
    "d_s = data_counts.std(axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure()\n",
    "plt.errorbar(range(4),d_m,yerr=d_s)\n",
    "plt.errorbar(range(4),p_m,yerr=p_s)\n",
    "plt.legend(['Real data','Pretrain data'])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure()\n",
    "plt.errorbar(range(4),dw_m,yerr=dw_s)\n",
    "plt.errorbar(range(4),pw_m,yerr=pw_s)\n",
    "plt.legend(['Real data weights','Pretrain data weights'])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1.0000000000000007"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "p_m.sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.29507895, 1.01768311, 0.69157511, 1.99566283])"
      ]
     },
     "execution_count": 93,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "raw_d_weights = 1 / d_m\n",
    "d_weights = raw_d_weights / raw_d_weights.sum() * 4\n",
    "d_weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.1789348 , 0.72904663, 0.52994981, 2.56206876])"
      ]
     },
     "execution_count": 94,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "raw_p_weights = 1 / p_m\n",
    "p_weights = raw_p_weights / raw_p_weights.sum() * 4\n",
    "p_weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.2953274  1.02232388 0.68529854 1.99705018] \n",
      " [0.18139336 0.71989869 0.52385821 2.57484974] \n",
      " [0.29507895 1.01768311 0.69157511 1.99566283] \n",
      " [0.1789348  0.72904663 0.52994981 2.56206876]\n"
     ]
    }
   ],
   "source": [
    "print(dw_m,'\\n',pw_m,'\\n',d_weights,'\\n',p_weights)"
   ]
  },
  {
   "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": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}