{"cells":[
 {
  "cell_type": "code",
  "execution_count": null,
  "metadata": {
   "collapsed": false
  },
  "outputs": [],
  "source": "import numpy as np\nimport scipy as sp\nimport pandas as pd\nimport matplotlib.pyplot as plt\n\ndf_train = pd.read_csv(\"../input/train.csv\",nrows=2000) \n#df_test = pd.read_csv(\"test.csv\")\nX_train = df_train.drop(\"label\",axis=1)\n\nX_train_n = [0]*10\nfor n in range(10):\n    X_train_n[n] = df_train.loc[df_train.label==n].drop(\"label\",axis=1)\n\nplt.figure(figsize=(12,12))\n\nfor i in range(10):\n    plt.subplot(10,4,i+1)\n    plt.imshow(X_train_n[i].mean(axis=0).reshape(28,28),interpolation='nearest')   \n    #plt.axis('tight')\nplt.figtext(0.5, 0.965,\"average image\",size='large')\n\nplt.show()\n\n# use corr\nnr = 0\nfor k in range(50):\n    maxcorr=0\n    for i in range(10):\n        a = X_train_n[i][:10].mean(axis=0)\n        corr = np.corrcoef(X_train.values[k,:],a)[0,1]\n        #print i,corr\n        if maxcorr <= corr:\n            maxcorr = corr\n            pred = i\n    if pred == df_train.label[k]:\n        nr += 1\n    print (k,maxcorr,pred,df_train.label[k])\nprint (nr / 50)"
 }
],"metadata":{"kernelspec":{"display_name":"Python 3","language":"python","name":"python3"}}, "nbformat": 4, "nbformat_minor": 0}