Evaluation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h1> Evaluate Training Process and Submit Test Predictions </h1>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "np.set_printoptions(suppress=True) # no exponential print\n",
    "import pandas as pd\n",
    "pd.set_option('max_columns', 300)\n",
    "import matplotlib.pyplot as plt\n",
    "plt.style.use('seaborn-white')\n",
    "%matplotlib inline\n",
    "import cv2\n",
    "import os, sys, time\n",
    "from multiprocessing import cpu_count\n",
    "# sklearn:\n",
    "from scipy.optimize import curve_fit\n",
    "from sklearn.metrics import accuracy_score, f1_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "RUN_NAME  = '04.01.19_2' ######################################## UPDATE THIS ########################################\n",
    "BEST_NAME = '04.01.19_1'\n",
    "\n",
    "LABELS_DIR = \"D:\\\\OneDrive - studium.uni-hamburg.de\\\\Programmierung\\\\Python\\\\Kaggle\\\\GalaxyZoo\\\\\"\n",
    "TRAIN_DIR  = \"C:\\\\ML-Datasets\\\\\\GalaxyZoo\\\\train\\\\\"\n",
    "TEST_DIR   = \"C:\\\\ML-Datasets\\\\\\GalaxyZoo\\\\test\\\\\"\n",
    "\n",
    "NET_FOLDER   = 'MyNet\\\\'\n",
    "IMG_SIZE     = (224, 224)\n",
    "DATA_TYPE = \".jpg\"\n",
    "    \n",
    "WEIGHTS_DIR = LABELS_DIR + 'results_and_weights\\\\' + NET_FOLDER"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2> Loss and Accuracy plotted over Epoch </h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load training log\n",
    "train_hist_curr = np.load(WEIGHTS_DIR + RUN_NAME  + '\\\\' + \"training_log.npy\")\n",
    "train_hist_best = np.load(WEIGHTS_DIR + BEST_NAME + '\\\\' + \"training_log.npy\") # currently the best run for comparison\n",
    "\n",
    "loss, val_loss, acc, val_acc, f1_score, val_f1_score = [train_hist_curr.item()[item] for item in ['loss', 'val_loss', 'acc', 'val_acc', 'f1_score', 'val_f1_score']]\n",
    "loss_best, val_loss_best, acc_best, val_acc_best, f1_score_best, val_f1_score_best = [train_hist_best.item()[item] for item in ['loss', 'val_loss', 'acc', 'val_acc', 'f1_score', 'val_f1_score']] "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Compare current run with best run:\n",
      "Check if runs had the same loss function and was trained only on one target images or on all images\n",
      "min val_loss      0.009774602982286665 at Epoch 91\n",
      "min val_loss_best 0.011917536359276719 at Epoch 63\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(\"Compare current run with best run:\")\n",
    "print(\"Check if runs had the same loss function and was trained only on one target images or on all images\")\n",
    "print(\"min val_loss     \", np.min(val_loss), \"at Epoch\", np.argmin(val_loss))\n",
    "print(\"min val_loss_best\", np.min(val_loss_best), \"at Epoch\", np.argmin(val_loss_best))\n",
    "plt.figure(figsize=(10,5))\n",
    "plt.plot(list(range(len(loss_best))),     loss_best, 'o-', color=\"g\",label='loss_best', alpha=0.6)\n",
    "plt.plot(list(range(len(val_loss_best))), val_loss_best, 'o-', color=\"c\",label='val_loss_best', alpha=0.6)\n",
    "plt.plot(list(range(len(loss))),     loss,'-o', color=\"darkorange\", label='loss', alpha=0.6)\n",
    "plt.plot(list(range(len(val_loss))), val_loss,'-o', color=\"peru\",label='val_loss', alpha=0.6)\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.ylim(0,0.2)\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "max val_acc 0.8054563170053929\n",
      "max val_acc_best 0.7689184800453409\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(\"max val_acc\", np.max(val_acc))\n",
    "print(\"max val_acc_best\", np.max(val_acc_best))\n",
    "\n",
    "plt.figure(figsize=(20,5))\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.plot(list(range(len(val_acc))), val_acc,'-o', label='val_acc')\n",
    "plt.plot(list(range(len(acc))), acc,'-o', label='acc')\n",
    "plt.plot(list(range(len(val_acc_best))), val_acc_best,'-o', label='val_acc_best')\n",
    "plt.plot(list(range(len(acc_best))), acc_best,'-o', label='acc_best')\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.legend()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2> Predict Test Data for Submission to Kaggle </h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "import keras.backend as K\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras import layers, models\n",
    "from keras.layers import Input, Dropout, Flatten, Conv2D, MaxPooling2D, Dense, BatchNormalization\n",
    "from keras.optimizers import Adam\n",
    "from keras.initializers import Orthogonal\n",
    "from keras.layers.advanced_activations import LeakyReLU\n",
    "\n",
    "labels = pd.read_csv(LABELS_DIR + \"training_solutions_rev1.csv\") # just for the shape and column names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 0 images belonging to 0 classes.\n",
      "Ignore prints above. The generator(s) got setup correctly.\n"
     ]
    }
   ],
   "source": [
    "test_IDs = [int(x.rpartition('.')[0]) for x in os.listdir(TEST_DIR)]\n",
    "\n",
    "test_gen  = ImageDataGenerator(rescale     = 1./255,\n",
    "                               fill_mode   = 'reflect',\n",
    "                               data_format = 'channels_last')\n",
    "# test data\n",
    "test_generator = test_gen.flow_from_directory(directory = TEST_DIR, \n",
    "                                              class_mode  = 'sparse',\n",
    "                                              target_size = IMG_SIZE,\n",
    "                                              color_mode  = 'rgb',\n",
    "                                              batch_size  = 128,\n",
    "                                              shuffle     = False)\n",
    "test_generator.filenames   = np.array([str(img_id)+DATA_TYPE for img_id in test_IDs])\n",
    "test_generator.classes     = np.zeros((len(test_IDs), 37))\n",
    "test_generator.samples     = len(test_IDs)\n",
    "test_generator.n           = len(test_IDs)\n",
    "\n",
    "print(\"Ignore prints above. The generator(s) got setup correctly.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "RGB-In (InputLayer)          (None, 224, 224, 3)       0         \n",
      "_________________________________________________________________\n",
      "conv2d_21 (Conv2D)           (None, 112, 112, 96)      10464     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_13 (MaxPooling (None, 37, 37, 96)        0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_27 (Batc (None, 37, 37, 96)        384       \n",
      "_________________________________________________________________\n",
      "conv2d_22 (Conv2D)           (None, 37, 37, 256)       221440    \n",
      "_________________________________________________________________\n",
      "max_pooling2d_14 (MaxPooling (None, 18, 18, 256)       0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_28 (Batc (None, 18, 18, 256)       1024      \n",
      "_________________________________________________________________\n",
      "conv2d_23 (Conv2D)           (None, 18, 18, 384)       885120    \n",
      "_________________________________________________________________\n",
      "batch_normalization_29 (Batc (None, 18, 18, 384)       1536      \n",
      "_________________________________________________________________\n",
      "conv2d_24 (Conv2D)           (None, 18, 18, 384)       1327488   \n",
      "_________________________________________________________________\n",
      "batch_normalization_30 (Batc (None, 18, 18, 384)       1536      \n",
      "_________________________________________________________________\n",
      "conv2d_25 (Conv2D)           (None, 18, 18, 256)       884992    \n",
      "_________________________________________________________________\n",
      "max_pooling2d_15 (MaxPooling (None, 8, 8, 256)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_31 (Batc (None, 8, 8, 256)         1024      \n",
      "_________________________________________________________________\n",
      "flatten_5 (Flatten)          (None, 16384)             0         \n",
      "_________________________________________________________________\n",
      "dense_11 (Dense)             (None, 128)               2097280   \n",
      "_________________________________________________________________\n",
      "dropout_7 (Dropout)          (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_32 (Batc (None, 128)               512       \n",
      "_________________________________________________________________\n",
      "dense_12 (Dense)             (None, 50)                6450      \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_4 (LeakyReLU)    (None, 50)                0         \n",
      "_________________________________________________________________\n",
      "dropout_8 (Dropout)          (None, 50)                0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_33 (Batc (None, 50)                200       \n",
      "_________________________________________________________________\n",
      "dense_13 (Dense)             (None, 37)                1887      \n",
      "=================================================================\n",
      "Total params: 5,441,337\n",
      "Trainable params: 5,438,229\n",
      "Non-trainable params: 3,108\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "img_shape       = (IMG_SIZE[0], IMG_SIZE[1], 3)\n",
    "img_in          = layers.Input(img_shape, name='RGB-In')\n",
    "\n",
    "orthogonal_init = Orthogonal(gain=1.0, seed=43)\n",
    "'''  CNN-Layers   ''' \n",
    "# AlexNet (changed strides first layer 4->3)\n",
    "conv_1          = BatchNormalization()(MaxPooling2D(pool_size=(3, 3))(Conv2D(96, (6,6), strides=2, activation = 'relu', kernel_initializer=orthogonal_init, padding='same')(img_in)))\n",
    "conv_2          = BatchNormalization()(MaxPooling2D(pool_size=(3, 3), strides=2)(Conv2D(256, (3,3),  activation = 'relu', kernel_initializer=orthogonal_init, padding='same')(conv_1)))\n",
    "conv_3          = BatchNormalization()(Conv2D(384, (3,3), activation = 'relu', kernel_initializer=orthogonal_init, padding='same')(conv_2))\n",
    "conv_4          = BatchNormalization()(Conv2D(384, (3,3), activation = 'relu', kernel_initializer=orthogonal_init, padding='same')(conv_3))\n",
    "conv_5          = BatchNormalization()(MaxPooling2D(pool_size=(3, 3), strides=2)(Conv2D(256, (3,3),  activation = 'relu', kernel_initializer=orthogonal_init, padding='same')(conv_4)))\n",
    "flatten         = Flatten()(conv_5)\n",
    "\n",
    "'''   Dense Layers   '''\n",
    "dense_1         = Dense(128, kernel_initializer=orthogonal_init, activation = 'relu')(flatten) # , activation = 'relu'\n",
    "# activation_1    = LeakyReLU(alpha=.01)(dense_1)\n",
    "dropout_1        = Dropout(0.4)(dense_1)\n",
    "bn_6            = BatchNormalization()(dropout_1)\n",
    "\n",
    "dense_2         = Dense(50, kernel_initializer=orthogonal_init, activation = 'linear')(bn_6) # , activation = 'relu'\n",
    "activation_2    = LeakyReLU(alpha=.01)(dense_2)\n",
    "dropout_2       = Dropout(0.4)(activation_2)\n",
    "bn_7            = BatchNormalization()(dropout_2)\n",
    "\n",
    "out_layer       = Dense(labels.shape[1]-1, activation = 'sigmoid')(bn_7) # 37 classes\n",
    "\n",
    "'''   Define and compile Model   '''\n",
    "galaxy_model   = models.Model(inputs = [img_in], outputs = [out_layer], name = 'galaxy_model')\n",
    "galaxy_model.compile(loss      = 'mean_squared_error',\n",
    "                      optimizer = Adam(lr=1e-5),\n",
    "                      metrics   = ['acc', 'binary_crossentropy'])\n",
    "galaxy_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "galaxy_model.load_weights(WEIGHTS_DIR + RUN_NAME + '\\\\best.weights.hdf5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "625/625 [==============================] - 81s 129ms/step\n",
      "80.96 s\n"
     ]
    }
   ],
   "source": [
    "t0 = time.time()\n",
    "test_generator.reset()\n",
    "predicted_test = galaxy_model.predict_generator(test_generator, workers=cpu_count(), use_multiprocessing=False, verbose=1)\n",
    "# np.save(WEIGHTS_DIR + RUN_NAME + '\\\\' + \"test_predictions\", predicted_test) # get saved below as csv\n",
    "print(np.round(time.time()-t0, 2),\"s\") # ~80s"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "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>GalaxyID</th>\n",
       "      <th>Class1.1</th>\n",
       "      <th>Class1.2</th>\n",
       "      <th>Class1.3</th>\n",
       "      <th>Class2.1</th>\n",
       "      <th>Class2.2</th>\n",
       "      <th>Class3.1</th>\n",
       "      <th>Class3.2</th>\n",
       "      <th>Class4.1</th>\n",
       "      <th>Class4.2</th>\n",
       "      <th>Class5.1</th>\n",
       "      <th>Class5.2</th>\n",
       "      <th>Class5.3</th>\n",
       "      <th>Class5.4</th>\n",
       "      <th>Class6.1</th>\n",
       "      <th>Class6.2</th>\n",
       "      <th>Class7.1</th>\n",
       "      <th>Class7.2</th>\n",
       "      <th>Class7.3</th>\n",
       "      <th>Class8.1</th>\n",
       "      <th>Class8.2</th>\n",
       "      <th>Class8.3</th>\n",
       "      <th>Class8.4</th>\n",
       "      <th>Class8.5</th>\n",
       "      <th>Class8.6</th>\n",
       "      <th>Class8.7</th>\n",
       "      <th>Class9.1</th>\n",
       "      <th>Class9.2</th>\n",
       "      <th>Class9.3</th>\n",
       "      <th>Class10.1</th>\n",
       "      <th>Class10.2</th>\n",
       "      <th>Class10.3</th>\n",
       "      <th>Class11.1</th>\n",
       "      <th>Class11.2</th>\n",
       "      <th>Class11.3</th>\n",
       "      <th>Class11.4</th>\n",
       "      <th>Class11.5</th>\n",
       "      <th>Class11.6</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>100018</td>\n",
       "      <td>0.387885</td>\n",
       "      <td>0.601414</td>\n",
       "      <td>0.024693</td>\n",
       "      <td>0.104442</td>\n",
       "      <td>0.518728</td>\n",
       "      <td>0.138393</td>\n",
       "      <td>0.405501</td>\n",
       "      <td>0.245445</td>\n",
       "      <td>0.303703</td>\n",
       "      <td>0.050673</td>\n",
       "      <td>0.239904</td>\n",
       "      <td>0.237411</td>\n",
       "      <td>0.043449</td>\n",
       "      <td>0.214783</td>\n",
       "      <td>0.784572</td>\n",
       "      <td>0.102933</td>\n",
       "      <td>0.258385</td>\n",
       "      <td>0.087883</td>\n",
       "      <td>0.053179</td>\n",
       "      <td>0.009528</td>\n",
       "      <td>0.036669</td>\n",
       "      <td>0.049618</td>\n",
       "      <td>0.067196</td>\n",
       "      <td>0.044228</td>\n",
       "      <td>0.001215</td>\n",
       "      <td>0.078608</td>\n",
       "      <td>0.009274</td>\n",
       "      <td>0.039519</td>\n",
       "      <td>0.125219</td>\n",
       "      <td>0.100370</td>\n",
       "      <td>0.044491</td>\n",
       "      <td>0.017482</td>\n",
       "      <td>0.101302</td>\n",
       "      <td>0.017367</td>\n",
       "      <td>0.003716</td>\n",
       "      <td>0.003290</td>\n",
       "      <td>0.119412</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>100037</td>\n",
       "      <td>0.377257</td>\n",
       "      <td>0.605967</td>\n",
       "      <td>0.021674</td>\n",
       "      <td>0.287966</td>\n",
       "      <td>0.277485</td>\n",
       "      <td>0.058013</td>\n",
       "      <td>0.224332</td>\n",
       "      <td>0.077310</td>\n",
       "      <td>0.186283</td>\n",
       "      <td>0.040223</td>\n",
       "      <td>0.124145</td>\n",
       "      <td>0.094234</td>\n",
       "      <td>0.020180</td>\n",
       "      <td>0.163061</td>\n",
       "      <td>0.838304</td>\n",
       "      <td>0.032466</td>\n",
       "      <td>0.104395</td>\n",
       "      <td>0.189025</td>\n",
       "      <td>0.014055</td>\n",
       "      <td>0.009382</td>\n",
       "      <td>0.032271</td>\n",
       "      <td>0.033471</td>\n",
       "      <td>0.057426</td>\n",
       "      <td>0.027331</td>\n",
       "      <td>0.001407</td>\n",
       "      <td>0.151234</td>\n",
       "      <td>0.021349</td>\n",
       "      <td>0.113677</td>\n",
       "      <td>0.032431</td>\n",
       "      <td>0.025986</td>\n",
       "      <td>0.022907</td>\n",
       "      <td>0.009609</td>\n",
       "      <td>0.032520</td>\n",
       "      <td>0.007953</td>\n",
       "      <td>0.002573</td>\n",
       "      <td>0.002770</td>\n",
       "      <td>0.052434</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>100042</td>\n",
       "      <td>0.449369</td>\n",
       "      <td>0.528660</td>\n",
       "      <td>0.023880</td>\n",
       "      <td>0.232809</td>\n",
       "      <td>0.230839</td>\n",
       "      <td>0.037952</td>\n",
       "      <td>0.186006</td>\n",
       "      <td>0.047115</td>\n",
       "      <td>0.174675</td>\n",
       "      <td>0.033204</td>\n",
       "      <td>0.091977</td>\n",
       "      <td>0.081029</td>\n",
       "      <td>0.019076</td>\n",
       "      <td>0.133437</td>\n",
       "      <td>0.867818</td>\n",
       "      <td>0.035734</td>\n",
       "      <td>0.153621</td>\n",
       "      <td>0.186303</td>\n",
       "      <td>0.009101</td>\n",
       "      <td>0.008860</td>\n",
       "      <td>0.027673</td>\n",
       "      <td>0.025700</td>\n",
       "      <td>0.051208</td>\n",
       "      <td>0.021611</td>\n",
       "      <td>0.001119</td>\n",
       "      <td>0.122189</td>\n",
       "      <td>0.016402</td>\n",
       "      <td>0.087094</td>\n",
       "      <td>0.021052</td>\n",
       "      <td>0.014990</td>\n",
       "      <td>0.014352</td>\n",
       "      <td>0.006544</td>\n",
       "      <td>0.017048</td>\n",
       "      <td>0.005210</td>\n",
       "      <td>0.001892</td>\n",
       "      <td>0.002115</td>\n",
       "      <td>0.039099</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>100052</td>\n",
       "      <td>0.443225</td>\n",
       "      <td>0.540962</td>\n",
       "      <td>0.028570</td>\n",
       "      <td>0.069488</td>\n",
       "      <td>0.482546</td>\n",
       "      <td>0.130451</td>\n",
       "      <td>0.398485</td>\n",
       "      <td>0.211564</td>\n",
       "      <td>0.312920</td>\n",
       "      <td>0.033603</td>\n",
       "      <td>0.216789</td>\n",
       "      <td>0.254294</td>\n",
       "      <td>0.045461</td>\n",
       "      <td>0.200052</td>\n",
       "      <td>0.799664</td>\n",
       "      <td>0.238906</td>\n",
       "      <td>0.208959</td>\n",
       "      <td>0.059162</td>\n",
       "      <td>0.051622</td>\n",
       "      <td>0.008035</td>\n",
       "      <td>0.033298</td>\n",
       "      <td>0.037669</td>\n",
       "      <td>0.067697</td>\n",
       "      <td>0.039352</td>\n",
       "      <td>0.000915</td>\n",
       "      <td>0.048968</td>\n",
       "      <td>0.006184</td>\n",
       "      <td>0.030772</td>\n",
       "      <td>0.111273</td>\n",
       "      <td>0.083885</td>\n",
       "      <td>0.032362</td>\n",
       "      <td>0.014937</td>\n",
       "      <td>0.080804</td>\n",
       "      <td>0.014366</td>\n",
       "      <td>0.002274</td>\n",
       "      <td>0.002515</td>\n",
       "      <td>0.106084</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>100056</td>\n",
       "      <td>0.301950</td>\n",
       "      <td>0.685563</td>\n",
       "      <td>0.012990</td>\n",
       "      <td>0.388408</td>\n",
       "      <td>0.256002</td>\n",
       "      <td>0.053268</td>\n",
       "      <td>0.207229</td>\n",
       "      <td>0.084657</td>\n",
       "      <td>0.146723</td>\n",
       "      <td>0.039161</td>\n",
       "      <td>0.125971</td>\n",
       "      <td>0.070082</td>\n",
       "      <td>0.013542</td>\n",
       "      <td>0.155289</td>\n",
       "      <td>0.845923</td>\n",
       "      <td>0.011065</td>\n",
       "      <td>0.053553</td>\n",
       "      <td>0.179152</td>\n",
       "      <td>0.009614</td>\n",
       "      <td>0.006060</td>\n",
       "      <td>0.029454</td>\n",
       "      <td>0.031519</td>\n",
       "      <td>0.047580</td>\n",
       "      <td>0.021683</td>\n",
       "      <td>0.000970</td>\n",
       "      <td>0.186816</td>\n",
       "      <td>0.024515</td>\n",
       "      <td>0.134099</td>\n",
       "      <td>0.029783</td>\n",
       "      <td>0.025864</td>\n",
       "      <td>0.023233</td>\n",
       "      <td>0.007588</td>\n",
       "      <td>0.035305</td>\n",
       "      <td>0.007227</td>\n",
       "      <td>0.002174</td>\n",
       "      <td>0.002150</td>\n",
       "      <td>0.047553</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   GalaxyID  Class1.1  Class1.2  Class1.3  Class2.1  Class2.2  Class3.1  \\\n",
       "0    100018  0.387885  0.601414  0.024693  0.104442  0.518728  0.138393   \n",
       "1    100037  0.377257  0.605967  0.021674  0.287966  0.277485  0.058013   \n",
       "2    100042  0.449369  0.528660  0.023880  0.232809  0.230839  0.037952   \n",
       "3    100052  0.443225  0.540962  0.028570  0.069488  0.482546  0.130451   \n",
       "4    100056  0.301950  0.685563  0.012990  0.388408  0.256002  0.053268   \n",
       "\n",
       "   Class3.2  Class4.1  Class4.2  Class5.1  Class5.2  Class5.3  Class5.4  \\\n",
       "0  0.405501  0.245445  0.303703  0.050673  0.239904  0.237411  0.043449   \n",
       "1  0.224332  0.077310  0.186283  0.040223  0.124145  0.094234  0.020180   \n",
       "2  0.186006  0.047115  0.174675  0.033204  0.091977  0.081029  0.019076   \n",
       "3  0.398485  0.211564  0.312920  0.033603  0.216789  0.254294  0.045461   \n",
       "4  0.207229  0.084657  0.146723  0.039161  0.125971  0.070082  0.013542   \n",
       "\n",
       "   Class6.1  Class6.2  Class7.1  Class7.2  Class7.3  Class8.1  Class8.2  \\\n",
       "0  0.214783  0.784572  0.102933  0.258385  0.087883  0.053179  0.009528   \n",
       "1  0.163061  0.838304  0.032466  0.104395  0.189025  0.014055  0.009382   \n",
       "2  0.133437  0.867818  0.035734  0.153621  0.186303  0.009101  0.008860   \n",
       "3  0.200052  0.799664  0.238906  0.208959  0.059162  0.051622  0.008035   \n",
       "4  0.155289  0.845923  0.011065  0.053553  0.179152  0.009614  0.006060   \n",
       "\n",
       "   Class8.3  Class8.4  Class8.5  Class8.6  Class8.7  Class9.1  Class9.2  \\\n",
       "0  0.036669  0.049618  0.067196  0.044228  0.001215  0.078608  0.009274   \n",
       "1  0.032271  0.033471  0.057426  0.027331  0.001407  0.151234  0.021349   \n",
       "2  0.027673  0.025700  0.051208  0.021611  0.001119  0.122189  0.016402   \n",
       "3  0.033298  0.037669  0.067697  0.039352  0.000915  0.048968  0.006184   \n",
       "4  0.029454  0.031519  0.047580  0.021683  0.000970  0.186816  0.024515   \n",
       "\n",
       "   Class9.3  Class10.1  Class10.2  Class10.3  Class11.1  Class11.2  Class11.3  \\\n",
       "0  0.039519   0.125219   0.100370   0.044491   0.017482   0.101302   0.017367   \n",
       "1  0.113677   0.032431   0.025986   0.022907   0.009609   0.032520   0.007953   \n",
       "2  0.087094   0.021052   0.014990   0.014352   0.006544   0.017048   0.005210   \n",
       "3  0.030772   0.111273   0.083885   0.032362   0.014937   0.080804   0.014366   \n",
       "4  0.134099   0.029783   0.025864   0.023233   0.007588   0.035305   0.007227   \n",
       "\n",
       "   Class11.4  Class11.5  Class11.6  \n",
       "0   0.003716   0.003290   0.119412  \n",
       "1   0.002573   0.002770   0.052434  \n",
       "2   0.001892   0.002115   0.039099  \n",
       "3   0.002274   0.002515   0.106084  \n",
       "4   0.002174   0.002150   0.047553  "
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "submission = pd.DataFrame(predicted_test, columns=labels.columns.values[1:])\n",
    "submission.insert(0, 'GalaxyID', test_IDs)\n",
    "submission.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(79975, 38)"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "submission.to_csv(WEIGHTS_DIR + RUN_NAME + '\\\\submission.csv', index=False)\n",
    "submission.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Best result so far (run 04.01.19_1): <b>0.11013</b> mean squared loss on kaggle -> <b>Rank 76/326</b>\n",
    "\n",
    "Best result from other Kagglers: 0.0749 msl\n",
    "\n",
    "All zeros benchmark: 0.27210 msl\n",
    "\n",
    "Central pixel benchmark: 0.16234 msl"
   ]
  },
  {
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}