VTB_classifier_playground.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "VTB_classifier_playground",
      "provenance": [],
      "collapsed_sections": [
        "NNJ1OZeRjE6f",
        "c-A_lSAmjwEb",
        "YOxkH3ocmxzx"
      ],
      "authorship_tag": "ABX9TyNDs4uV9oSwXR/NFY8Tqh39",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/pren1/VAST/blob/master/VTB_classifier_playground.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "d9A4Tbg0gN8r",
        "colab_type": "code",
        "outputId": "2e353bd9-6902-4bc2-cea3-f342c2774354",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 399
        }
      },
      "source": [
        "!rm -rf VAST/\n",
        "!git clone https://github.com/pren1/VAST.git\n",
        "!pip3 install soundfile\n",
        "!pip3 install librosa"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Cloning into 'VAST'...\n",
            "remote: Enumerating objects: 94, done.\u001b[K\n",
            "remote: Counting objects: 100% (94/94), done.\u001b[K\n",
            "remote: Compressing objects: 100% (68/68), done.\u001b[K\n",
            "remote: Total 94 (delta 40), reused 50 (delta 18), pack-reused 0\u001b[K\n",
            "Unpacking objects: 100% (94/94), done.\n",
            "Requirement already satisfied: soundfile in /usr/local/lib/python3.6/dist-packages (0.10.3.post1)\n",
            "Requirement already satisfied: cffi>=1.0 in /usr/local/lib/python3.6/dist-packages (from soundfile) (1.14.0)\n",
            "Requirement already satisfied: pycparser in /usr/local/lib/python3.6/dist-packages (from cffi>=1.0->soundfile) (2.20)\n",
            "Requirement already satisfied: librosa in /usr/local/lib/python3.6/dist-packages (0.6.3)\n",
            "Requirement already satisfied: numpy>=1.8.0 in /usr/local/lib/python3.6/dist-packages (from librosa) (1.18.4)\n",
            "Requirement already satisfied: resampy>=0.2.0 in /usr/local/lib/python3.6/dist-packages (from librosa) (0.2.2)\n",
            "Requirement already satisfied: scipy>=1.0.0 in /usr/local/lib/python3.6/dist-packages (from librosa) (1.4.1)\n",
            "Requirement already satisfied: six>=1.3 in /usr/local/lib/python3.6/dist-packages (from librosa) (1.12.0)\n",
            "Requirement already satisfied: joblib>=0.12 in /usr/local/lib/python3.6/dist-packages (from librosa) (0.14.1)\n",
            "Requirement already satisfied: numba>=0.38.0 in /usr/local/lib/python3.6/dist-packages (from librosa) (0.48.0)\n",
            "Requirement already satisfied: decorator>=3.0.0 in /usr/local/lib/python3.6/dist-packages (from librosa) (4.4.2)\n",
            "Requirement already satisfied: audioread>=2.0.0 in /usr/local/lib/python3.6/dist-packages (from librosa) (2.1.8)\n",
            "Requirement already satisfied: scikit-learn!=0.19.0,>=0.14.0 in /usr/local/lib/python3.6/dist-packages (from librosa) (0.22.2.post1)\n",
            "Requirement already satisfied: llvmlite<0.32.0,>=0.31.0dev0 in /usr/local/lib/python3.6/dist-packages (from numba>=0.38.0->librosa) (0.31.0)\n",
            "Requirement already satisfied: setuptools in /usr/local/lib/python3.6/dist-packages (from numba>=0.38.0->librosa) (46.1.3)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zAOEgEC3gY2-",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from VAST.util import *\n",
        "import pdb\n",
        "import pprint\n",
        "from tqdm import tqdm\n",
        "import numpy as np\n",
        "import IPython\n",
        "import soundfile as sf\n",
        "import matplotlib.pyplot as plt\n",
        "import librosa\n",
        "import os"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CFkmpVp7KwZ1",
        "colab_type": "text"
      },
      "source": [
        "### Prepare dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wTdWN9Tmgacc",
        "colab_type": "code",
        "outputId": "14438f8a-aaec-4970-f530-6ea85154516c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/gdrive')\n",
        "\n",
        "# !gdown https://drive.google.com/uc?id=14AT3AGwNtPMRaFGN7VE2rf2BHrN4ImSd\n",
        "# !gdown https://drive.google.com/uc?id=16124JngZYtz-xt5p_KBdsuo4lXN-TDCW"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/gdrive; to attempt to forcibly remount, call drive.mount(\"/content/gdrive\", force_remount=True).\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WOGaNMM3SCuK",
        "colab_type": "text"
      },
      "source": [
        "### Set critical path and parameters"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4s-w0yI4SHdw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "regenerate_data_to_folder = False\n",
        "\n",
        "debug_name_list = ['宝鐘マリン', '白上フブキ', '天音かなた', '夏色まつり']\n",
        "save_path = \"gdrive/My Drive/Free_RAM_VAST/\"\n",
        "sub_size = 512\n",
        "train_ratio = 0.8\n",
        "batch_size = 2048\n",
        "dim = 8000 \n",
        "sample_rate = 8000\n",
        "n_classes = 4\n",
        "'How many sub_batch should we use for each vtb?'\n",
        "sub_value = int(batch_size/(n_classes * sub_size))\n",
        "dropout_rate = 0.1"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2Uc2ZwwtXAUL",
        "colab_type": "text"
      },
      "source": [
        "### We save the data as some batches"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "86Zsq_l_qIN2",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def split_data_into_folders(debug_name_list, save_path, Batch_size):\n",
        "  'Save data in a special way'\n",
        "  data, label = load_data()\n",
        "  label_list = set(label)\n",
        "  assert len(label_list) == len(debug_name_list)\n",
        "  create_folders(save_path)\n",
        "  whole_label = np.asarray(label)\n",
        "  for single in label_list:\n",
        "    whole_index = np.where(label == single)[0]\n",
        "    save_split_small_batch(data[whole_index], whole_label[whole_index], save_path + f\"{debug_name_list[single]}/\", Batch_size)\n",
        "\n",
        "def load_data():\n",
        "  'Just load data, be careful about your RAM'\n",
        "  data_folder = \"gdrive/My Drive/CLIPPED_VAST/\"\n",
        "  label_folder = \"gdrive/My Drive/CLIPPED_VAST/\"\n",
        "  data = []\n",
        "  label = []\n",
        "  for i in tqdm(range(6)):\n",
        "    data.append(load_data_array_from_npy(data_folder + f\"dataset_{i}.npy\"))\n",
        "    label.extend(load_data_array_from_npy(label_folder + f\"labelset_{i}.npy\"))\n",
        "\n",
        "  # data.append(load_data_array_from_npy(data_folder + f\"add_data.npy\"))\n",
        "  # label.extend(load_data_array_from_npy(label_folder + f\"add_label.npy\"))\n",
        "  data = np.vstack(data)\n",
        "  return data, label\n",
        "\n",
        "def save_split_small_batch(target_data, target_label, save_path, Batch_size = 128):\n",
        "  'save data, each npy file contains Batch_size seq of data'\n",
        "  save_index = 0\n",
        "  create_folders(save_path)\n",
        "  for index in tqdm(range(0, len(target_label), Batch_size)):\n",
        "    data_clip = target_data[index:index + Batch_size]\n",
        "    label_clip = target_label[index:index + Batch_size]\n",
        "    if len(data_clip) == Batch_size:\n",
        "      'save data'\n",
        "      save_data_array_as_npy(data_clip, save_path + f'data_{save_index}')\n",
        "      save_data_array_as_npy(label_clip, save_path + f'label_{save_index}')\n",
        "      save_index += 1"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nV5k-iHvNj2r",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "if regenerate_data_to_folder:\n",
        "  split_data_into_folders(debug_name_list = debug_name_list, save_path = save_path, Batch_size = sub_size)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NNJ1OZeRjE6f",
        "colab_type": "text"
      },
      "source": [
        "### Take a look at the loaded data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vOGoPzW_jEGN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# for i in range(3):\n",
        "#   plt.plot(data[i], color = 'orange', label='read in sample data')\n",
        "#   plt.title(f\"Label: {label[i]}\")\n",
        "#   plt.show()\n",
        "#   IPython.display.display(IPython.display.Audio(np.asarray(data[i]).T, rate=8000))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "c-A_lSAmjwEb",
        "colab_type": "text"
      },
      "source": [
        "### See if the preprocess part works correct"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Eb9zkyCHjgNq",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# from keras.models import Sequential\n",
        "# from kapre.time_frequency import Melspectrogram, Spectrogram\n",
        "# from kapre.utils import Normalization2D\n",
        "\n",
        "# melspecModel = Sequential() \n",
        "\n",
        "# melspecModel.add(Melspectrogram(n_dft=1024, n_hop=128, input_shape=(1, data.shape[-1]),\n",
        "#                          padding='same', sr=sample_rate, n_mels=80,\n",
        "#                          fmin=40.0, fmax=sample_rate/2, power_melgram=1.0,\n",
        "#                          return_decibel_melgram=True, trainable_fb=False,\n",
        "#                          trainable_kernel=False,\n",
        "#                          name='mel_stft') )\n",
        "\n",
        "# melspecModel.add(Normalization2D(int_axis=0))\n",
        "\n",
        "# melspecModel.summary()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wOzBT3d6mBPs",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# test_input = np.expand_dims(np.expand_dims(data[0], 0), 0)\n",
        "# print(f\"test_input shape: {test_input.shape}\")\n",
        "# melspec = melspecModel.predict(test_input)\n",
        "# print(f\"melspec shape: {melspec.shape}\")\n",
        "\n",
        "# plt.figure(figsize=(17,6))\n",
        "# plt.pcolormesh(melspec[0,:,:,0])\n",
        "\n",
        "# plt.title('Spectrogram visualization')\n",
        "# plt.ylabel('Frequency')\n",
        "# plt.xlabel('Time')\n",
        "\n",
        "# plt.show()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YOxkH3ocmxzx",
        "colab_type": "text"
      },
      "source": [
        "### Check if mel spectrogram matches the one computed with librosa"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "W19dlYOTmwVK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# librosa_melspec = librosa.feature.melspectrogram(y=data[0], sr=sample_rate, n_fft=1024,\n",
        "#                                                  hop_length=128, power=1.0, #window='hann',\n",
        "#                                                  n_mels=80, fmin=40.0, fmax=sample_rate/2)\n",
        "# S_dB = librosa.power_to_db(librosa_melspec, ref=np.max)\n",
        "# plt.figure(figsize=(17,6))\n",
        "# plt.pcolormesh(S_dB)\n",
        "\n",
        "# plt.title('Spectrogram visualization - librosa')\n",
        "# plt.ylabel('Frequency')\n",
        "# plt.xlabel('Time')\n",
        "\n",
        "# plt.show()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-Qanl5IE1UZ-",
        "colab_type": "text"
      },
      "source": [
        "### A toy CNN + LSTM + attention classifier "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_7YclNci1ZUr",
        "colab_type": "code",
        "outputId": "8b1efd03-5cf8-41fc-c999-3554fedd9348",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "from keras import layers as L\n",
        "import keras.backend as K\n",
        "from keras.models import Model\n",
        "from keras.callbacks import LearningRateScheduler\n",
        "import math\n",
        "import keras\n",
        "from kapre.time_frequency import Melspectrogram, Spectrogram\n",
        "from kapre.utils import Normalization2D\n",
        "\n",
        "'define model'\n",
        "def AttRNNSpeechModel(nCategories, dropout_rate, samplingrate=8000,\n",
        "                      inputLength=8000, rnn_func=L.LSTM):\n",
        "    # simple LSTM\n",
        "    sr = samplingrate\n",
        "    iLen = inputLength\n",
        "    dropout_rate = dropout_rate\n",
        "\n",
        "    inputs = L.Input((inputLength,), name='input')\n",
        "\n",
        "    x = L.Reshape((1, -1))(inputs)\n",
        "\n",
        "    m = Melspectrogram(n_dft=1024, n_hop=128, input_shape=(1, iLen),\n",
        "                       padding='same', sr=sr, n_mels=80,\n",
        "                       fmin=40.0, fmax=sr / 2, power_melgram=1.0,\n",
        "                       return_decibel_melgram=True, trainable_fb=False,\n",
        "                       trainable_kernel=False,\n",
        "                       name='mel_stft')\n",
        "    m.trainable = False\n",
        "\n",
        "    x = m(x)\n",
        "\n",
        "    x = Normalization2D(int_axis=0, name='mel_stft_norm')(x)\n",
        "\n",
        "    # note that Melspectrogram puts the sequence in shape (batch_size, melDim, timeSteps, 1)\n",
        "    # we would rather have it the other way around for LSTMs\n",
        "\n",
        "    x = L.Permute((2, 1, 3))(x)\n",
        "\n",
        "    x = L.Conv2D(10, (5, 1), activation='relu', padding='same')(x)\n",
        "    x = L.BatchNormalization()(x)\n",
        "    x = L.Dropout(dropout_rate)(x)\n",
        "    \n",
        "    x = L.Conv2D(1, (5, 1), activation='relu', padding='same')(x)\n",
        "    x = L.BatchNormalization()(x)\n",
        "    x = L.Dropout(dropout_rate)(x)\n",
        "\n",
        "    # x = Reshape((125, 80)) (x)\n",
        "    # keras.backend.squeeze(x, axis)\n",
        "    x = L.Lambda(lambda q: K.squeeze(q, -1), name='squeeze_last_dim')(x)\n",
        "\n",
        "    x = L.Bidirectional(rnn_func(64, return_sequences=True)\n",
        "                        )(x)  # [b_s, seq_len, vec_dim]\n",
        "    \n",
        "    x = L.Dropout(dropout_rate)(x)                    \n",
        "    x = L.Bidirectional(rnn_func(64, return_sequences=True)\n",
        "                        )(x)  # [b_s, seq_len, vec_dim]\n",
        "\n",
        "    xFirst = L.Lambda(lambda q: q[:, -1])(x)  # [b_s, vec_dim]\n",
        "    query = L.Dense(128)(xFirst)\n",
        "\n",
        "    # dot product attention\n",
        "    attScores = L.Dot(axes=[1, 2])([query, x])\n",
        "    attScores = L.Softmax(name='attSoftmax')(attScores)  # [b_s, seq_len]\n",
        "\n",
        "    # rescale sequence\n",
        "    attVector = L.Dot(axes=[1, 1])([attScores, x])  # [b_s, vec_dim]\n",
        "\n",
        "    x = L.Dense(64, activation='relu')(attVector)\n",
        "    x = L.Dense(32)(x)\n",
        "\n",
        "    output = L.Dense(nCategories, activation='softmax', name='output')(x)\n",
        "\n",
        "    model = Model(inputs=[inputs], outputs=[output])\n",
        "\n",
        "    return model\n",
        "\n",
        "'learning rate decay'\n",
        "def step_decay(self, epoch):\n",
        "    initial_lrate = 0.01\n",
        "    drop = 0.5\n",
        "    epochs_drop = 2.0\n",
        "    lrate = initial_lrate * math.pow(drop, math.floor((1 + epoch) / epochs_drop))\n",
        "    return lrate    "
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Rv8LeJyK2Vqk",
        "colab_type": "code",
        "outputId": "e2c9f096-fbb0-47d1-e123-71b611d7da1a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "model = AttRNNSpeechModel(n_classes, dropout_rate=dropout_rate, samplingrate=sample_rate, inputLength=dim)\n",
        "model.compile(optimizer='adam', loss=['categorical_crossentropy'], metrics=['categorical_accuracy'])\n",
        "model.summary()\n",
        "lrate = LearningRateScheduler(step_decay)"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "tracking <tf.Variable 'mel_stft/real_kernels:0' shape=(1024, 1, 1, 513) dtype=float32> dft_real_kernels\n",
            "tracking <tf.Variable 'mel_stft/imag_kernels:0' shape=(1024, 1, 1, 513) dtype=float32> dft_imag_kernels\n",
            "tracking <tf.Variable 'mel_stft/Variable:0' shape=(513, 80) dtype=float32> freq2mel\n",
            "Model: \"model_1\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input (InputLayer)              (None, 8000)         0                                            \n",
            "__________________________________________________________________________________________________\n",
            "reshape_1 (Reshape)             (None, 1, 8000)      0           input[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "mel_stft (Melspectrogram)       (None, 80, 63, 1)    1091664     reshape_1[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "mel_stft_norm (Normalization2D) (None, 80, 63, 1)    0           mel_stft[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "permute_1 (Permute)             (None, 63, 80, 1)    0           mel_stft_norm[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_1 (Conv2D)               (None, 63, 80, 10)   60          permute_1[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_1 (BatchNor (None, 63, 80, 10)   40          conv2d_1[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "dropout_1 (Dropout)             (None, 63, 80, 10)   0           batch_normalization_1[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_2 (Conv2D)               (None, 63, 80, 1)    51          dropout_1[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_2 (BatchNor (None, 63, 80, 1)    4           conv2d_2[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "dropout_2 (Dropout)             (None, 63, 80, 1)    0           batch_normalization_2[0][0]      \n",
            "__________________________________________________________________________________________________\n",
            "squeeze_last_dim (Lambda)       (None, 63, 80)       0           dropout_2[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "bidirectional_1 (Bidirectional) (None, 63, 128)      74240       squeeze_last_dim[0][0]           \n",
            "__________________________________________________________________________________________________\n",
            "dropout_3 (Dropout)             (None, 63, 128)      0           bidirectional_1[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "bidirectional_2 (Bidirectional) (None, 63, 128)      98816       dropout_3[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "lambda_1 (Lambda)               (None, 128)          0           bidirectional_2[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "dense_1 (Dense)                 (None, 128)          16512       lambda_1[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "dot_1 (Dot)                     (None, 63)           0           dense_1[0][0]                    \n",
            "                                                                 bidirectional_2[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "attSoftmax (Softmax)            (None, 63)           0           dot_1[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "dot_2 (Dot)                     (None, 128)          0           attSoftmax[0][0]                 \n",
            "                                                                 bidirectional_2[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "dense_2 (Dense)                 (None, 64)           8256        dot_2[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "dense_3 (Dense)                 (None, 32)           2080        dense_2[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "output (Dense)                  (None, 4)            132         dense_3[0][0]                    \n",
            "==================================================================================================\n",
            "Total params: 1,291,855\n",
            "Trainable params: 200,169\n",
            "Non-trainable params: 1,091,686\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "O4ewvrg2cO2t",
        "colab_type": "text"
      },
      "source": [
        "### Define data generator"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YW-LGoxbTauh",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class reiterate_list():\n",
        "  'Iterate over the id list provided'\n",
        "  def __init__(self, ID_list, return_number, is_shuffle=True):\n",
        "    self.ID_list = np.asarray(ID_list)\n",
        "    self.list_length = len(self.ID_list)\n",
        "    self.return_number = return_number\n",
        "    self.order = np.arange(self.list_length)\n",
        "    self.shuffled = False\n",
        "    self.cursor = 0\n",
        "    if is_shuffle:\n",
        "        self.shuffle()\n",
        "\n",
        "  def shuffle(self):\n",
        "      np.random.shuffle(self.order)\n",
        "      self.cursor = 0\n",
        "\n",
        "  def next(self):\n",
        "      if self.shuffled==True:\n",
        "          self.shuffled = False\n",
        "      if self.cursor + self.return_number > self.list_length:\n",
        "          self.shuffle()\n",
        "          self.shuffled = True\n",
        "      idx = self.order[self.cursor:self.cursor + self.return_number]\n",
        "      self.cursor += self.return_number\n",
        "      return self.ID_list[idx]\n",
        "\n",
        "def build_train_test_id_dicts():\n",
        "  'randomly assign each id to either train set or test set'\n",
        "  first_level_folders = [f.path.split('/')[-1] for f in os.scandir(save_path) if f.is_dir()]\n",
        "  assert first_level_folders == debug_name_list, \"Saved folder inconsistent, check your code\"\n",
        "  train_dict = {}\n",
        "  test_dict = {}\n",
        "  for single_vtb in debug_name_list:\n",
        "    all_files = files_in_target_folder(save_path + single_vtb, '.npy')\n",
        "    range_index = np.arange(len(all_files)//2)\n",
        "    np.random.shuffle(range_index)\n",
        "    train_dict[single_vtb] = range_index[:int(train_ratio * len(range_index))]\n",
        "    test_dict[single_vtb] = range_index[int(train_ratio * len(range_index)):]\n",
        "  return train_dict, test_dict\n",
        "\n",
        "def DataGenerator(target_dict):\n",
        "  id_dict = {}\n",
        "  for single_vtb in target_dict:\n",
        "    id_dict[single_vtb] = reiterate_list(target_dict[single_vtb], sub_value)\n",
        "  \n",
        "  # Create empty arrays to contain batch of features and labels#\n",
        "  X = np.empty((batch_size, dim))\n",
        "  y = np.empty((batch_size), dtype=int)\n",
        "  while True:\n",
        "    'get a bunch of ids, then read in them'\n",
        "    data_list = []\n",
        "    label_list = []\n",
        "    for single_vtb in id_dict:\n",
        "      target_id_list = id_dict[single_vtb].next()\n",
        "      data_list.extend([save_path + single_vtb + \"/\" + f'data_{ID}.npy' for ID in target_id_list])\n",
        "      label_list.extend([save_path + single_vtb + \"/\" + f'label_{ID}.npy' for ID in target_id_list])\n",
        "      \n",
        "    'Then, fill in the read in results to X and y'\n",
        "    for index, (data_path, label_path) in enumerate(zip(data_list, label_list)):\n",
        "      X[index * sub_size:(index + 1) * sub_size] = load_data_array_from_npy(data_path)\n",
        "      y[index * sub_size:(index + 1) * sub_size] = load_data_array_from_npy(label_path)\n",
        "    'To one-hot'\n",
        "    yield X, keras.utils.to_categorical(y, num_classes=n_classes)\n",
        "\n",
        "def get_step_from_dictionary(input_dict):\n",
        "  'When will the whole dataset run out?'\n",
        "  res = []\n",
        "  for vtb in input_dict:\n",
        "    res.append(len(input_dict[vtb])//sub_value)\n",
        "  return np.max(res)  \n",
        "\n",
        "def generate_generator():\n",
        "  train_dict, test_dict = build_train_test_id_dicts()\n",
        "  train_generator = DataGenerator(train_dict)\n",
        "  test_generator = DataGenerator(test_dict)\n",
        "  return train_generator, test_generator, train_dict, test_dict"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "I7Ts922Dgled",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "train_generator, test_generator, train_dict, test_dict = generate_generator()\n",
        "train_steps = get_step_from_dictionary(train_dict)\n",
        "validation_steps = get_step_from_dictionary(test_dict)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5FWfXquew1xB",
        "colab_type": "text"
      },
      "source": [
        "### Calculate class weights"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OnPy2sgRvjWn",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "f4ea583f-fd9f-4021-d94c-1aa17314ea9b"
      },
      "source": [
        "from sklearn.utils import class_weight\n",
        "def build_label_for_class_weight_calculation(input_dict):\n",
        "  label = []\n",
        "  for index, vtb in enumerate(input_dict):\n",
        "    label.extend([index] * len(input_dict[vtb]))\n",
        "  return label  \n",
        "\n",
        "label = build_label_for_class_weight_calculation(train_dict)\n",
        "\n",
        "class_weights = class_weight.compute_class_weight('balanced',\n",
        "                                                 np.unique(label),\n",
        "                                                 label)\n",
        "class_weights"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([0.82621951, 1.27830189, 0.82621951, 1.25462963])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "N0dQbE_fLK2h",
        "colab_type": "text"
      },
      "source": [
        "### Train model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bFxHhBQo4I8o",
        "colab_type": "code",
        "outputId": "b2381328-b722-4cf9-9cfc-45edff41b332",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 619
        }
      },
      "source": [
        "# results = model.fit(x = data[train_idx], y = to_categorical(label[train_idx]),\n",
        "# \t\t\t\tepochs=30,\n",
        "# \t\t\t\tbatch_size=2048,\n",
        "# \t\t\t\tshuffle=True,\n",
        "# \t\t\t\tvalidation_data=(data[test_idx], to_categorical(label[test_idx])),\n",
        "# \t\t\t\tcallbacks=[lrate],\n",
        "# \t\t\t\tclass_weight=class_weights\n",
        "# \t\t\t\t)\n",
        "\n",
        "# Train model on dataset\n",
        "results = model.fit_generator(generator=train_generator,\n",
        "                    epochs=20,\n",
        "\t\t\t\t\t\t\t\t\t\tvalidation_data=test_generator,\n",
        "\t\t\t\t\t\t\t\t\t\tvalidation_steps = validation_steps,\n",
        "\t\t\t\t\t\t\t\t\t\tsteps_per_epoch = train_steps,\n",
        "\t\t\t\t\t\t\t\t\t\tshuffle=True,\n",
        "\t\t\t\t\t\t\t\t\t\tcallbacks=[lrate],\n",
        "\t\t\t\t\t\t\t\t\t\tclass_weight=class_weights\n",
        "                    )"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/15\n",
            "82/82 [==============================] - 264s 3s/step - loss: 0.0232 - categorical_accuracy: 0.9917 - val_loss: 0.1008 - val_categorical_accuracy: 0.9750\n",
            "Epoch 2/15\n",
            "82/82 [==============================] - 226s 3s/step - loss: 0.0284 - categorical_accuracy: 0.9901 - val_loss: 0.1027 - val_categorical_accuracy: 0.9702\n",
            "Epoch 3/15\n",
            "82/82 [==============================] - 251s 3s/step - loss: 0.0233 - categorical_accuracy: 0.9919 - val_loss: 0.0712 - val_categorical_accuracy: 0.9712\n",
            "Epoch 4/15\n",
            "82/82 [==============================] - 231s 3s/step - loss: 0.0271 - categorical_accuracy: 0.9905 - val_loss: 0.0804 - val_categorical_accuracy: 0.9734\n",
            "Epoch 5/15\n",
            "53/82 [==================>...........] - ETA: 1:10 - loss: 0.0213 - categorical_accuracy: 0.9925"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/keras/utils/data_utils.py:718: UserWarning: An input could not be retrieved. It could be because a worker has died.We do not have any information on the lost sample.\n",
            "  UserWarning)\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "82/82 [==============================] - 248s 3s/step - loss: 0.0214 - categorical_accuracy: 0.9925 - val_loss: 0.1097 - val_categorical_accuracy: 0.9763\n",
            "Epoch 6/15\n",
            "82/82 [==============================] - 245s 3s/step - loss: 0.0210 - categorical_accuracy: 0.9927 - val_loss: 0.1055 - val_categorical_accuracy: 0.9743\n",
            "Epoch 7/15\n",
            "82/82 [==============================] - 231s 3s/step - loss: 0.0227 - categorical_accuracy: 0.9922 - val_loss: 0.1092 - val_categorical_accuracy: 0.9734\n",
            "Epoch 8/15\n",
            "82/82 [==============================] - 218s 3s/step - loss: 0.0213 - categorical_accuracy: 0.9924 - val_loss: 0.1352 - val_categorical_accuracy: 0.9756\n",
            "Epoch 9/15\n",
            "82/82 [==============================] - 204s 2s/step - loss: 0.0188 - categorical_accuracy: 0.9933 - val_loss: 0.0802 - val_categorical_accuracy: 0.9769\n",
            "Epoch 10/15\n",
            "82/82 [==============================] - 222s 3s/step - loss: 0.0177 - categorical_accuracy: 0.9938 - val_loss: 0.0881 - val_categorical_accuracy: 0.9768\n",
            "Epoch 11/15\n",
            "82/82 [==============================] - 230s 3s/step - loss: 0.0151 - categorical_accuracy: 0.9947 - val_loss: 0.0802 - val_categorical_accuracy: 0.9756\n",
            "Epoch 12/15\n",
            "82/82 [==============================] - 233s 3s/step - loss: 0.0146 - categorical_accuracy: 0.9948 - val_loss: 0.0827 - val_categorical_accuracy: 0.9786\n",
            "Epoch 13/15\n",
            "82/82 [==============================] - 224s 3s/step - loss: 0.0172 - categorical_accuracy: 0.9939 - val_loss: 0.1250 - val_categorical_accuracy: 0.9757\n",
            "Epoch 14/15\n",
            "82/82 [==============================] - 259s 3s/step - loss: 0.0160 - categorical_accuracy: 0.9944 - val_loss: 0.0838 - val_categorical_accuracy: 0.9771\n",
            "Epoch 15/15\n",
            "82/82 [==============================] - 230s 3s/step - loss: 0.0156 - categorical_accuracy: 0.9946 - val_loss: 0.0748 - val_categorical_accuracy: 0.9779\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bFuCpAWPDklM",
        "colab_type": "text"
      },
      "source": [
        "### Training loss/acc curve"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ilHuvXUYAPd6",
        "colab_type": "code",
        "outputId": "d82e8a2f-a117-4846-8849-1424b231867b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 613
        }
      },
      "source": [
        "# summarize history for categorical accuracy\n",
        "plt.plot(results.history['categorical_accuracy'])\n",
        "plt.plot(results.history['val_categorical_accuracy'])\n",
        "plt.title('Categorical accuracy')\n",
        "plt.ylabel('accuracy')\n",
        "plt.xlabel('epoch')\n",
        "plt.legend(['train', 'test'], loc='upper left')\n",
        "plt.show()\n",
        "# summarize history for loss\n",
        "plt.plot(results.history['loss'])\n",
        "plt.plot(results.history['val_loss'])\n",
        "plt.title('model loss')\n",
        "plt.ylabel('loss')\n",
        "plt.xlabel('epoch')\n",
        "plt.legend(['train', 'test'], loc='upper left')\n",
        "plt.show()"
      ],
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZ4AAAEqCAYAAAAlJzZsAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nOzdd3wUdf748dfMbElITwgkgQRSSGihlwgYQPSoSrFxnAd4CHdGQQVURO/Or3C2s3H4U7DcHXd6hyCYU4QoYAlFeu+QQkKHtCV1y8zvj5jVNYS0zW7K5/l48NDMzGfm/SE89r3zqZKmaRqCIAiC4CKyuwMQBEEQWhaReARBEASXEolHEARBcCmReARBEASXEolHEARBcCmReARBEASXEolHEBqRc+fOERcXx9q1a2tddufOncTFxbFz584Ge4YgOINIPIIgCIJLicQjCIIguJRIPIIgCIJLicQjCD+zdOlS4uLiSEtLY/78+fTt25eBAwfyxhtvoGkaly9fJikpiT59+jBo0CA++OCDSvfIzc3lj3/8I4MHDyY+Pp5x48axatWqSteZTCYWLFhA37596devH08//TTXr1+/YVwZGRk89thjDBw4kPj4eCZMmEBKSopT637ixAlmzpxJnz596NWrF7/97W/Zs2ePwzVWq5V33nmHkSNH0qNHDwYMGMC9997L119/bb/m2rVrPPvsswwdOpTu3bszaNAgHnroIU6fPu3UeIWmS+fuAAShMZo7dy5RUVHMmzeP1NRUli9fjp+fH2vWrKFfv37Mnz+fL774gr/+9a9069aNW265BYCysjKmTZtGeno6U6ZMISIigk2bNvHHP/6R/Px8Zs2aBYCmaSQlJbF3717uv/9+YmJi2Lx5M08//XSlWNLS0pg8eTKtW7dmxowZeHl5sXHjRh577DFeffVVxo8fX+/6pqWlMWXKFFq1asWMGTMwGo2sXr2a6dOn849//IP+/fsD8Pbbb7Ns2TLuueceevToQXFxMcePH+fQoUP86le/AmDOnDmcOnWK3/zmN7Rv3568vDx27dpFZmYmnTp1qnesQjOgCYJg97e//U2LjY3VnnnmGfsxq9WqJSYmanFxcdo777xjP15QUKD16NFDmzdvnv3YihUrtNjYWG3t2rUO5adNm6Z1795dy83N1TRN0zZu3KjFxsZq7733nsN1DzzwgBYbG6utWbPGfvzBBx/UxowZo5WUlDjE+uCDD2q33nqrpqqqpmmatmPHDi02NlbbsWPHTeuYnZ1d6RmPPPKI1q1bNy0jI8N+LCcnRxswYIA2ceJE+7Hx48drs2bNqvLeBQUFWmxsrPbBBx/cNAahZRNNbYJwA/fee6/9/xVFoXv37miaxj333GM/7uvrS2RkJOfOnbMf+/777wkMDOSuu+5yKD9t2jTMZjM//PADAKmpqciyzK9//WuH637zm984xJGfn8/27dsZPXo0xcXF5Obm2v/ceuutXL58mYyMjHrV1WazsXXrVoYPH07Hjh3txwMDA5k0aRJHjx7l2rVrAPj4+HD69Okqn+nh4YFer2fXrl3k5+fXKy6h+RJNbYJwA2FhYQ4/+/j4oNfrCQ4OrnS84kMZ4Pz583To0AFFURyui46OBrAnqfPnz9O6dWu8vb0drvv5Bz9AVlYWmqaxdOlSli5desNYc3NziYqKqnnlblC+pKSEyMjISucq7lsR75w5c3jkkUcYNWoUMTExDBkyhHHjxhEfHw+AwWBg/vz5vPrqqwwePJgePXowdOhQxo8fT2hoaJ1jFJoXkXgE4QZkuXJjgCRJN7xWa8AtrVRVBWD69OkMHTr0hte4st+kf//+bNy4kW+//ZatW7eSnJzMihUrmDdvHjNnzrTHOmLECDZv3sz27dt55513WL58OcuWLWPgwIEui1VovETiEQQnateuHceOHcNmszm89aSnpwPQvn17+3Xbt2+nsLDQ4a0nMzPT4X7h4eFAeTPcoEGDGiTmwMBAPD09b9h8VhF3u3bt7Mf8/PyYMGECEyZMoLS0lJkzZ7J06VJ+97vf2escHh7O9OnTmT59OpcuXWL8+PEi8Qh2oo9HEJxo2LBh5Obm8sUXX9iPqarKv/71LwwGg330W2JiIqqq8t///tfhuo8//tjhfkFBQQwcOJDVq1dz+fLlSs/Lzc2td8yKonDrrbfy7bffkpWVZT+en59PcnIy3bt3p3Xr1gDk5eU5lPXw8CAqKoqysjJKS0spKSmhtLTU4ZqQkBCCgoIwmUz1jlVoHsQbjyA40X333ceqVat47rnnOH78OOHh4WzatIkffviBefPmERAQAMBtt91Gnz59eOONNzh//jydOnVi06ZNFBQUVLrn888/z69//Wvuuusu7r33XiIiIsjJyeHgwYOkpaWxcePGesf9+OOPs3XrVqZMmcKUKVMwGo2sWrWK69evs2DBAvt1Y8aMoX///nTv3p2AgABOnjzJp59+yvDhw/Hy8uL48eNMmzbN3gdkMBj4/vvvSUtLu+FQcaFlEolHEJzIaDSyYsUK3njjDb744gtMJhMdOnRg0aJF3HffffbrZFnm3Xff5cUXX+SLL75AkiRuu+02nnrqKSZMmOBwz6ioKNasWcP/+3//j+TkZPLz8wkICKBz58489thjTok7Ojqa//73v7z++uu8//77aJpG9+7dWbx4sX0OD8C0adP45ptv2LFjB6WlpYSGhjJz5kx7/05ISAh33nknP/zwg71eHTt25C9/+YvDiEChZZO0huwZFQRBEIRfEH08giAIgku5NfFkZmYyY8YMevfuTUJCAosWLaKkpKTacuvXr2f27NkkJiYSFxfHhx9+WG2ZxYsXExcXxwsvvOCM0AVBEIQ6clviMZlMTJ06laKiIpYsWcKCBQtYt24dCxcurLZsSkoK2dnZDBs2rEbPOnHiBGvWrKk0WU8QBEFwPbcNLli5ciUmk4nk5GQCAwOB8mGd8+fPJykp6aaT4t566y37BL9PPvnkps/RNI3nn3+eGTNmiB0XBUEQGgG3vfGkpqaSkJBgTzoAI0eOxGAwkJqaetOyN5pVXpU1a9Zw5coVHnrooTrHKgiCIDiP2xJPWloaMTExDscMBgMRERH22dL1lZ+fz2uvvcYzzzyDh4eHU+4pCIIg1I/bmtpMJhO+vr6Vjvv6+t5wEl1dvPHGG3Tp0oU77rij3vfKyytCVWs/8jwoyJucnMJ6P7+xaa71guZbN1Gvpqcp102WJQICvG54rtlOID106BCfffYZycnJTrlfVX+BNREU1DwHNTTXekHzrZuoV9PTHOvmtsTj6+t7w7WbTCZTvZZ4r7B48WImTpxIcHCw/TmqqmKxWDCZTHh5eVVauv5mcnIK6/TGExzsw9WrN97OuClrrvWC5ls3Ua+mpynXTZalKpOm2/p4oqOjSUtLczhmNpvJyspySuJJT0/nk08+oX///vY/Fy9eZNWqVfTv359Tp07V+xmCIAhC7bntjScxMZF3332XvLw8+8KJGzduxGw2V7nvSG0sW7YMm83mcGzu3Ln06tWLqVOnEhERUe9nCIIgCLXntsQzefJkPvroI5KSkkhKSiInJ4eXX36ZMWPGOIx2W7hwIcnJyRw7dsx+7MyZM5w5c8b+86lTp0hJSQFg1KhRAPTr16/SM41GI23bthV7ggiCILiRW/t4VqxYweLFi5k9ezZGo5GxY8fy5JNPOlynqmqlN5cNGzbw9ttv239OTk62DyI4efJkwwcvCIIg1JlYnbqGqhtcUFJSRGFhPjab1eG4LMv27YubE1fXS1F0eHv74+lZ99GFNdWUO3RvRtSr6WnKdbvZ4IJmO5zalUpKirh+PQ9//2D0egOSJNnP6XQyVmvzSzyurJemaVgsZvLzrwK4JPk0BeZDX2E5tYVWd7+AJImF5oWmQ/xrdYLCwnz8/YMxGIwOSUdwDkmSMBiM+PsHU1iY7+5wGgWtrIiyvcmouedQ8867OxxBqBWReJzAZrOi1xvcHUazp9cbKjVltlTmQylgKd9CxHZR9GsKTYtIPE4i3nQanvg7LqeWXsd8ZCO6qAFIXoHYLoo5aULTIhKPIDQx5gPrwVKGoe8ElNA4bBdPIsYICU2JSDyCXWrqd6xdu9qp93z00Vk89dTjTr1nS6YWF2A5uhldTAJKQBhKaBxaSQGa6Yq7QxOEGhOj2gS7LVu+48SJY0yadK/T7jlv3gIURXy/cRbzgS9BtWLsOx4AJSQWKO/nkf3aujM0Qagx8Ykg1IqmaZSVldX4+sjIKCIiOjZcQC2IWpSH5fg36DoNRvYLAUD2D0Xy8MF6SQwwEJoO8cYjAPCXvzzPhg3rABgypHy5odGjxwFw4sQxZs+ey7JlS0lPT2PBgj9y++13sHTpW+zevZPLly/h7x9Av34DeOSRx/D19bPf99FHZ9GqVSteffUtAD78cDkrV37E8uX/5PXXX+bEiWOEhIQyc+bDDBs2wsW1blrM+9eBqmHsc5f9mCRJKCGxYoCB0KSIxCMAMH36Q+Tn53H2bCZ/+tNiAAICAvjnPz/g2rVr/PWvLzFt2u8IC2tHUFBrysrKsFqtPPTQwwQGBnLlyhU++ugfzJs3m/ff/9dNn2W1Wnn++YVMmnQf06bN4NNPV/LnPy9k5crPCA0Nc0V1mxy1MAfLie/Rx92K7BvscE4JjcOauRe1MBfZO7CKOwhC4yESTwPZdvgiWw9dRJLA1QOOhvQIZXB8aK3KtGvXHn//AC5dukj37vEO565fN/Hqq28SH9/Tfkynk3nyyYX2n61WK5GRkTz44G84efIEcXGdq3yWxWLh979/lMGDbwUgLq4zd901ki1bvuO++6bUKu6WwrzvcwAMfe6sdE4J/bGf59Ip5JgEl8YlCHUhEo9QLT8/P4ekUyEl5UtWrfoP2dnZlJQU249nZ5+9aeKRZZn+/X9aIdzPzx9//wCuXBEjs25ENV3BcnIr+q7DkL2DKp2XAyNA74Ht4kn0IvEITYBIPA1kcHz5W0dzWKstIKDyh913333D4sV/5s47JzBz5sP4+vpTWHidefNmYzabb3o/o9GIweC40oNer8dsrvmghZakbN/nIMsYeo274XlJlsv7ecQAA6GJEIlHqNaNFgz45ptNxMTE8vTTz9mPnThx3IVRtQxq/iWsp7eh7/4rZK+AKq9TQmMx7zqEWnod2cPHhREKQu2J4dSCnU6nr/ZtpUJZWRl6vd7h2MaNGxoirBatbF8yKHoMvcbe9DolJA4o7+cRhMZOJB7BrmPHjly8eIGvv97AiRPHuHjxQpXX9u8/kOPHj/Lhh8vZvXsn7767lK1bU10YbfNnyz2P9cxODN1uR/b0vem1SnBHUPRiWLXQJIimNsFu3LjxHDt2lCVLXqOgoMA+j+dGJk68m/Pnz/O//61l5cqP6du3H4sXv8KDD/7GhRE3b+Z9yaA3ou85utprJUWP0iZarFTdyKklJiRFh2Ro5e5Q3ErsQFpDN9uB9NKls4SEdLjhueYwuOBG3FWvm/1dO0tj2PXRlpNF8Zo/Yeh9J8b+d9eoTNmezzDv/xzvae8gGTwrnW8M9WoITaVemqZRvPpZZL+2eI58rEZlmkrdbuRmO5CKpjZBaITMez4DgyeGHqNqXEYJiQVNw3b5TANGJtSVmncONf8C1uxDaGVF7g7HrUTiEYRGxnY1A+vZ/Rh6jEIy1nybb6VtDEiKaG5rpKyZ+8r/R7VhzTro3mDczK2JJzMzkxkzZtC7d28SEhJYtGgRJSUl1ZZbv349s2fPJjExkbi4OD788MNK16Snp/PCCy8wZswYevXqxfDhw1m4cCFXr15tiKoIgtOU7fkMjF4Yuv+qVuUkvRE5uIMY2dZIWTP3IbeJRmrljzVjr7vDcSu3JR6TycTUqVMpKipiyZIlLFiwgHXr1rFw4cJqy6akpJCdnc2wYcOqvGb79u3s3r2b++67j+XLl/P444+ze/du7r//foqKWvZrrtB42S6fwZZ9CEPP0Tfsp6mOEhKH7Uo6mrVmw+IF11CvX0O9dhZ9ZF90HftizT6MZmm5E6bdNqpt5cqVmEwmkpOTCQwsX9hQURTmz59PUlISnTp1qrLsW2+9hSyX58xPPvnkhteMGTOG3/zmNw7bJcfFxTF+/Hi+/vprJk6c6MTaCIJzlO35DMnDB0O32+tUXhcai+XQBmxXM9CFxjk5OqGurGf3A6Dr2Ae1MBfLsc1Yzx1GH9nPzZG5h9veeFJTU0lISLAnHYCRI0diMBhITb35fJCKpHMzgYGBDkkHyhOPoihiTTChUbJePInt/FEMvcYg6T3qdI/yjeEk0c/TyFgz9yEHtEP2C0EJjUMyemPN2OPusNzGbYknLS2NmJgYh2MGg4GIiAjS09Mb5Jn79+/HZrMRHR3dIPcXhLrSNA3znrVInn7ou95W5/tIRi/kwPYi8TQiWmkhtosn0XXsA4AkK+g69sZ69iCazeLm6NzDrX08vr6VZ2P7+vpSUFDg9OdZLBZefPFFIiMjb9o3JAjuYLtwHNvFkxh6j0PSGet1LyU0FtvlM2iq1UnRCfVhzToAmoquY1/7MV1kX7CUYDt/zI2RuU+LWblg0aJFnD59mo8++gidrvbVrmoiFMCVKzI6XdU5/GbnmjJ31EuWZYKDG34RTFc8o4KmaVz48n8oPkGE3joOWWeovtBNFMb25MrRzfhar+HRzrGv1JX1cqXGXK9L3x1E8W1N2y7d7c3/WkACmd94ort4iOC+Q25avjHXra7clnh8fX0xmUyVjptMJqKiopz6rLfffptPP/2UpUuXEh8fX32BG7jZygWqqlY5i1+sXOBcqqo2+ExuV88Wt2Yfouz8SYxDppKTVwbUb7ST6hUBQM6J/RgMIfbjTXkW/M005npp1jKK0w6i75zItWuFDueU8J4UntiJ1n8KkqzcsHxjrlt1GuXKBdHR0aSlpTkcM5vNZGVlOTXxfPzxxyxdupTnn3+eESNGOO2+zVFq6nesXbva6fe9fv06H364nIyMhum7a8o0TSsfyebTGn1colPuKbfyR/JrKxYMbQSs546AzWzv3/k5XWRftLLCFjnvym2JJzExkR07dpCXl2c/tnHjRsxmM0OHDnXKM7788ksWL17MnDlzuO+++5xyz+Zsy5bv+Owz5yeewsLr/OMf75OZKRLPL9nOHkC9moGx911IivMaIHQhcVgvnULTmt/bdlNizdgHRi/79uQ/pwvvAYoBa3rLG93mtsQzefJkfHx8SEpKYsuWLSQnJ7No0SLGjBnjMNpt4cKFdO3a1aHsmTNnSElJISUlBYBTp045/Aywa9cunn76afr168fgwYM5cOCA/U9WVpZrKikIN6FpKmV71yL5tkEXO9ip91ZC46CsCDXvvFPvK9ScptqwZh1AF9ELSa78pULSG9GFx2PN3NviviC4tY9nxYoVLF68mNmzZ2M0Ghk7dixPPvmkw3WqqmKz2RyObdiwgbffftv+c3JyMsnJyQCcPFk+jHTnzp1YLBZ27drF/fff71B+4sSJvPzyyw1RrSbrL395ng0b1gEwZEj5pLbRo8fx7LPPc+zYEd5//12OHDkMwIABA5k37ykCA4Pt5f/zn3/x+eefceXKZTw9PYmKimHevAUYjUbuvfcuAP74xwX261ev/pzQ0DBXVa9RsmbsRc3JxmPYzCrb+OuqfD4P2C6eQgkMd+q9hZqxXToFZUU3bGaroIvsizVzL+qV9PK19loIsS1CDTX3bRHOnz/HW2/9lbNnM/nTnxYDEBAQQEFBPo8+Oot+/QZy110TsFqt/P3v76FpGv/8539RFIWUlC956aUXeOihP9CtWzzFxUUcPnyIxMThdOoUyw8/bOPZZ5/k979/hN69y5NabGwcBkPtR281l20RNFWleM0fQVNpdc9fkGowKbpW99c0iv4zD6VtDJ63JwFNu6P6ZhprvUq3fYTlxPd4T30bSX/jIfJaWRGF/56DvvsdeCRMrnS+sdatJm42uKDFDKd2NcupbVhOpiJJEq7O7fq4RPS1bLpp1649/v4BXLp0ke7dfxr599JLLxATE8srr7xhHwrauXNXfv3rSWza9BUjR47h2LEjREd34re/fdBebsiQn/rpYmPLl25p3z7c4d4tmTV9F2reeTxGPOz0pAMgSVL5fJ4LJ9A0rdIqHkLD0jQNa+Y+dO27V5l0oHzCr9KuK9aMvWgD728xv6fmOcFEcIqyslIOHz7IiBF3YLPZsFqtWK1WWrcOJiKiA8ePHwUgNrYzp0+f5G9/e50DB/ZhsbTM2dg1pak2yvYmIwe0RxfVv8Geo4TEoRXno5nEElGupuacRSvKvWkzWwVdZD+061dRc1pO37N442kg+tjB6GMHN5mmthsxmUzYbDaWLn2TpUvfrHQ+LKwdAGPG3ElJSQmff76WVav+S6tWXowcOYakpDl4etZ+heXmznpmB1rBJYx3PIokNdx3v4qRVLZLp5D92jbYc4TKrBl7QZJQOvSq9lpdh96USf/EmrkXpXXDNiM3FiLxCFXy9vZBkiQeeGA6Q4cOdzinKDKenuWblMmyzL33Tubeeydz7do1vvtuE++88zdatWrFww/PdkfojZamWsvfdoI6OCyh0hBk/zAkDx+sF0+ij7u1QZ8lOLJm7kcJiUP2qH7VAdnTFyUkDmvGHoz9JrkgOvcTiUew0+n0mM0/7ePi6elJ9+49yMzM4Pe/f+QX1974Ta5169bcc89kvv12s33ejl6vB3C4d0tlObUN7fpVPEY+3uDt+ZIkoYTEigVDXUwtuIyadw7jLVNqXEYX2Zey7R9jy7+A4t/8R3uKxCPYdezYkS+//B9ff72BiIgO+Pn588gjjzNnzh947rmnuOOOUfj4+HLt2jUOHNhLQsJgEhOH8eqrf8Hb24du3eLx9fXl2LEjHD16mEcffRyAwMAgvL19+OqrDYSGhmEwGIiO7mRPSC2FZrNi3vc5cnAUSkRPlzxTCY0tH65bmAvNcM2vxqhii+ua9O9U0HUsTzzWjL0ovUXiEVqQcePGc+zYUZYseY2CggL7PJ533/2Qv/99OS+9tAizuYzWrYPp27cfkZHlSxvFx/fkiy+SWbfuf5SWlhIaGsof/vAo99xTPjxUlmUWLvwz7733/3j88STMZnOLnMdjOZmKVpiDx63TXTZ6SQkpH1Fou3QKIltG/4G7WTP3IQd1QPZpXeMysncgcpsorBl7Mfa+swGjaxxE4hHsvLy8+b//e7HS8bi4zrzyiuPggp83tY0ePY7Ro8fd9N6JicNITBzmtFibGs1qxrz/C5S2nVDad3fZc+WgcNB7/Lge2B0ue25LpRYXYLt8BkPfCbUuq4/sR9nOVajXryL7BFdfoAkTw6mFJkmzlqGZS9wdRo1Zjn+HVpSHof8kl87VkGQFJaST6OdxkfItrjV0kTVvZqug+3EbbGvGPidH1fiINx6hydE0jZL1r2O7nIYS1hldxz7oOvZB9gpwd2iVaJqGmpOF+cA6lNDO6MK6uDwGJSQO8+5PsRWbgJYxQdFdrJn7kHyCkQPa17qs7NsGOSgca8YeDD1GNkB0jYdIPEKTY7t8GtulUyjhPVBNVyjb9m/Ktv0bOTiqPAlF9nHryCCttBDruSNYzx3Gln0EraQAZB3GAfe4JZ6K+Tyl2cchsGs1Vwt1pZnLdxTVdxtR57daXcd+mPcmoxbnI7fyd3KEjYdIPEKTYzm4AcnojecdjyDpjNjyLmDN3Ic1cx/m3Z9i3v0psl8Iusi+5W9CwZENOlFTU1XUq+lYsw9jPXcY9UoGoIHRC127bujC41HC4932QaIER4KiE4mngVmzD4NqLd/Wuo50kf0w7/0Ma+Y+DF1vc2J0jYtIPEKToqk2rGcPYOhzJ5KufA0sJSAMJSAMY+9xqIW5WM/uL09CB1MwH/gSqZW/vTlOCe3slH1v1KI8bOeOlCeb80ehrAgkCTk4CkPf8ejC45FbRzbIOmy1JSl6lDbRlGYdw9DzbneH02xZM/chefigtKn7KtNyQBiSXwjWjL0i8QjVEwsxNjxN08BcAoqCvuuNd5OVvQMxdBuBodsItLIirFkHsWbuw3JqK5Zj34DBE11Ez/JE1D4eyVCzJX00mwXbpdNYsw9jO3cYNfccQHlS69AHXXg8unZdkTxuvBqvuymhcZTtX4feXFLjOgs1p9msWLMOoo/qX68vG5IkoY/sh/ngerTSQqB5zr0SiccJFEWHxWLGYKh6FVqh/sxlJUhFOeg7DUZu5Vft9ZLRC32nQeg7DUKzmrGdP1reJHf2ANYzO0DRobTrVp6EOvRG9vR1KK+arpS/0WQfxnbhOFjLQFZQQmIxDrwPpX08cmD7JvGFQwmJBU3FdvkMunCxQriz2S4cB0tJnUaz/ZIusi/mA+vKR8iFhzohusZHJB4n8Pb2Jz//Kv7+wej1hibxQdSUaJqGxWIm/2o2hhPfoR/xUK3vIekM6Dr0RtehN5qqYrt8+sd+ob2UZR2kjH+ihHRCiejJNbWQwlP70EyXy8v6BJcv+BoejxLWBUnv4ewqNjilbQxIMrZLp0TiaQDWzL2gM6KE1b8PTW7dEck7CEvGXhgyxgnRNT4i8ThBxWKZBQXXsNmsDudkWUZVm+bq1Dfj6nrJsoLhwOcYvXzqPWJNkmV0oXHoQuPQEiaj5mZjzdiL9ew+zLtWY9EbkUM7o+t+R3lfTTNY2VnSe2AMjcYq5vM4naapWDP3owuPR9LVfnPDX5IkCV3HvliOf4Na1nTmqtWGSDxO4unpZU9AP9eUdxC8GVfXy3zsW8ouHsUwbkH1F9eCJEkoQREoQREY+01ELc4nuF1bcvLKnPqcxsAjogsFu9ajWc1O+YAUyqlX0tFKCuo1mu2XdJF9sRz5muK0fRDcw2n3bSzcP+RGEKqhaSrmwynIwZEooXEN+iy5lT9yM/1Q9gjvCqoV29UMd4fSrFgz94GkoAt3XoJQ2nZC8vSl6MQPTrtnYyISj9DoWc8eQCu4jKHHKNF/Vg8e4Z0BxPI5TqRpGpbMvSjtuiAZK7d41JUky+g69qH4zH40a/PbTsStiSczM5MZM2bQu3dvEhISWLRoESUl1bdprl+/nrxh5HEAACAASURBVNmzZ5OYmEhcXBwffvjhDa+zWCy8/vrrDBkyhJ49e/LAAw9w/PhxZ1ejwWmlhdgun8Fyahtle9ZS8s1yrJdOuTssl7EcSkHyDrKvZSXUjeLpgxzY/scFQwVnUPMvohVcrtUWCDWli+yHZinFeu6I0+/tbm7r4zGZTEydOpWwsDCWLFlCbm4uL730Erm5ubz5ZuVtln8uJSWF7Oxshg0bxieffFLldS+99BLJycksWLCAdu3a8cEHHzB9+nQ+//xz2rZtXB3Gmrm4fAMp+59LqKby/6es6KcLJQkkBdV0Bd2EP7ovYBexXUnDdukUxlumIMmKu8Np8pSQOCyntqKpNvH36QTWzL1A+fbVzqaEdUb28MaasRd9AyQ2d3Jb4lm5ciUmk4nk5GQCAwMBUBSF+fPnk5SURKdOnaos+9ZbbyH/OEmrqsRz+fJlVq5cybPPPst9990HQM+ePRkxYgQrVqzgqaeecnKNqqdZyhwSilpwGe3HJKOV/ryjXkLyDkT2a4s+agCyX9sf/4Qg+QRjOf5t+W6FVzPKl0NpxsyHUsDQSmzd7CRKaCyWY5tRr51FaRPl7nCaPGvmPuQ2UQ2yQK0k62gV24/CE7vQbFanrLjRWLitJqmpqSQkJNiTDsDIkSNZuHAhqampN008cg1mBm/duhWbzcaYMT+Ng/f29mb48OGkpqa6JPHYrmZwddc2ii9nlyeZ4nyH81Irf2S/tug69kbyDfkpwfi2uemoI33sYMp2fYr56GY8h9V+TktToZqulK/U23OMmG3vJEpI+YKhtksnReKpJ7UwF/VqBoYGXPzVKy6BwkPfYbt4Ap0L93FqaG5LPGlpadx9t+O6UQaDgYiICNLT051y/9atWxMQ4PhNJCYmhnXr1qGqao0SWH1Yzx2l7PRuNO9glPbdkP1+TC6+5QmmrhMRJUMr9LGDsZxMRU24H9mjeS6rYT78NUgy+m63uzuUZkP2CkDybYvt4inoMdrd4TRp1rO13+K6tjyjeoLOiDV9j0g8zmAymfD19a103NfXl4KCAqfc38en8geyn58fFouF4uJivL1rvq5WUFAd1uD61a/L/zQA85A7OXfsG4zZO/EfNLFBnlGd4OCGS3i2kutkndqCd/dE2nSMaLDnVKUh6+ZOwcE+ENmNolO7aN3aq0FX7XYld/y+Ll44hD6oHSGdGnaIv1envpRm7ad1UFKz6ZdrPo2GDSwnpxBV1WpdruEmWgaghHUhb/cGzNG3uXwV5IaeQFq2/4vyPrHY21w+Abe5T/q1BEShlnzD5VMnUQJrv2FZY+OO35dWVkRJ5lEMPUc16LODg32whfXCdnw7l47sR9fA89icSZalKr+wu+3rjq+vLyaTqdJxk8mEn1/1C0DW5P7Xr1f+B1FQUIBer6dVq1b1foa76buNQCvMwZp1wN2hOJVms2A5shGlfXeUwHB3h9Ps2Pt5xHyeOrNmHQTN1qDNbBV04fGg6LBm7GnwZ7mK2xJPdHQ0aWlpDsfMZjNZWVlERdW/0zM6OpqcnBzy8x079NPS0ujYsWOD9++4gq5DbySvQCxHN7s7FKeynv4BrcSEQfRBNAjJJxjJK0DM56kHa+a+8sFBLhhVKhk80bWPx5qxt3xrkGbAbZ++iYmJ7Nixg7y8PPuxjRs3YjabGTp0aL3vP2TIEGRZZsOGDfZjRUVFfPPNNyQmJtb7/o2BJCvouw7Hdv4otvwL7g7HKTRNK18eJygcpZ3YLbMhSJKEEhKH7eLJZvNB5kqa1Yw1+xC6jn1c1kemi+yLVlQ+iq45cFvimTx5Mj4+PiQlJbFlyxaSk5NZtGgRY8aMISbmpx38Fi5cSNeujh9AZ86cISUlhZSUFABOnTrl8DNA27ZtmTx5Mq+99hqrV69m27ZtzJkzB4Bp06a5oIauoe88FGQdlqPfuDsUp7BlH0bNu4Chx2ixPE4DUkJj0Yrz0a5fdXcoTY7t/FGwml3SzFZBF9ELJMU+YbWpc9vgAl9fX1asWMHixYuZPXs2RqORsWPH8uSTTzpcp6oqNpvN4diGDRt4++237T8nJyeTnJwMwMmTP7VbP/PMM7Rq1Yq33nqL69evEx8fzz/+8Y9Gt2pBfcievuiiB2A5tRVj/7ub/HwX86ENSF4B6KIHuDuUZk0JKe+ktl08iezbxs3RNC3WzH1g8EQJ7eyyZ0oe3ihhnbFk7MHQ/54m/6VM0sS7do00vlFtP7FdSac4+QWMg3+LoduNt4R2toaol+1aJsVrn8c48H4MPd3Xv9PcR7VB+YrfRf+ag9KhV5OfhOzK35em2ij66HGU9t3wvO0PDf68n9fNfOxbyrauoNU9i5vEaMRGOapNcB6lTRRycCSWo5ubdJu9+VAK6D3Qd6l/H59wc5Iko4TGigEGtWS7fAat9Dq6js7be6emdB17A1KzGN0mEk8zYeh2O2r+hfK935sgtTAHa9ou9F2GIRma/lD3pkAJiUMzXUEtyqv+YgH4sZlN0bllFQG5lT9KSCesGU2/n0cknmZCF9UfycOnyQ6tNh/+GpAwdL/D3aG0GEqomM9TG5qmYc3ci9Kum9v6UnWRfVFzy9d+bMpE4mkmJJ0BfedErGf3oRbmuDucWtHKirCc+B5d9ABk7yB3h9NiyEERoPcQzW01pOZmo12/5tLRbL9U0cRnaeJvPSLxNCP6LsMBsBz71s2R1I7lxPdgKcXQY5S7Q2lRJFlBaRtTvmCoUK3yJi6pQfbeqSnZpzVycGST7+cRiacZkX1ao+vQG8uJ75vMdrmazYr5yEaUsC4orTu4O5wWRwmNQ807h1Za6O5QGj3r2X0oIZ2QPSsvbuxKuo59Ua+mN7mWjZ8TiaeZ0Xe7Ha30Otb03e4OpUas6bvQivLE8jhuUrFuW0vaSr0uVNNV1JxstzazVdD/uAW8NXOfmyOpO5F4mhklrAuyfyjmo5vcHUq1NE3DfGgDckAYSni8u8NpkZTgSFB0YoBBNSo+5BtD4pH9Q5AD2jXp5jaReJoZSZLQdxuBejUD25X6b6jXkGznj6HmZIvlcdxI0hlQ2kSLAQbVsJ7dhxzYvtGs8qCL7Ift0inUksor/DcFIvE0Q/pOg0Hv0ejfesyHNiB5+qGLSXB3KC2aEhKLeu0smrnE3aE0SmqJCdulU26ZNFoVXWRf0LQm29wmEk8zJBk80ccOxpq2q9F+I7LlZGM7dwR99zuQFL27w2nRlNA40FRsV9Kqv7gFsp09AJrWKJrZKsiB4Ui+bZrsoqEi8TRT+m4jQLWWD1VuhMyHU0BnxNBlmLtDafGUNtEgyQ3az2PLO09p6j+xnNyCpqkN9pyGYMnch+QdVD7vqZGQJAldx77Yzh9DKytydzi1JhJPM6X4h6G064bl2Ldoqq36Ai6kFuVhPbMDfedEJI8bLyIouI5k8ERu3aFBEo9mLqb0h/9S/OmfsJzcQun3H1KcvLjR9z9W0Cyl2M4f+XHvncbVD6mP6geqrXw31CZGJJ5mTN9tBFpRLtaz+90digPLkY2gqRi6/8rdoQg/UkLjsF1Nd9r8L01TsZzaStEnC7Ac/hp93BC8HngTj+Gz0ApzKE5+gdLv/95om4IrWLMPg83aqJrZKsjBkUheAU1ydJvb9uMRGp4uoheSdxCWo5vtY//dTTOXYD7+LbrI/si+we4OR/iREhKL5VAKtqsZ6ELj6nUv29VMSrf9G/VKGnKbaDxHPVE+bBuQOw1C16E3Zfv+h+XwRiwZuzH2m4S+621IsuKMqjiVNXMfktHbPt+pMZEkGV3HvuUTxi1lSHqju0OqMfHG04xJsoy+623YLhzHlnfe3eEAYDm5BcwlYnmcRkYXUv8FQ9USE6Wp/6D4s/9Du34Vj2EP0Wr8s/akU0EyeOKRMJlW9y5CCY6ibPvHFK/9M9YLJ+pVB2fTVCvWrIMoHXo1yqQIP45us1mwZh9ydyi1IhJPM6fvnAiKrlGsWq2pNsyHv0IJiUVpE+XucISfkTy8kQPa12k+j6baMB/ZVN6sdnIr+vhf4XX/y+hjhyBJVX/EKP5heI6Zj8cds9EspZSse5mSTe+gFubWpypOY7twEszFjbKZrYISEofk4dPktkoQTW3NnOzhgy46AcupbRgH3OPWvW6sGXvQCnMwDHrAbTEIVVNCY7Gc3o6m2mr8Dd964QRl2z9CzT2H0q4bxkG/QQkIq/EzJUlCH9kXXXg85oPrMR/4EmvWAQy978LQY6Rbh9pbM/eBzuCWvXdqSpJldB17Y0nbhWazNJmpCeKNpwUwdLsdrGVYTm1zWwyapmE+uAHJLwSlQ0+3xSFUTQmJBUspak5WtdeqhbmUbH6XknUvo5lL8LjjUTzHzK9V0vk5SWfA2HcCXve9iK59PObdn1K0+jmsWQfqdL/60jQV69l96NrHI+kMbomhpnSR/cBSiuX4d01mB2LxxtMCKMEdkdtEYz66GX23ETdt/mgotosnUa9lYrx1ulueL1RP+XFQge3iyUr9MhU0mwXzoa8w7/+8fGRin/EYeo1B0jmnY1v2CcbzV7OxnjtC2faPKUl5CyWiJx63TEH2a+uUZ9SEejUTrSgPXf/G28xWQQnrihISS9n2j7FdOoVxyFRkDx93h3VTbv0EyMzMZMaMGfTu3ZuEhAQWLVpESUnNlu1ITk5m1KhRxMfHM3bsWNavX1/pmnPnzjF37lyGDBlC7969mTRpEl9++aWzq9EkGLqNQCu4hO38Mbc833xoA5KHD/pOg9zyfKF6slcAkm+bKgcYWLMOULT6Ocy7P0XXvjte976Esd9EpyWdn9O1706ruxdhHHg/tosnKVr9LGW7PkWzlDn9WZqmoV6/hiVzH2V7PqPkqyWUfLUEJBldRON/O5cUHZ7jnsbQ/x6smfsoXv1so5tC8Uu1fuMxm80UFxfj7+9vP5abm8vq1asxmUyMHDmSHj16VHsfk8nE1KlTCQsLY8mSJeTm5vLSSy+Rm5vLm2++edOyKSkpPP3008yaNYvBgwezadMm5s6di5eXF0OHDgWgrKyMGTNmALBw4UL8/f1Zt24dc+fOxWg0cvvtt9e26k2aLqo/0g//xXJ0s8vbrG15F7BlHcTQd2Kjb7Zo6ZSQOGxn96Npqv3NVC24TOkP/8GWdRDZLwSPMfNd8m9IUnQYeo5G1+kWynauwnxgHZbT2zEm3I8uakCdJnRqqhU1/yLqtSxsOVmoOeX/xT77X0L2D0EJ64IuekCTmeAsyQrG3uPQRfSk9Nv3KPlqCbrYW/EYNMVt23TfTK0Tz3PPPceZM2dYu3YtACUlJdx///1kZ2cDsGLFClasWEHfvjdfUG/lypWYTCaSk5MJDAwEQFEU5s+fT1JSEp06daqy7JIlSxg1ahTz5s0DICEhgfT0dJYuXWpPPIcPHyYzM5MVK1aQkFC+COWgQYM4ePAg69ata3GJR1L06LsMw7x/Her1q8g+rptDYzmcAooefbfbXPZMoW50obFYT21BzbuI7NMa8/4vMB9KAUWHceD9P66t59oWermVP57DZ2HtMpyybf+mdPO7KMe+xTj4AZTA9lWW08wl2HLPoV47+2OCOYuaex5Ua/kFih45MBx9VH/koAiU1h2QA9o3qfkwv6QEhdNq4p8x703GfPBLii4cw2PoDHTturo7NAe1/he0Z88eJk2aZP953bp1ZGdn8/7779OlSxdmzJjBu+++ywcffHDT+6SmppKQkGBPOgAjR45k4cKFpKamVpl4srOzSU9P54knnnA4Pm7cOJ555hlyc3MJDAzEai3/x+Xj49jW6ePj02Q64JxN32UY5gNfYjn2LcaB97nkmWpxAZbT29HHJTb6dmfhp34e877/Ybt8uryfo9MgjAPvQ27lX03phqUL6YQy8XksJ76jbPcaitf8CX23EdhG/ha1OP/Ht5gfk8y1LDTTZXtZycMHOSgCffc7UFpHIAd1QPZr22jn59SHpOgwDrgHXYdelHz3ASVfvoq+2+0YB97bIM2idVHrxHPt2jVCQkLsP2/evJlevXpx6623AjBp0iTee++9au+TlpbG3Xff7XDMYDAQERFBenrV6zhVnIuOjnY4HhMTYz8fGBhInz59iIqK4o033uD555/H39+f9evXc+TIER577LGaVbaZkb2D0HXsg/nE9xj6TmjwZi/VdIWSb98Dmw1DvFgepymQfIKRWvljTd+FHNQBzxFJKCFVtz64miTLGLrehj5qAGW712A5somzx76Bn61HKPm2QQmKQI4dbE8yUiv/RrfWWkNT2sbgdff/UbbrUyxHNmI9dwTPYQ+htI1xd2i1TzxeXl6YTOXrK1mtVnbv3s306dPt5z08PCgsrH7/dpPJhK9v5b3LfX19KSgoqLJcxblflvXz83M4bzAY+Pe//83DDz9sb1YzGAy8+uqr3HLLLdXG90tBQXVv6w0Objzf9EsG38nFj/bgeeUgPj3r1/RVVb00TeP6gU3kbPwnkizTZsJjeMc0ng+vmmhMvzNnqkm9isY9jFpyHe/uiY34jcAHwh+l7NIYCg9/j86/LYa2HTG26YDs4eXu4Jyqfv8WfWD8HyjpOZirX7xN8ecv4n/LBAIS73PrnJ9aJ57u3buzevVqEhIS+OabbyguLmbYsGH281lZWbRu3dqZMdZJaWkpjz32GKqq8vbbb+Pt7U1KSgrPPPMMrVu3ZsCAAbW6X05OIapa+ya64GAfrl69XutyDUXzjEAOaEfOD+soCe1X52+BVdVLLc6nNPUf2LIOooR1wWPYQ5R4B1HSiP4OqtPYfmfOUuN6+ceBP5TmFDd8UPWlBBN8x4NcvXodM1B4XYXrzed357R/i14d8Zj0AqXb/0v+9rWYTuzGY/gslKDw+t+7CrIsVfmFvdaJ54knnuDBBx/k7rvvRtM0+5DmCps2baJPn+rHvvv6+trfnH7OZDIRFVX1cioVbzYmk4ng4J86yCvedCrOf/rppxw8eJDU1FR7P9Itt9xCVlYWb7zxBitXrqxBbZufiq2xy7b+C/VKmlNfuy3puynbsgLNWobxlinou98u5uwIQiMhGVrhOWwG1sg+P66p9zyGvhMx9Bzt8jfbWieerl27smHDBvbv34+Pj4/Dm4PJZGLKlCk1epuIjo4mLc1xx0Oz2UxWVpbD4IVfqkhK6enpDv08FfeqOH/mzBnatm3rMHgBoEuXLiQnJ1cbX3Om7zSIsp2rMR/dhKcTEo9mLqZ020dYT29Hbt0Rz+Gz6jyDXRCEhqXr0JtW98ZQtvVfmHd/ivXsfjyHzUT2D6m+sJPU6etoYGAgI0aMqJRgfH19mTZtGl26dKn2HomJiezYsYO8vDz7sY0bN2I2m+1Dom8kPDycqKioShNG161bR3x8vD3RhIWFcfnyZXJychyuO3r0KO3atas2vuZM0nugjxuCNX03anF+ve5lPX+sfGmTMzsw9BlPqwnPiaQjCI2c7OGDx4gkPG77A2rBJYrW/AnzkY0u2x221oknLS2NTZs2ORzbvXs3M2bM4N577+Wf//xnje4zefJkfHx8SEpKYsuWLSQnJ7No0SLGjBljH6EG5ZM/u3Z1HIM+Z84cNmzYwJtvvsnOnTt58cUX2bZtG7Nnz7Zfc9ddd+Hh4cHMmTNJSUlh69atPPfcc+zYsYMHHhCLVBq6jgDVVuetsVVLGaXbP6bky1dBZ6DV+OfKZ7HLYhUmQWgKJElCH5OA1z2LUcI6ly9R9OVfUQtzqi9c32drtZzUMnPmTCRJsg+ZvnTpEqNHj8ZoNBIUFER6ejovvvgiEydOrPZeGRkZLF68mL1792I0Ghk7dixPPvkknp4/zbRdsGABn332GSdPOi7j8dlnn7Fs2TLOnz9PREQEjzzyCGPHjnW45vjx47z11lscOXKE4uJiOnbsyLRp05gwYUJtqgw0n8EFP1e8/jXU3HN4TXmtVgnDdjUTy5YPsFw7h77bCIwD72s08wOcoTH/zupD1KvpcVXdNE3DcuJ7ynasBCQ8Bk1BFzukXkPQbza4oNaJZ/DgwUyfPp2ZM2cCsHz5ct59912++uor2rZty8MPP8zVq1f59NNP6xxwY9QcE4/17AFKvnoLj9uT0EdV3y+nqTbMB9Zh3vs5ircfhlt/16iXjK+rxvw7qw9Rr6bH1XVTTVcp/f4DbBdPouvQG+Ot0+o8cfhmiafWTW0mk8mhw/77778nISGBtm3LV44dNmwYmZmZdQpUcC0lvAeST3CNNolT8y9R/L+/YN7zGbro/rSf+WazTDqC0JLJvsF4jnsaY8KvsZ47TPGaP6NZnb8wa60b5IOCgjh/vnwb5YKCAg4dOsRTTz1lP282m1vskjRNTcUs8LKdn2DLzUYJrDymX9M0LMc2U7ZjFej0eIxIQh89AMXTGwqb57dMQWjJJEnG0GMkSng8tnOHoQEmmtY68QwePJiPPvoIHx8fdu7cCcCIESPs50+fPk1oaKjzIhQalD7uVsr2rMVydDPKrdMdzqlFeZR+/yG2c0dQwuPxSPwdsleAewIVBMGllICwBhuhWuvEM3fuXDIyMnjllVfQ6/U8+eST9uHJZWVlfPXVV9x5551OD1RoGJKHN/qYW8qXmx9wL5KxfLkRy5kdlG77N9gsGIdMRd9leItb60oQhIZRp6a2//znP1y/fh2j0YjB8NNCk5qmsWLFCodFRIXGT99tBJaTqVhObUXfaTClW/9Vvkhkm2g8h89E9hO/T0EQnKfOky5+ud0AlC8Q2rlz53oFJLie0roDctsYzIdSMB/cgFZyHUO/SRh6jW3Ei0QKgtBU1Snx5Ofn89577/H999/bBxq0a9eO4cOH89BDDznsTio0DYZut1P6zTLkgDA8Rz2B0rqDu0MSBKGZqnXiuXjxIlOmTOHixYt069aNX/2qfJ+V9PR0PvjgA7788kv+85//iAEGTYwueiCeHj4oIZ3E9tSCIDSoWiee1157jYKCAv71r39VWqttz549/P73v+e1117j9ddfd1qQQsOTJAld+27uDkMQhBag1hNIt27dytSpU2+4AnW/fv144IEH2Lp1q1OCEwRBEJqfWiee0tLSSlsN/FxQUBClpaX1CkoQBEFovmqdeGJiYvjiiy8wm82VzpnNZj7//HM6dWpa2xwLgiAIrlPrPp5Zs2bx2GOPcffddzN58mQiIyOB8pWmV65cyZkzZ/jb3/7m9EAFQRCE5qHWiWfkyJG88sor/PWvf2XRokX22eyaptG6dWtefvll7rjjDqcHKgiCIDQPdZrHM378eMaOHcuRI0e4cOECUL7jZ/fu3dHpxEZggiAIQtWqzRIVieVG2rRpQ5s2bew/X7lyxf7/YWFi+2NBEAShsmoTz2233VanxSGPHz9ep4AEQRCE5q3axPPiiy+KVYkFQRAEp6k28UyaNMkVcQiCIAgtRK3n8ThTZmYmM2bMoHfv3iQkJLBo0SJKSkpqVDY5OZlRo0YRHx/P2LFjWb9+/Q2vS09PZ/bs2QwYMICePXty55138tVXXzmzGoIgCEItuG0ImslkYurUqYSFhbFkyRJyc3N56aWXyM3N5c0337xp2ZSUFJ5++mlmzZrF4MGD2bRpE3PnzsXLy4uhQ4farzt9+jRTpkwhISGBV155BaPRyMmTJykrc/4e4oIgCELNuC3xrFy5EpPJRHJysn0JHkVRmD9/PklJSTdd/WDJkiWMGjWKefPmAZCQkEB6ejpLly51SDx//vOfueWWWxwmtA4aNKiBaiQIgiDUhNua2lJTU0lISHBY923kyJEYDAZSU1OrLJednU16ejpjx451OD5u3DgOHz5Mbm4uAGlpaezdu5epU6c2TAUEQRCEOnFb4klLSyMmJsbhmMFgICIigvT09CrLVZyLjo52OF5xr4rzBw8eBKC4uJhJkybRtWtXEhMTefvtt1FV1Wn1EARBEGrHbYnHZDLh6+tb6bivry8FBQVVlqs498uyfn5+DuevXbsGwPz587ntttv4+9//zj333MM777zDBx984JQ6CIIgCLXXbNe3qXirGT9+PI8++ihQ3heUn5/PsmXLeOihh5DlmufdoCDvOscSHOxT57KNWXOtFzTfuol6NT3NsW5uSzy+vr6YTKZKx00mE1FRUVWWq3izMZlMBAcH249XvOlUnK94I/rlYIJbbrmFjz/+mPPnzxMeHl7jeHNyClFVrcbXVwgO9uHq1eu1LtfYNdd6QfOtm6hX09OU6ybLUpVf2N3W1BYdHU1aWprDMbPZTFZW1k0TT8W5X/YDVdyr4nx1ewLdaD8hQRAEoeG5LfEkJiayY8cO8vLy7Mc2btyI2Wx2GBL9S+Hh4URFRVWaMLpu3Tri4+Pto+R69eqFv78/27Ztc7hu27Zt+Pj4EBER4cTaCIIgCDXltqa2yZMn89FHH5GUlERSUhI5OTm8/PLLjBkzxmG028KFC0lOTubYsWP2Y3PmzOGJJ54gIiKCQYMGsXnzZrZt28by5cvt1+j1embPns2LL76Ij48P/fv3Z+fOnaxatYrHH38cvV7v0voKgiAI5dzax7NixQoWL17M7NmzMRqNjB07lieffNLhOlVVsdlsDsdGjx5NaWkpy5Yt48MPPyQiIoLXX3+90pvSAw88gCRJrFixgvfff5+QkBCeeuoppk+f3tDVEwRBEKogaZpW+x7zFkgMLnDUXOsFzbduol5NT1OuW6McXCAIgiC0TCLxCIIgCC4lEo8gCILgUiLxCIIgCC4lEo8gCILgUiLxCIIgCC4lEo8gCILgUiLxCIIgCC4lEo8gCILgUiLxCIIgCC4lEo8gCILgUiLxCIIgCC4lEo8gCILgUiLxCIIgCC4lEo8gCILgUiLxCIIgCC4lEo8gCILgUiLxCIIgCC4lEo8gCILgUiLxCIIgCC7l1sSTmZnJjBkz6N27NwkJCSxatIiSkpIalU1OTmbUqFHEx8czduxY1q9ff9PrFy9eQoGFYAAAIABJREFUTFxcHC+88IIzQhcEQRDqSOeuB5tMJqZOnUpYWBhLliwhNzeXl156idzcXN58882blk1JSeHpp59m1qxZDB48mE2bNjF37ly8vLwYOnRopetPnDjBmjVr8Pb2bqjqCIIgCDXktsSzcuVKTCYTycnJBAYGAqAoCvPnzycpKYlOnTpVWXbJkiWMGjWKefPmAZCQkEB6ejpLly6tlHg0TeP5559nxowZrF27tuEqJAiCINSI25raUlNTSUhIsCcdgJEjR2IwGEhNTa2yXHZ2Nunp6YwdO9bh+Lhx4zh8+DC5ubkOx9esWcOVK1d46KGHnFsBQRAEoU7clnjS0tKIiYlxOGYwGIiIiCA9Pb3KchXnoqOjHY5X3OvnZfPz83nttdd45pln8PDwcFbogiAIQj24LfGYTCZ8fX0rHff19aWgoKDKchXnflnWz8/P4TzAG2+8QZcuXbjjjjucEbIgCILgBG7r42lohw4d4rPPPiM5Odkp9wsKqvvAhOBgH6fE0Ng013pB862bqFfT0xzr5rbE4+vri8lkqnTcZDIRFRVVZbmKNxuTyURwcLD9eMWbTsX5xYsXM3HiRIKDg+3PUVUVi8WCyWTCy8sLRVFqHG9OTiGqqtX4+grBwT5cvXq91uUau+ZaL2i+dRP1anqact1kWaryC7vbmtqio6NJS0tzOGY2m8nKyrpp4qk498t+oIp7/fz8J598Qv/+/e1/Ll68yKpVq+jfvz+nTp1yZnUEQRCEGnLbG09iYiLvvvsueXl5BAQEALBx40bMZvMN5+JUCA8PJyoqivXr1zv03axbt474+Hj7KLlly5Zhs9kcys6dO5devXoxdepUIiIiGqBWgiAIQnXclngmT57MRx99RFJSEklJSeTk5PDyyy8zZswYh9FuCxcuJDk5mWPHjtmPzZkzhyeeeIKIiAgGDRrE5s2b2bZtG8uXL7df069fv0rPNBqNtG3bloEDBzZs5QRBEIQqubWPZ8WKFSxevJjZs2djNBoZO3YsTz75pMN1qqpWenMZPXo0paWlLFu2jA8//JCIiAhef/31m74pCYIgCI2DpGla7XvMWyAxuMBRc60XNN+6iXo1PU25bo1ycIEgCILQMonEIwiCILiUSDyCIAiCS4nEIwiCILiUSDyCIAiCS4nEIwiCILiUSDyCIAiCS4nEIwiCILiUSDyCIAiCS4nEIwiCILiUSDyCIAiCSzXbHUibo1Kzld3Hr7Dr+GUCfT0Y0iOUmHZ+SJLk7tAEQRBqTCSeRk7TNDIuXif14AV2Hr9MmdlGmwBPzpw3seXQRdoGtmJIfAiDuocS4GN0d7iCIAjVEomnkSossfDD0UtsOXiBc1eLMOhl+nduQ2LPMGLa+VFmsbHn/7d351FNXnkfwL9ZWRLCEnYksivFiK3Uqih1qccq2Gq1VWZcWmntGax2bG3V1rdzpnbU7m7HpVV7dKhn5rjh0tYO6oyII/K2brxaqwUx7EsCWcCQ7Xn/iImEQEAkgdjf5xzOSe5zb7g/HpJf7n2e5z436lFQXI2DZ0pxKL8UQ6LFGDM0DMPiAsHj9u0squauHtduK3C1RA5VcytGS8Pw5OBgcDk0u0vI7x3dFqGbXHFbBBPD4FdZE85eqcJPv9bDYDQhKtQHacnhGJEYAm/Pjr8n1Da24FxxNc4V16BR3QqBJxcjHwvFmKFhkIQInTIV1z4uE8OgvFaDqyUNKC5VoKRKCYYBhF48eHlwUN+kha+Aj/GPR+DpxyPgK+D3ep96izsvRe8IxeV+3Dk2R7dFoMTTTc5MPE2aVpwrrsbZK9Woa7oLbw8uRiWFYmxyGCQhPt3+XSYTg+t3FCi4Wo2LNxtgMJowIEiIsUPDMDIpBD7evfdhHxTkg7JyBa7dVqC4VI7iUgVUzToAQHSYD6QxYkhjxIgOEwEs4NptBU7+VIHiUjm4HBZGJIZgUkokBoZ2Pz5Xcec3uyMUl/tx59go8fSC3k48RpMJxSUK5F+pwtUSOUwMg8ESP4xNDsfwhCDweZyH6m+zVo+i67UoKK7G7Wo1OGwWhsUFInVoGKQxAeCwH3zKi2EYlNdpcLVEjhvlTbhR1ggTw0DgyUVSdACGxooxJFoMkYPRTLW8Gad/rkRBcTVa9UbED/DFMymReCIhsEd9cgZ3frM7QnG5H3eOjRJPL+itxFPXdBdnr1ShoLgaSo0OIgEfqdJQpA0NR0iAd2922aqiXoOCq9UovFYDVYsevgI+Rg0JxRhpGMIDBQ7btmgNuF6mwNVSOYpL5VBqzKOa2AG+SJT4Y2iMGNHhPg+cNFq0BhQUV+PUz+Wob9LC38cDE56IwNPDIiD04vU41t7gzm92Rygu9+POsVHi6QUPk3iqqptw8WYD8q9U4Zc7jWCxAGmMGGnJ4RgaK3bZAXeD0YTiEjkKiqtxtUQOo4lBTLgIY4aGYcRg8zEkhmFQUd+M4lI5rpbIUVKphNHEwNvDPKoxT6EFIC46sFfeECYTg6slcuT9VI5f7jSCx2VjVFIInhkeiQHBHf/TOps7v9kdobjcjzvHRomnF/Qk8TSqW3HmajVO/a8MzVoDAn09MXZoGFKlYQgQeTqpp92jatbh/LUaFBRXo7K+GXwuG4MH+qO8ToNGdSsAQBIshDTWfKwmNkJkM6pxxhuiol6DUz9X4Pz/1UBnMGGwxA+TUiKRHBcINtt11yq585vdEYrL/bhzbP028ZSVlWHNmjW4ePEiPDw8kJ6ejuXLl8PLy6vLtrm5udi+fTsqKyshkUiwePFiTJ061bq9tLQUOTk5KCwsRFVVFfz9/TFq1CgsW7YMQUFBD9zXniSeb/Nu4szlSjweH4S05HAkRvmD3c8u9mQYBmU1ahQUV+NaqQKSECGkMWIMiRE7vC7ImW8IzV09zl6pwqmLFVCoWhHo64mJwwdg7NAweHs6fxruYWNjGAbNWgPqm+5af5o0Ovh48eDn4wH/Nj/eHlyXXQDszh9ijjyqcQHuHVu/TDwqlQoZGRkIDw9HdnY2FAoF1q1bh9GjR+PLL7902PbEiRN48803sWjRIqSmpuLkyZPIycnBjh078PTTTwMAcnJy8M9//hMzZ85EYmIiampqsGXLFhiNRhw7dgwCgeNjG+31JPEYjCb4+nmjWa19oHbuwBVvCKPJhEs3G3Dyp3LcrFDCg8fBaGkonhk+AGHiB9t/D6I7sRmMJihUWtQ3aW0STH2TFnVNd3G31WBT38uDg7utRrvX4XPZ1iRkTUpCS2LyhL+PB3wF/F4Z8bnzh5gjj2pcgHvH1i8Tz1dffYWtW7fi9OnTCAgIAAAcO3YMy5cvx/HjxxEfH99p2ylTpiAhIQEbN260li1cuBAqlQoHDhwAACgUCvj7+9t8m7xx4waef/55rF+/HjNmzHig/rriOh534uq47tSocfLncly4XguDkcGQ6AAkxwWCz2WDx2WDx+WAz2ODz2WDy2WDz+W02WZ+zuOxuzXitMTWrNXfTyaNLTZJRqFqhanNW4fLYSHQ1wtBfl4I9vNCkJ8ngvzMzwP9POHJ58JgNKFJ3YpGTSsa1fd/mjStUKhbzdvUrTC2+z9jsQA/oQf8hLajJX+hB8S+nogMFsLLo+trwel/0f24c2yOEk+frVyQn5+PkSNHWpMOAEyePBnvvfce8vPzO0085eXlKC0txbJly2zKMzIysGrVKigUCgQEBNi8rsWgQYPA4XBQV1fXu8EQpxsY6oOs9Mfw4rg4nLlcidOXKvF/txUP/DpcDut+omqfmLhscDlstOgMqK5vRku7UYuPNw/Bfl6Ii/BFYJI5uQTfSy5+Ph5dJjUuh41APy8E+nU+lWxiGGha9Oak1CZBmZOSFjWKFvxyp9FuRBXs54XIECEkwUJEhvhAEiyEv48HreNH+qU+SzwlJSWYOXOmTRmfz4dEIkFpaWmn7SzbYmNjbcrj4uKs2ztKOgBw6dIlGI1Gu7bEfYgEfExLjUb6qChotHro9SboDEboDSboDSboDCboDUbo9CbojffK9Lbb7errjdAbzY9bWg0Q+3lBEixEkK8Xgv3vjVp8Pbs1qnhYbBYLIgEfIgEfA9H5xbVanQGN6lbUNd5FeZ0GsjoNZLVq/PxrvbWO0IuHyGAhBob4IDJEiOTBDDxYTL+5Xor8fvVZ4lGpVBCJRHblIpEISqWy03aWbe3b+vr62mxvT6/XY+3atYiOjsa4ceN62GvSX7DZLIh6cSWGttxhesOTz0WYmIswsQDJcYHW8rutBlTUayCr1aC8Tg1ZrQYnf66AwWgCjl0Hl8PGgCABJCFCRAb7YGCIDwYEC+DJp2Ubiev8bv7b1qxZg1u3biEnJwdc7oOH3dlcZXcEBfW/ZWF6w6MaF+DesUkG+GN0m+dGowkV9RqUVipRWqnE7SolLt1qQP6VagDmY0hhYgGiI3wRE+6LmAhfhAcK4Cv0gLen6866exjuvL+68ijG1meJRyQSQaVS2ZWrVCrExMR02s4yslGpVDanRVtGOpbtbW3ZsgUHDhzA5s2bIZVKe9RfOrnA1qMaF/BoxubNYWH88EgMkfgBMJ/y3ahuhaxOg/JaNWR1Gty8o8C5K1U27bgcNkQCHny8+RB58yHy5kEk4JufC3gQeVse8+HjzeuT1ccfxf1l4c6x9cuTC2JjY1FSUmJTptPpIJPJ8MILL3TazpKUSktLbY7VWF6rfdL69ttvsXnzZqxZswYTJ07sre4T4tZYLBYCRJ4IEHliWJupuhateaquQXkXqmY91C06qJp1ULXooWrRobJBA1WzDgZjx1/CBJ7ce0nqXoIS2CasMLEAoQHeLr0gmPQ/fZZ40tLSsG3bNjQ2NsLf3x8AkJeXB51OZ70WpyORkZGIiYnB999/j0mTJlnLjx8/DqlUanNiwXfffYePPvoIS5cuxUsvveS8YAh5RHh7cpEQ6YeESL9O6zAMg7utRnNSatG1S1DmJKVu1qGyoRmqO41o1tqegcfnsSG5d3xpYKj5J0zs/Ujfq0mpacWdWvMJIAajyXwCyb2RouWxlwfHLaY1e0OfJZ45c+YgJycH2dnZyM7Ohlwux/r16zF16lTrGWoA8N577yE3NxfXr1+3li1duhTLli2DRCLB6NGjcerUKZw7dw47duyw1ikqKsKKFSuQkpKC1NRUXL582botICAAEonENYES8ohhsVjw9uTC25PbrYVtDUYTNHf1UGp0qKjX4E6NGndqzatlnLpYAcA8pRcZLMDAEB9IQu+d9BAkAI/7cKu0uxrDMJArtbhTq8GdWjVkteZYLYvrAgALQEfjRcu0pjUhefMRGiQElwXriNFSLvTiufWosU+P8ezZswcfffQRlixZYl0y55133rGpZzKZYDTaXvE9ZcoUaLVabN++Hbt27YJEIsHnn39uM1K6cOEC9Ho9ioqKMHv2bJv2M2bMwPr1650XHCHEisthWy+AHRjqg1RpGADzNUu1ihbzB3SN+YO66Jc6/Oey+TgTh81CeKDg/sgoxAeRwUJ48PtHMjKZGNQ22vZfVqu2jvDYLBbCAr2RFBUASYgPBt47k9CDz4a6RW8dIaqb9VBaH+ugbNGhSdOK8joNCq/XdDityWIBPl62yciDzwGPc//aNMv1ajwu21re9to17r1yPs++HYfNcuroixYJ7SY6ucDWoxoX8OjG5g5xMQyDBqXWOiq6U6vGnRo11C16APfPwBsYIrQmpOTEUDSr7zr1+iSD0YSqhub7SaZOjfJaDVr15i/FXA4LA4KEGBjqcy/JmEdsD3tfrcBAIe5UNJqTlOVYW7MOymadzfSmulmPVsv1aveuX3sYLBbA47IR5OuF/1mQ0qM4+uXJBYQQ0h6LxbIuNZQyOBiAORk1aXT3k1GNGjdkTTh/rdamLYfNurdcErvNt3yO3Tf9tqMBu/I2I4C7OoN5qqxGg8oGjXXk4cHnQBJsvrOvJMS5x6hYLBYEnjwIPHkPtD4hwzAw3EtAthdX309M+jYXUlvK2154bTCYIPDiOiUuSjyEkH6NxWJZ16cbFn//DDxVsw6yWjWUWgPkjS22H6JtV6gwmqDXm1el0DebP4ANBuP9D2KDyW59PAuBJxcDQ30wKSXSmmSC/b363Srz7bFYrHtJt39MS7ZHiYcQ4pZEAj6GxIh7ZQrRaDLBYGDufes3L6HE47BpvTsnocRDCPnd47DZ4PABD3CAPr71+u/Bo3viPCGEkH6JEg8hhBCXosRDCCHEpSjxEEIIcSlKPIQQQlyKEg8hhBCXotOpu+lhFuRz58X8HHlU4wIe3dgoLvfjrrE56jet1UYIIcSlaKqNEEKIS1HiIYQQ4lKUeAghhLgUJR5CCCEuRYmHEEKIS1HiIYQQ4lKUeAghhLgUJR5CCCEuRYmHEEKIS9GSOT1UVlaGNWvW4OLFi/Dw8EB6ejqWL18OLy+vLtvm5uZi+/btqKyshEQiweLFizF16lQX9NqxH374AceOHcO1a9egVCoRGRmJzMxMzJkzB2x2599R5s2bh6KiIrvyAwcOQCqVOrPL3Xbo0CGsWrXKrvyPf/wjPvjgA4dt++v+Ajr/2wPA22+/jUWLFnW4beXKlTh8+LBd+caNG/Hss8/2ah+7486dO9i1axeuXLmCW7duISYmBsePH7erd+bMGWzYsAG//fYbQkJCsGDBAsybN6/L19fr9di0aRMOHz4MtVoNqVSK999/H4mJic4Ix6qruIxGI3bv3o0zZ87gt99+g9FoREJCAt544w2MGjWqy9cfNGiQXZm3tzcuXbrUq3H0Nko8PaBSqTB//nyEh4dj48aNUCgUWLduHRQKBb788kuHbU+cOIEVK1Zg0aJFSE1NxcmTJ/HWW29BIBDg6aefdlEEHfvmm28QHh6Od999F2KxGBcuXMDf/vY3lJeXY8WKFQ7bPvHEE3Z1YmNjndndHtm5cyd8fHyszwMDAx3W78/7CwD+8pe/QKPR2JQdOXIE+/btQ1pamsO2kZGR+Oyzz2zKoqKieruL3XLr1i2cOXMGycnJMJlM6Gglr0uXLiE7OxvPP/88VqxYgYsXL2Lt2rXgcrnIzMx0+Prr1q1Dbm4uVq5ciYiICOzcuRMvv/wyjh49ipCQEGeF1WVcWq0WO3bswPTp05GVlQUul4vDhw/jlVdewbZt2zB+/Pguf8e8efOQkZFhfe7oS2K/wZAHtmPHDiY5OZmRy+XWsqNHjzIJCQnMzZs3HbZ99tlnmaVLl9qUvfLKK8zMmTOd0tcH0TYei7Vr1zJSqZRpbW3ttN3cuXOZRYsWObNrD+3gwYNMQkJChzE60p/3V2dmzpzJZGRkOKyzYsUKJj093UU96prRaLQ+7qxvWVlZzKxZs2zKVq9ezaSmptq0b6+mpoZJTExkcnJyrGVqtZoZMWIE8/HHH/dC7zvXVVwGg4FpamqyKTOZTMyMGTOYuXPndvn6CQkJzM6dO3unsy7kBqmx/8nPz8fIkSMREBBgLZs8eTL4fD7y8/M7bVdeXo7S0lKkp6fblGdkZKC4uBgKhcJpfe6OtvFYJCYmorW1FU1NTX3Qo77V3/dXR8rKylBcXIznnnuur7vyQLr6lq7T6VBYWGg3xZmRkYH6+npcu3at07YFBQUwGo02bYVCIcaPH+/w/dobuoqLw+HA19fXpozFYmHw4MGoq6tzZtf6FCWeHigpKUFcXJxNGZ/Ph0QiQWlpaaftLNvaT0FZXstR277y888/w8/PD2Kx2GG9oqIiPP7445BKpcjMzMT58+dd1MMHM23aNCQmJmLChAnYsmULDAZDp3XdcX8dPXoUbDYb06ZN67KuTCZDSkoKkpKSMH36dHz//fcu6GHPyGQy6PV6u30RHx8PwPG+KCkpQWBgIPz9/W3K4+LiUFZWBpPJ1PsdfggmkwmXLl3q9lT1V199haSkJKSkpGDJkiWQyWRO7uHDo2M8PaBSqSASiezKRSIRlEplp+0s29q3tXzjcdS2LxQXF+PQoUNYvHgxOBxOp/WefPJJPPfcc4iKikJDQwP27NmDhQsXYvfu3d06QOoKQUFBWLJkCYYOHQoOh4P8/Hxs3boVFRUVWL9+fYdt3G1/AcCxY8fw5JNPIjQ01GG9xMRESKVSxMXFQa1W48CBA1i2bBm0Wi1eeOEFF/W2+zrbF5bnjvaFSqWyOa5n4evrC71ej5aWFgiFwl7s7cP5+9//jtu3b2PNmjVd1p0+fTrGjRuHoKAglJSUYNu2bcjMzMSRI0e6PH7ZlyjxkA7V19dj6dKlkEqleO211xzWXbp0qc3ziRMn4rnnnsOWLVv6TeIZO3Ysxo4da32empoKHx8fbN68GdnZ2ZBIJH3Yu95x+fJlyGQyvP76613WXbBggc3zZ555BvPnz8fmzZv7ZeL5vSgqKsKnn36KhQsXIiUlpcv6H3/8sfVxSkoKRowYgWnTpuHbb7/Fm2++6cyuPhSaausBkUgElUplV65Sqezma9uybGvf1vJtzVFbV1Kr1Xjttdfg6emJbdu2gcfjPVB7Pp+PiRMnOpx37w+mTJkCAJ320132l8XRo0fh4eHR49Ohn332WVRVVfXLY1ed7QvLc0f7QiQSQa1W25UrlUrweDx4e3v3Yk977saNG8jOzsYzzzyDd955p0evER0djcTExH7/3qPE0wOxsbEoKSmxKdPpdJDJZIiJiem0nWVb+/loy2s5ausqra2t+NOf/gS5XI6dO3fazYv/nrjD/rIwGAz4/vvvMX78+H41bdRbJBIJeDye3b747bffADjeF7GxsZDL5XYnyJSUlCAqKqpfnH4sk8nw6quv4rHHHsMnn3wCFss9b3fdXX3/F3dDaWlpKCwsRGNjo7UsLy8POp3O4bUdkZGRiImJsTuIe/z4cUil0g7PKnMlg8GAN998E7/++iu+/vprRERE9Oh1dDodTp482W8uHu3Md999BxaLhSFDhnS4vb/vr7YKCgrQ2NjY47PZGIbBDz/8gIiIiH4VlwWfz8fIkSPxww8/2JQfP34cQUFBSEpK6rTtmDFjwGazbdo2Nzfj9OnTXV7r5Ar19fVYuHAhAgMDsXXrVvD5/B6/VmlpKX755Zd+/96jYzw9MGfOHOTk5CA7OxvZ2dmQy+VYv349pk6danO223vvvYfc3Fxcv37dWrZ06VIsW7YMEokEo0ePxqlTp3Du3Dns2LGjL0Kx8eGHH+Lf//433nnnHWi1Wly+fNm6LS4uDkKh0C6mn376CTt37sSkSZMQERGBhoYG7N27FxUVFfjwww/7KhQ7WVlZeOqpp5CQkAAWi4WzZ89i3759mDVrFiIjIwG43/5q6+jRo/Dz8+vwg3TLli3YunUr8vLyEBERgcrKSqxcuRLp6ekYOHAgVCoV9u/fj6KiInzyySd90Hvg7t27OHPmDACgsrISGo0GJ06cAABIpVJERERg8eLFmDt3LlavXo1p06bh4sWL2L9/Pz744AObUcukSZMQHh6OPXv2AABCQkIwZ84cfPbZZ+ByuQgPD8fu3bsB2B/rcnVcYrEYr776KuRyOVauXGkdwVkMGzas07h27doFmUyGp556CgEBASgpKcH27dvh7++PP/zhD06N62FR4ukBkUiEPXv24KOPPsKSJUusS+a0n5c1mUwwGo02ZVOmTIFWq8X27duxa9cuSCQSfP755/3iKviCggIAwKeffmq3be/evXjqqafsYgoKCoJer8eXX36JpqYmeHp6Ijk5GXv37sXw4cNd1veuxMTE4ODBg6itrYXBYEBUVBSWL19u88HjbvvLwvLtffr06R0ej2MYBkaj0XrVvEAggFAoxLZt2yCXy8Hj8fDYY49h27ZtmDBhgqu7DwCQy+V2B8Mtz9etW4cXXngBjz/+OLZu3YovvvgCubm5CA4OxqpVq+xWLTAajXanSK9atQre3t7YsGGDdcmcb775xqmrFnQnrhEjRuDGjRsAgMWLF9u1//XXX62P28cVHR2Nf/3rXzhx4gQ0Gg38/f2RmpqKP//5z11e/tDXWAzTwdoUhBBCiJPQMR5CCCEuRYmHEEKIS1HiIYQQ4lKUeAghhLgUJR5CCCEuRYmHEEKIS1HiIeR37NChQxg0aJDNxcKEOBslHkIIIS5FiYcQQohLUeIhhBDiUpR4CHGBuro6vP/++0hNTcWQIUMwZcoU7Nu3z7r9woULGDRoEI4ePYpNmzZhzJgxSE5ORlZWFsrKyuxer6ioCHPnzsWwYcMwfPhwvP7667h582aHv/eDDz5AWloahgwZggkTJmD16tXQaDQ29XQ6HdatW4eRI0di2LBhWLx4cb+8Lw95NNAioYQ4mVwux+zZs2E0GpGZmQmxWIzz58/jr3/9K5qampCdnW2t+/XXX8NkMiErKwtKpRJ79+7F/PnzratPA0BhYSGysrIwYMAAvPHGG2htbcW+ffuQmZmJAwcOIDo6GoB5uf0XX3wRjY2NeOmllxAfH4+6ujrk5eWhqanJ5r49a9euha+vL9544w1UVlZiz549+PDDD7FhwwbX/rHI7wNDCHGq1atXM6NHj2bkcrlN+fvvv88MHTqUUSqVTGFhIZOQkMCMGjWKUSqV1jr//e9/mYSEBOaLL76wlk2fPp0ZMWIEo1AorGW3b99mkpKSmCVLlljLVqxYwQwePJi5fPmyXZ9MJhPDMAxz8OBBJiEhgVmwYIG1jGEYZu3atUxiYiKjUqke/g9ASDs01UaIEzEMgx9//NF6GwWFQmH9SU1NhVarxZUrV6z1n3/+eYhEIuvzUaNGIT4+Hv/5z38AmKfOrl+/jhkzZtjcHTYqKgoTJkzA2bNnrcvn5+XlIS0tDcnJyXb9an+Hy1mzZtmUpaSkwGg0orKyslf+DoS0RVNthDiRQqGAUqnEwYMHcfDgwQ7ryOVyhIWFATAnkPaioqJQWFgIAKiqqgIA63RaWzExMfjxxx+td8bVaDSIj4/vVj8tv9/CkvyPlETTAAACKElEQVRUKlW32hPyICjxEOJElht3ZWRkYObMmR3WiYuLw+3bt13ZLTscDqfDcoZu10WcgBIPIU4UEBAAgUAAg8GA0aNHd1rPkng6OoOtrKwMERERAIDw8HCb+m2VlpbC29sb/v7+YLFYEAqFuHXrVi9EQUjvomM8hDgRh8PB5MmTcerUKestjttqf8rykSNHbKa3zp8/j1u3bmHcuHEAgODgYCQlJSE3NxdNTU3WejKZDKdPn8bYsWPB4XDAZrMxadIk5Ofn2xxDsqCRDOlLNOIhxMmWL1+OoqIizJ49Gy+++CLi4+OhVCpx48YN5OXlobi42FpXLBYjMzMTs2bNgkqlwp49exAUFISXX37ZWufdd99FVlYWZs+ejZdeesl6OrWHhweWLVtmrffWW2/h3LlzmDdvHmbPno24uDg0NDQgLy8PW7ZswYABA1z5ZyDEihIPIU4mFouxf/9+bN26FadOncI//vEP+Pr6IiYmBitXrrSp+9prr+H27dvYtWsXVCoVUlJSsHr1apsz2EaOHIndu3dj06ZN2LRpEzgcDlJSUvD222/bnHQQHByM/fv3Y+PGjfjuu++gUqkQHByMMWPG2LweIa7GYmjMTUifu3DhAubPn48vvvgC6enpfd0dQpyKjvEQQghxKUo8hBBCXIoSDyGEEJeiYzyEEEJcikY8hBBCXIoSDyGEEJeixEMIIcSlKPEQQghxKUo8hBBCXIoSDyGEEJf6f3eOaiL1/HCqAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zUGdtNa0S5BI",
        "colab_type": "text"
      },
      "source": [
        "### Save model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9zmmbqtdS35Z",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "model.save_weights('gdrive/My Drive/four_vtb_classifier_model.h5')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OYOdPaPCdDCj",
        "colab_type": "code",
        "outputId": "e35a2163-3f11-4e1e-843d-6c4695cf8513",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 72
        }
      },
      "source": [
        "evaluate_model = AttRNNSpeechModel(n_classes, dropout_rate=dropout_rate)\n",
        "evaluate_model.load_weights('gdrive/My Drive/four_vtb_classifier_model.h5')"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "tracking <tf.Variable 'mel_stft_2/real_kernels:0' shape=(1024, 1, 1, 513) dtype=float32> dft_real_kernels\n",
            "tracking <tf.Variable 'mel_stft_2/imag_kernels:0' shape=(1024, 1, 1, 513) dtype=float32> dft_imag_kernels\n",
            "tracking <tf.Variable 'mel_stft_2/Variable:0' shape=(513, 80) dtype=float32> freq2mel\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7H806kWpDoMV",
        "colab_type": "text"
      },
      "source": [
        "### Draw confusion matrix"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "x471y59uDpqj",
        "colab_type": "code",
        "outputId": "d511786f-0bb2-4f87-94a8-0e2ec9053f78",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "from sklearn.metrics import confusion_matrix\n",
        "# Get y_pred_test\n",
        "y_pred = []\n",
        "y_test = []\n",
        "for i in tqdm(range(validation_steps)):\n",
        "  data, label = next(test_generator)\n",
        "  four_class_probs = evaluate_model.predict(data)\n",
        "  y_pred.extend(np.argmax(four_class_probs, axis=-1))\n",
        "  y_test.extend(np.argmax(label, axis=-1))\n",
        "\n",
        "confusion_mat = confusion_matrix(y_test, y_pred)"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 21/21 [01:40<00:00,  4.80s/it]\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MNMl9ygFE2ek",
        "colab_type": "code",
        "outputId": "3a978c7a-492d-42a8-caee-6e36b2ef0669",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 273
        }
      },
      "source": [
        "import seaborn as sn\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "df_cm = pd.DataFrame(confusion_mat, range(confusion_mat.shape[0]), range(confusion_mat.shape[1]))\n",
        "sn.set(font_scale=1.4) # for label size\n",
        "sn.heatmap(df_cm, annot=True, cmap='Blues', fmt='d')\n",
        "plt.show()"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oPu8HeZhdrdJ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# debug_name_list = ['宝鐘マリン', '白上フブキ', '天音かなた', '夏色まつり']\n",
        "\n",
        "# up_total = 0\n",
        "\n",
        "# for i in range(len(y_pred)):\n",
        "#   if up_total == 50:\n",
        "#     pdb.set_trace()\n",
        "#   if y_pred[i] != y_test[i]:\n",
        "#     plt.plot(X_test[i], color = 'orange', label='incorrect data')\n",
        "#     plt.title(f\"pred: {y_pred[i]}, real: {y_test[i]}\")\n",
        "#     plt.show()\n",
        "#     print(f\" 预测: {debug_name_list[y_pred[i]]}, 实际: {debug_name_list[y_test[i]]}\")\n",
        "#     for j in range(4):\n",
        "#       print(f\"{debug_name_list[j]}: {four_class_probs[i][j]}\")\n",
        "#     IPython.display.display(IPython.display.Audio(np.asarray(X_test[i]).T, rate=8000))\n",
        "#     up_total += 1\n",
        "    "
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}