test_deep_SVDD.ipynb

{"nbformat":4,"nbformat_minor":0,"metadata":{"accelerator":"GPU","colab":{"name":"test_deep_SVDD_end.ipynb","provenance":[],"collapsed_sections":[]},"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":"3.6.9"}},"cells":[{"cell_type":"code","metadata":{"colab_type":"code","id":"nzHCR0JfI8VY","colab":{}},"source":["#!pip uninstall tensorflow"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"colab_type":"code","id":"s_ZglR8zI9ON","colab":{}},"source":["#!pip install tensorflow-gpu==2.0.0"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"colab_type":"code","id":"a80IPFtiYlm-","colab":{"base_uri":"https://localhost:8080/","height":33},"executionInfo":{"status":"ok","timestamp":1599008397776,"user_tz":-480,"elapsed":2957,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"b67b90ba-62c1-4554-93f9-e74e03135c31"},"source":["import tensorflow as tf\n","import tensorflow_probability as tfp\n","from tensorflow.keras import layers\n","print(tf.__version__)\n","import time\n","import numpy as np\n","from pathlib import Path\n","import matplotlib.pyplot as plt\n","from matplotlib import gridspec\n","import matplotlib.patches as mpatches\n","from sklearn.metrics import roc_auc_score\n","import random\n","import itertools\n","from matplotlib.patches import Ellipse\n","from mpl_toolkits.mplot3d import Axes3D\n","from scipy.optimize import curve_fit\n","from sklearn.metrics import confusion_matrix\n","\n"],"execution_count":1,"outputs":[{"output_type":"stream","text":["2.3.0\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"colab_type":"code","id":"dGMSBTSpj6jK","colab":{},"executionInfo":{"status":"ok","timestamp":1599008397778,"user_tz":-480,"elapsed":2950,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["size = 25"],"execution_count":2,"outputs":[]},{"cell_type":"code","metadata":{"colab_type":"code","id":"A2EyZoJUj6jM","colab":{"base_uri":"https://localhost:8080/","height":84},"executionInfo":{"status":"ok","timestamp":1599008403145,"user_tz":-480,"elapsed":8311,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"f6427126-1cb9-42a6-9ee4-57d56c1b3dac"},"source":["tf.test.is_gpu_available()"],"execution_count":3,"outputs":[{"output_type":"stream","text":["WARNING:tensorflow:From <ipython-input-3-17bb7203622b>:1: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version.\n","Instructions for updating:\n","Use `tf.config.list_physical_devices('GPU')` instead.\n"],"name":"stdout"},{"output_type":"execute_result","data":{"text/plain":["True"]},"metadata":{"tags":[]},"execution_count":3}]},{"cell_type":"markdown","metadata":{"colab_type":"text","id":"b-cF_s9oj6jO"},"source":["![title](擷取.PNG)"]},{"cell_type":"code","metadata":{"colab_type":"code","id":"IIift702j6jP","colab":{},"executionInfo":{"status":"ok","timestamp":1599008403146,"user_tz":-480,"elapsed":8305,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["def gernate_data(n, nu, seed, range_, multi, size):\n","    ## normal\n","    np.random.seed(seed=seed)\n","    y_train = np.ones((int(n*(1-nu)),))\n","    mean = np.random.uniform(-5, 5, size)\n","    cov_  = np.random.uniform(0, 1, (size, size))* 3\n","    cov = np.dot(cov_.T,cov_)\n","    x_train = np.random.multivariate_normal(mean, cov, (int(n*(1-nu))))\n","    x_train = x_train.astype('float32')\n","\n","    \n","    \n","    np.random.seed(seed=seed)\n","    x_trains_ = None ## normal + outlier\n","    for i in range(multi):\n","        mean = np.random.uniform(-range_, range_, size)\n","        cov_ = np.random.randn(size, size)\n","        cov = np.dot(cov_.T,cov_)\n","        x_train_ = np.random.multivariate_normal(mean, cov, (int(n*nu/multi)), check_valid  = 'raise')\n","        try:\n","            x_trains_ = np.vstack((x_trains_, x_train_))\n","        except:\n","            x_trains_ = x_train_\n","    ## 畫圖\n","    '''fig = plt.figure()\n","    ax = Axes3D(fig)\n","    ax.scatter(x_train[:, 0], x_train[:, 1],x_train[:, 2], s=0.9, label = 'normal')\n","    ax.scatter(x_trains_[:,0], x_trains_[:,1], x_trains_[:,2], s=0.9, label = 'outlier')\n","    plt.legend()\n","    plt.title('normal + outlier')\n","    plt.show()'''\n","\n","    y_trains_ = np.zeros(((len(x_trains_),)))\n","    x_train_mix = np.vstack((x_trains_, x_train)).astype('float32')\n","    y_train_mix = np.hstack((y_trains_, y_train)).astype('float32')\n","    \n","    return x_train_mix, y_train_mix"],"execution_count":4,"outputs":[]},{"cell_type":"markdown","metadata":{"colab_type":"text","id":"B1hkHlXiQ1io"},"source":["## 參數設定"]},{"cell_type":"code","metadata":{"colab_type":"code","id":"VW-G91MqQ1ip","colab":{},"executionInfo":{"status":"ok","timestamp":1599008403147,"user_tz":-480,"elapsed":8300,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["h_dim = [10, 5]\n","z_dim = h_dim[1]\n","regular = 1e-5\n","max_lr = 1e-6\n","base_lr = 1e-5\n","\n","\n","n_epochs = 1000\n","\n","optimizer = tf.keras.optimizers.Adam(lr = base_lr)"],"execution_count":5,"outputs":[]},{"cell_type":"code","metadata":{"colab_type":"code","id":"-mwMy0auj6je","colab":{},"executionInfo":{"status":"ok","timestamp":1599008403147,"user_tz":-480,"elapsed":8296,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["def result_plot (x_train_mix, y_train_mix, ae_model, center, radius = None):\n","    output = ae_model(x_train_mix, training=False)\n","    if radius:\n","        trained_radius = radius.numpy()\n","    else:\n","        trained_radius = get_radius(output, nu)\n","    r = trained_radius**2\n","    dists = distance(x_train_mix, ae_model, center)\n","    fig = plt.figure()\n","    ax = Axes3D(fig)\n","    ax.scatter(x_train_mix[dists<=r][:,0], x_train_mix[dists<=r][:,1],x_train_mix[dists<=r][:,2] ,s=0.9, label='normal')\n","    ax.scatter(x_train_mix[dists>=r][:,0], x_train_mix[dists>=r][:,1],x_train_mix[dists>=r][:,2] ,s=0.9, label='anomaly')\n","    plt.legend()\n","    plt.show()\n","    result(dists, y_train_mix.astype(int), r )\n","    return trained_radius"],"execution_count":6,"outputs":[]},{"cell_type":"markdown","metadata":{"colab_type":"text","id":"HQeT_nSgj6jg"},"source":["## 利用aurtoencoder"]},{"cell_type":"code","metadata":{"colab_type":"code","id":"wqAPZUPyj6jg","colab":{"base_uri":"https://localhost:8080/","height":318},"executionInfo":{"status":"ok","timestamp":1599008403148,"user_tz":-480,"elapsed":8290,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"c79d80f2-3da8-4376-ecdc-4581db44e8f3"},"source":["def make_ae_model():\n","\n","    inputs = tf.keras.Input(shape=(size,), name='inputs')\n","\n","    x = tf.keras.layers.Dense(h_dim[0], activation='linear', use_bias=False)(inputs)\n","    x = tf.keras.layers.Dense(h_dim[1], activation='linear', use_bias=False, name = 'hidden')(x)\n","    x = tf.keras.layers.Dense(h_dim[0], activation='linear', use_bias=False)(x)\n","    x = tf.keras.layers.Dense(size, activation='linear', use_bias=False)(x)\n","    model = tf.keras.Model(inputs=inputs, outputs=x)\n","\n","    return model\n","\n","ae_model = make_ae_model()\n","ae_model.summary()"],"execution_count":7,"outputs":[{"output_type":"stream","text":["Model: \"functional_1\"\n","_________________________________________________________________\n","Layer (type)                 Output Shape              Param #   \n","=================================================================\n","inputs (InputLayer)          [(None, 25)]              0         \n","_________________________________________________________________\n","dense (Dense)                (None, 10)                250       \n","_________________________________________________________________\n","hidden (Dense)               (None, 5)                 50        \n","_________________________________________________________________\n","dense_1 (Dense)              (None, 10)                50        \n","_________________________________________________________________\n","dense_2 (Dense)              (None, 25)                250       \n","=================================================================\n","Total params: 600\n","Trainable params: 600\n","Non-trainable params: 0\n","_________________________________________________________________\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"colab_type":"code","id":"iVQ8iex6j6ji","colab":{"base_uri":"https://localhost:8080/","height":318},"executionInfo":{"status":"ok","timestamp":1599008403149,"user_tz":-480,"elapsed":8284,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"b673b82c-562f-45b4-f728-34743379071c"},"source":["ae_model = make_ae_model()\n","ae_model.summary()"],"execution_count":8,"outputs":[{"output_type":"stream","text":["Model: \"functional_3\"\n","_________________________________________________________________\n","Layer (type)                 Output Shape              Param #   \n","=================================================================\n","inputs (InputLayer)          [(None, 25)]              0         \n","_________________________________________________________________\n","dense_3 (Dense)              (None, 10)                250       \n","_________________________________________________________________\n","hidden (Dense)               (None, 5)                 50        \n","_________________________________________________________________\n","dense_4 (Dense)              (None, 10)                50        \n","_________________________________________________________________\n","dense_5 (Dense)              (None, 25)                250       \n","=================================================================\n","Total params: 600\n","Trainable params: 600\n","Non-trainable params: 0\n","_________________________________________________________________\n"],"name":"stdout"}]},{"cell_type":"markdown","metadata":{"colab_type":"text","id":"uwrm-Sw7Q1ir"},"source":["## 建構模型"]},{"cell_type":"code","metadata":{"colab_type":"code","id":"iroW56blj6jq","colab":{"base_uri":"https://localhost:8080/","height":251},"executionInfo":{"status":"ok","timestamp":1599008403150,"user_tz":-480,"elapsed":6792,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"d2e0b82c-7d27-40a1-b6c3-dfd4d38e420a"},"source":["def make_model(model):\n","    \n","    model.trainable = True\n","    x = model.get_layer('hidden').output\n","\n","    model_out = tf.keras.Model(inputs = model.input, outputs = x)\n","    \n","    return model_out\n","\n","model = make_model(ae_model)\n","model.summary()"],"execution_count":9,"outputs":[{"output_type":"stream","text":["Model: \"functional_5\"\n","_________________________________________________________________\n","Layer (type)                 Output Shape              Param #   \n","=================================================================\n","inputs (InputLayer)          [(None, 25)]              0         \n","_________________________________________________________________\n","dense_3 (Dense)              (None, 10)                250       \n","_________________________________________________________________\n","hidden (Dense)               (None, 5)                 50        \n","=================================================================\n","Total params: 300\n","Trainable params: 300\n","Non-trainable params: 0\n","_________________________________________________________________\n"],"name":"stdout"}]},{"cell_type":"markdown","metadata":{"colab_type":"text","id":"GxKBUdIwj6js"},"source":["## Train deep-SVDD"]},{"cell_type":"code","metadata":{"id":"vMQpWAr7-JAi","colab_type":"code","colab":{},"executionInfo":{"status":"ok","timestamp":1599008403621,"user_tz":-480,"elapsed":5379,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["class train_SVDD():\n","    \n","    def __init__(self , objective , encoder, nu , z_dim = z_dim):\n","        \n","        self.objective = objective\n","        self.nu = nu\n","        self.z_dim = z_dim\n","        self.encoder = encoder\n","        self.optimizer = tf.keras.optimizers.Adam(lr = 1e-3, epsilon = 1e-6)\n","        self.center =  None #np.array(init_center_c(x_train, model = encoder, eps = 0.1))\n","        \n","        \n","    def init_center_c(self, data, eps = 0.1):\n","    \n","        #positive = tf.fill([self.z_dim], eps)\n","        #negative = tf.fill([self.z_dim], -eps)\n","        c = tf.zeros(shape = [self.z_dim], dtype = tf.float32, name = 'c')\n","        c = tf.reduce_mean( self.encoder(data, training = False), axis = 0, name = 'outputs') \n","        # If c_i is too close to 0, set to +-eps. Reason: a zero unit can be trivially matched with zero weights.\n","\n","        #c = tf.where(tf.logical_and(tf.less(c, 0), tf.greater(c, -eps)), negative, c)\n","        #c = tf.where(tf.logical_and(tf.greater(c, 0), tf.less(c, eps)), positive, c)\n","        \n","        return c\n","        \n","    def get_radius(self,batch_x):\n","        output = self.encoder(batch_x, training=False)\n","        dist = tf.reduce_sum(tf.math.square(output - self.center) , axis=1)\n","\n","        return tfp.stats.percentile(tf.sqrt(dist), (1-self.nu)*100)\n","\n","\n","    def loss(self, objective, dist, R = None):\n","\n","        if objective == 'soft-boundary':\n","            scores = dist - R**2\n","            avg_loss = R**2 + (1/nu) * tf.reduce_mean(tf.maximum(tf.zeros_like(scores), scores))\n","\n","        if objective=='deep-SVDD':\n","            avg_loss = tf.reduce_mean(dist)\n","        return avg_loss\n","\n","    @tf.function\n","    def train_model(self, batch_x, center, objective, R = None):\n","        with tf.GradientTape() as tape:\n","            output = self.encoder(batch_x, training=True)\n","            dist = tf.reduce_sum(tf.math.square(output - self.center) , axis=1)\n","            output_loss = self.loss(self.objective, dist, R)\n","\n","        grads = tape.gradient(output_loss, self.encoder.trainable_variables)\n","        self.optimizer.apply_gradients(zip(grads, self.encoder.trainable_variables))\n","\n","        return output_loss, dist\n","    def train(self, train_dataset, x_train, n_epochs = 150, warm_up_n_epochs = 25):\n","        \n","        if self.objective == 'soft-boundary':\n","            self.R = tf.zeros(shape = [1], dtype = tf.float32, name = 'R')\n","        else:\n","            self.R = None\n","        loss_list = []\n","        pre_loss = 100\n","        loss_now = 10\n","        epoch = 0\n","        self.center = np.array(self.init_center_c(tf.convert_to_tensor(x_train), eps = 0.1))\n","        #for epoch in range(n_epochs):\n","        while abs(pre_loss - loss_now) > 1e-6:\n","            pre_loss = loss_now\n","            epoch_loss_avg = tf.metrics.Mean()\n","\n","            for batch, (batch_x) in enumerate(train_dataset):\n","                loss_, dist = self.train_model(batch_x, self.center, self.objective, self.R )\n","                epoch_loss_avg(loss_)\n","                if (self.objective == 'soft-boundary') and (epoch >= warm_up_n_epochs):\n","                    self.R = self.get_radius(batch_x)\n","            #loss_list.append(loss_)\n","            loss_now = epoch_loss_avg.result()\n","            epoch +=1\n","        self.encoder.save_weights('oneclass_{}' .format(self.objective))\n","        #plt.plot(loss_list)\n","        #plt.show()\n","        \n","    def distance(self, data, load_model = True):\n","        if load_model:\n","            self.encoder.load_weights('oneclass_{}' .format\n","                                  (self.objective))\n","        output = self.encoder(data, training=False)\n","        dist = tf.reduce_sum(tf.math.square(output - self.center) , axis=1) ## dist**2\n","        \n","        return dist\n","    \n","    def decide_treshold(self, x_train):\n","        \n","        if self.objective == 'soft-boundary':\n","            radius_end = self.R.numpy()\n","        else:\n","            radius_end = self.get_radius(tf.convert_to_tensor(x_train)).numpy()\n","            \n","        return float(radius_end)\n","    def result_plot (self, dists, x_train_mix, radius ):\n","        r = radius**2\n","        fig = plt.figure()\n","        ax = Axes3D(fig)\n","        ax.scatter(x_train_mix[dists<=r][:,0], x_train_mix[dists<=r][:,1],x_train_mix[dists<=r][:,2] ,s=0.9, label='normal')\n","        plt.show()\n","        fig = plt.figure()\n","        ax = Axes3D(fig)\n","        ax.scatter(x_train_mix[dists<=r][:,0], x_train_mix[dists<=r][:,1],x_train_mix[dists<=r][:,2] ,s=0.9, label='normal')\n","        ax.scatter(x_train_mix[dists>=r][:,0], x_train_mix[dists>=r][:,1],x_train_mix[dists>=r][:,2] ,s=0.9, label='anomaly')\n","        plt.legend()\n","        plt.show()\n","    \n","    def train_result(self, x_train_mix, show_img = False, load_model = True):\n","        \n","        radius_end = self.decide_treshold(x_train_mix)\n","        if show_img:\n","            dist = self.distance(x_train_mix, load_model = load_model).numpy()\n","            #self.result_plot (dist, x_train_mix, radius_end )\n","\n","        return radius_end\n","    \n","    def result(self, dist, y_test, radius_end, show_img = False):\n","        \n","        y_true = y_test.astype(int)\n","        scores = dist - radius_end ** 2\n","        y_prob = 1/dist\n","        auc = roc_auc_score(y_true, y_prob)\n","        confusionmatrix = confusion_matrix(y_true, scores<0)\n","        if show_img:\n","            print('AUC:', auc)\n","            print('confusion_matrix:')\n","            print(confusionmatrix)\n","\n","        return auc, confusionmatrix \n","    \n","    def test_result(self, x_test_mix, y_test_mix, radius_end, show_img = False, load_model = True):\n","        \n","        dist = self.distance(x_test_mix, load_model = load_model).numpy()\n","        test_auc, confusionmatrix = self.result(dist, y_test_mix, radius_end, show_img = show_img)\n","        if show_img :\n","            self.result_plot (dist, x_test_mix, radius_end )\n","\n","        return test_auc, confusionmatrix \n","\n","\n","    "],"execution_count":10,"outputs":[]},{"cell_type":"code","metadata":{"id":"kp7oWKV-Lssh","colab_type":"code","colab":{},"executionInfo":{"status":"ok","timestamp":1599008403622,"user_tz":-480,"elapsed":4312,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["nu = 0.1\n","x_train_mix, y_train_mix = gernate_data(6000, nu, 16, range_ = 10, multi = 15, size = size)\n","x_test_mix, y_test_mix = gernate_data(10000, nu, 16,range_ = 10, multi = 15, size = size)\n","n_samples = len(x_train_mix)\n","\n","\n","batch_size = 2000\n","train_buf = x_train_mix.shape[0]\n","train_dataset = tf.data.Dataset.from_tensor_slices(x_train_mix)\n","train_dataset = train_dataset.shuffle(buffer_size=train_buf)\n","train_dataset = train_dataset.batch(batch_size)"],"execution_count":11,"outputs":[]},{"cell_type":"code","metadata":{"id":"J4_GcB1bu49m","colab_type":"code","colab":{},"executionInfo":{"status":"ok","timestamp":1599008403622,"user_tz":-480,"elapsed":3191,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["result = np.zeros((25, 3))"],"execution_count":12,"outputs":[]},{"cell_type":"code","metadata":{"id":"KSsqCd_K_mke","colab_type":"code","colab":{"base_uri":"https://localhost:8080/","height":435},"executionInfo":{"status":"ok","timestamp":1599009866123,"user_tz":-480,"elapsed":1465284,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"29473bbd-4a13-4c61-bce9-edc81f057a52"},"source":["objective='deep-SVDD'\n","for i in range(25):\n","    ## 訓練 autoencoder\n","    pre_time = time.time()\n","    ae_model = make_ae_model()\n","    ae_model.compile(optimizer=\"Adam\", loss=\"mse\")\n","    ae_model.fit(x_train_mix, x_train_mix, epochs = 100 ,verbose = 0)\n","    ## 訓練 SVDD\n","    model = make_model(ae_model)\n","    SVDD = train_SVDD(objective, model, nu = nu, z_dim = z_dim)\n","    SVDD.train(train_dataset, x_train_mix)\n","    radius_end = SVDD.train_result(x_train_mix)\n","    auc , matrix = SVDD.test_result(x_test_mix, y_test_mix, radius_end)\n","    train_time = time.time() - pre_time\n","    print('epoch = ', i, 'auc = ', auc)\n","    result[i,0] = i\n","    result[i,1] = auc\n","    result[i,2] = train_time"],"execution_count":13,"outputs":[{"output_type":"stream","text":["epoch =  0 auc =  0.8917977553310886\n","epoch =  1 auc =  0.8923272727272727\n","epoch =  2 auc =  0.8914381593714927\n","epoch =  3 auc =  0.8875933782267116\n","epoch =  4 auc =  0.8873414141414141\n","epoch =  5 auc =  0.8872951739618405\n","epoch =  6 auc =  0.8887519640852974\n","epoch =  7 auc =  0.8887196408529742\n","epoch =  8 auc =  0.887490684624018\n","epoch =  9 auc =  0.8873345679012347\n","epoch =  10 auc =  0.8932391694725028\n","epoch =  11 auc =  0.8878255892255892\n","epoch =  12 auc =  0.8877620089786756\n","epoch =  13 auc =  0.8927526374859709\n","epoch =  14 auc =  0.8896177328843995\n","epoch =  15 auc =  0.8911936026936027\n","epoch =  16 auc =  0.8873379349046016\n","epoch =  17 auc =  0.891829292929293\n","epoch =  18 auc =  0.8891592592592593\n","epoch =  19 auc =  0.8874216610549944\n","epoch =  20 auc =  0.8872253647586981\n","epoch =  21 auc =  0.88850645342312\n","epoch =  22 auc =  0.8896350168350168\n","epoch =  23 auc =  0.8941353535353534\n","epoch =  24 auc =  0.8872497194163861\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"bE1ACHlVQPUS","colab_type":"code","colab":{"base_uri":"https://localhost:8080/","height":33},"executionInfo":{"status":"ok","timestamp":1599009866127,"user_tz":-480,"elapsed":1462992,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"06309130-590d-45c7-c800-32a3d55bae5e"},"source":["np.mean(result, 0)"],"execution_count":14,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([12.        ,  0.88939923, 58.49415525])"]},"metadata":{"tags":[]},"execution_count":14}]},{"cell_type":"code","metadata":{"id":"TGXrUOrM0_D2","colab_type":"code","colab":{"base_uri":"https://localhost:8080/","height":33},"executionInfo":{"status":"ok","timestamp":1599009866127,"user_tz":-480,"elapsed":1462293,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"af79bb27-4b65-47f5-96e0-2adf629a3fc1"},"source":["np.std(result, 0)"],"execution_count":15,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([7.21110255e+00, 2.19416915e-03, 4.32613925e+01])"]},"metadata":{"tags":[]},"execution_count":15}]},{"cell_type":"code","metadata":{"id":"CGJQB6BX4djD","colab_type":"code","colab":{},"executionInfo":{"status":"ok","timestamp":1598980066343,"user_tz":-480,"elapsed":1227,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["nu = 0.2\n","x_train_mix, y_train_mix = gernate_data(6000, nu, 16, range_ = 10, multi = 15, size = size)\n","x_test_mix, y_test_mix = gernate_data(10000, nu, 16,range_ = 10, multi = 15, size = size)\n","n_samples = len(x_train_mix)\n","\n","\n","batch_size = 2000\n","train_buf = x_train_mix.shape[0]\n","train_dataset = tf.data.Dataset.from_tensor_slices(x_train_mix)\n","train_dataset = train_dataset.shuffle(buffer_size=train_buf)\n","train_dataset = train_dataset.batch(batch_size)"],"execution_count":138,"outputs":[]},{"cell_type":"code","metadata":{"id":"xMoCmxID-TI_","colab_type":"code","colab":{},"executionInfo":{"status":"ok","timestamp":1598980066345,"user_tz":-480,"elapsed":822,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["result = np.zeros((25, 3))"],"execution_count":139,"outputs":[]},{"cell_type":"code","metadata":{"id":"ws5CzLbW-VrU","colab_type":"code","colab":{"base_uri":"https://localhost:8080/","height":435},"executionInfo":{"status":"ok","timestamp":1598982183186,"user_tz":-480,"elapsed":2116265,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"3d587f8b-b526-4cd5-d57f-ed30ce7c606d"},"source":["objective='deep-SVDD'\n","for i in range(25):\n","    ## 訓練 autoencoder\n","    pre_time = time.time()\n","    ae_model = make_ae_model()\n","    ae_model.compile(optimizer=\"Adam\", loss=\"mse\")\n","    ae_model.fit(x_train_mix, x_train_mix, epochs = 100, verbose = 0)\n","    ## 訓練 SVDD\n","    model = make_model(ae_model)\n","    SVDD = train_SVDD(objective , model, nu = nu, z_dim = z_dim)\n","    SVDD.train(train_dataset, x_train_mix)\n","    radius_end = SVDD.train_result(x_train_mix)\n","    auc , matrix = SVDD.test_result(x_test_mix, y_test_mix, radius_end)\n","    train_time = time.time() - pre_time\n","    print('epoch = ', i, 'auc = ', auc)\n","    result[i,0] = i\n","    result[i,1] = auc\n","    result[i,2] = train_time"],"execution_count":140,"outputs":[{"output_type":"stream","text":["epoch =  0 auc =  0.901297932330827\n","epoch =  1 auc =  0.9033935463659146\n","epoch =  2 auc =  0.9011051378446114\n","epoch =  3 auc =  0.9009980576441103\n","epoch =  4 auc =  0.9009515037593985\n","epoch =  5 auc =  0.9054993107769422\n","epoch =  6 auc =  0.9010295112781955\n","epoch =  7 auc =  0.900843045112782\n","epoch =  8 auc =  0.9009862155388471\n","epoch =  9 auc =  0.9008941102756891\n","epoch =  10 auc =  0.9009328947368421\n","epoch =  11 auc =  0.9035059523809524\n","epoch =  12 auc =  0.9012363408521303\n","epoch =  13 auc =  0.9025967418546367\n","epoch =  14 auc =  0.901037656641604\n","epoch =  15 auc =  0.9034134085213033\n","epoch =  16 auc =  0.905084335839599\n","epoch =  17 auc =  0.9028522556390978\n","epoch =  18 auc =  0.9013080827067669\n","epoch =  19 auc =  0.9012379699248121\n","epoch =  20 auc =  0.9009692982456141\n","epoch =  21 auc =  0.901091290726817\n","epoch =  22 auc =  0.9010911967418546\n","epoch =  23 auc =  0.9010055764411028\n","epoch =  24 auc =  0.902831015037594\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"60Vz4frA-YWv","colab_type":"code","colab":{"base_uri":"https://localhost:8080/","height":33},"executionInfo":{"status":"ok","timestamp":1598982183197,"user_tz":-480,"elapsed":2106647,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"388af189-9db5-43e2-c430-cbb12a328ff8"},"source":["np.mean(result, 0)"],"execution_count":141,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([12.        ,  0.9018877 , 84.59634295])"]},"metadata":{"tags":[]},"execution_count":141}]},{"cell_type":"code","metadata":{"id":"17EZtHxe-com","colab_type":"code","colab":{"base_uri":"https://localhost:8080/","height":33},"executionInfo":{"status":"ok","timestamp":1598982183199,"user_tz":-480,"elapsed":2106203,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"bb961e2a-31f6-4397-d17e-36ef84c7879a"},"source":["np.std(result, 0)"],"execution_count":142,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([7.21110255e+00, 1.33808815e-03, 5.46605662e+01])"]},"metadata":{"tags":[]},"execution_count":142}]},{"cell_type":"code","metadata":{"id":"4mJbACBQey_c","colab_type":"code","colab":{},"executionInfo":{"status":"ok","timestamp":1598982183201,"user_tz":-480,"elapsed":2105687,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["nu = 0.3\n","x_train_mix, y_train_mix = gernate_data(6000, nu, 16, range_ = 10, multi = 15, size = size)\n","x_test_mix, y_test_mix = gernate_data(10000, nu, 16,range_ = 10, multi = 15, size = size)\n","n_samples = len(x_train_mix)\n","\n","\n","batch_size = 2000\n","train_buf = x_train_mix.shape[0]\n","train_dataset = tf.data.Dataset.from_tensor_slices(x_train_mix)\n","train_dataset = train_dataset.shuffle(buffer_size=train_buf)\n","train_dataset = train_dataset.batch(batch_size)"],"execution_count":143,"outputs":[]},{"cell_type":"code","metadata":{"id":"TfwlUAiDzV7U","colab_type":"code","colab":{},"executionInfo":{"status":"ok","timestamp":1598982183203,"user_tz":-480,"elapsed":2104865,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}}},"source":["result = np.zeros((25, 3))"],"execution_count":144,"outputs":[]},{"cell_type":"code","metadata":{"id":"Bpi8r1FfzXiy","colab_type":"code","colab":{"base_uri":"https://localhost:8080/","height":435},"executionInfo":{"status":"ok","timestamp":1598984208782,"user_tz":-480,"elapsed":4126517,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"87f01746-a547-47af-ad75-f4dafda823b3"},"source":["objective='deep-SVDD'\n","for i in range(25):\n","    ## 訓練 autoencoder\n","    pre_time = time.time()\n","    ae_model = make_ae_model()\n","    ae_model.compile(optimizer=\"Adam\", loss=\"mse\")\n","    ae_model.fit(x_train_mix, x_train_mix, epochs = 100, verbose = 0)\n","    ## 訓練 SVDD\n","    model = make_model(ae_model)\n","    SVDD = train_SVDD(objective , model, nu = nu, z_dim = z_dim)\n","    SVDD.train(train_dataset, x_train_mix)\n","    radius_end = SVDD.train_result(x_train_mix)\n","    auc , matrix = SVDD.test_result(x_test_mix, y_test_mix, radius_end)\n","    train_time = time.time() - pre_time\n","    print('epoch = ', i, 'auc = ', auc)\n","    result[i,0] = i\n","    result[i,1] = auc\n","    result[i,2] = train_time"],"execution_count":145,"outputs":[{"output_type":"stream","text":["epoch =  0 auc =  0.8695532857142857\n","epoch =  1 auc =  0.8686713333333334\n","epoch =  2 auc =  0.8685629047619048\n","epoch =  3 auc =  0.8707925714285715\n","epoch =  4 auc =  0.8682555238095239\n","epoch =  5 auc =  0.8687573333333333\n","epoch =  6 auc =  0.8689124285714287\n","epoch =  7 auc =  0.8689834761904763\n","epoch =  8 auc =  0.8689150476190477\n","epoch =  9 auc =  0.8722992857142855\n","epoch =  10 auc =  0.8688273333333333\n","epoch =  11 auc =  0.8690736190476192\n","epoch =  12 auc =  0.8681613809523809\n","epoch =  13 auc =  0.869062523809524\n","epoch =  14 auc =  0.868827857142857\n","epoch =  15 auc =  0.8693151904761907\n","epoch =  16 auc =  0.8687817619047619\n","epoch =  17 auc =  0.8693990476190476\n","epoch =  18 auc =  0.8686228571428573\n","epoch =  19 auc =  0.8686087142857142\n","epoch =  20 auc =  0.8686744285714285\n","epoch =  21 auc =  0.8723288095238095\n","epoch =  22 auc =  0.8682281428571429\n","epoch =  23 auc =  0.8686672380952382\n","epoch =  24 auc =  0.8679190238095238\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"p2Zon_7kzYwI","colab_type":"code","colab":{"base_uri":"https://localhost:8080/","height":33},"executionInfo":{"status":"ok","timestamp":1598984208786,"user_tz":-480,"elapsed":4126146,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"614d36f5-1188-4a32-e13f-5ca13c953da4"},"source":["np.mean(result, 0)"],"execution_count":146,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([12.        ,  0.86912804, 81.01051138])"]},"metadata":{"tags":[]},"execution_count":146}]},{"cell_type":"code","metadata":{"id":"AJQTNTIbzbZ9","colab_type":"code","colab":{"base_uri":"https://localhost:8080/","height":33},"executionInfo":{"status":"ok","timestamp":1598984208786,"user_tz":-480,"elapsed":4125778,"user":{"displayName":"Sunday michelle","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GgXr6CAFYJOVfTG2Le69HDmPDFgum5Bq4cKrdlwBg=s64","userId":"16105043169600541688"}},"outputId":"6556aa32-d1ad-4ecf-8434-e4f4cf309abf"},"source":["np.std(result, 0)"],"execution_count":147,"outputs":[{"output_type":"execute_result","data":{"text/plain":["array([7.21110255e+00, 1.08365651e-03, 5.12258896e+01])"]},"metadata":{"tags":[]},"execution_count":147}]},{"cell_type":"code","metadata":{"id":"b5PKSw1U2qZ6","colab_type":"code","colab":{}},"source":[""],"execution_count":null,"outputs":[]}]}