AlexNet網絡模型的詳細介紹參考ImageNet Classification with Deep Convolutional Neural Networks論文解讀，
TensorFlow的介紹參考TensorFlow簡介，
MNIST數據集的介紹參考MNIST數據集。
實驗大致步驟如下，
1 #coding=utf-8 2 from __future__ import print_function 3 4 from tensorflow.examples.tutorials.mnist import input_data 5 mnist = input_data.read_data_sets("/tmp/data", one_hot=True) 6 7 import tensorflow as tf 8 # 初始化網絡超參數 9 learning_rate = 0.001 10 training_iters = 200000 11 batch_size = 64 12 display_step = 20 13 14 # 初始化網絡參數 15 n_input = 784 # 輸入的維度 16 n_classes = 10 # 標籤的維度，即輸出的個數 17 dropout = 0.75 # Dropout隨機失活率 18 19 # 佔位符 20 x = tf.placeholder(tf.float32, [None, n_input]) 21 y = tf.placeholder(tf.float32, [None, n_classes]) 22 keep_prob = tf.placeholder(tf.float32) 23 24 # 卷積 25 def conv2d(name, l_input, w, b): 26 return tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(l_input, w, strides=[1, 1, 1, 1], padding='SAME'),b), name=name) 27 28 # 最大池化 29 def max_pool(name, l_input, k): 30 return tf.nn.max_pool(l_input, ksize=[1, k, k, 1], strides=[1, k, k, 1], padding='SAME', name=name) 31 32 # 歸一化 33 def norm(name, l_input, lsize=4): 34 return tf.nn.lrn(l_input, lsize, bias=1.0, alpha=0.001 / 9.0, beta=0.75, name=name) 35 36 # 初始化網絡的權重和偏置項 37 weights = { 38 'wc1': tf.Variable(tf.random_normal([11, 11, 1, 64])), 39 'wc2': tf.Variable(tf.random_normal([5, 5, 64, 192])), 40 'wc3': tf.Variable(tf.random_normal([3, 3, 192, 384])), 41 'wc4': tf.Variable(tf.random_normal([3, 3, 384, 384])), 42 'wc5': tf.Variable(tf.random_normal([3, 3, 384, 256])), 43 'wd1': tf.Variable(tf.random_normal([4*4*256, 4096])), 44 'wd2': tf.Variable(tf.random_normal([4096, 4096])), 45 'out': tf.Variable(tf.random_normal([4096, 10])) 46 } 47 biases = { 48 'bc1': tf.Variable(tf.random_normal([64])), 49 'bc2': tf.Variable(tf.random_normal([192])), 50 'bc3': tf.Variable(tf.random_normal([384])), 51 'bc4': tf.Variable(tf.random_normal([384])), 52 'bc5': tf.Variable(tf.random_normal([256])), 53 'bd1': tf.Variable(tf.random_normal([4096])), 54 'bd2': tf.Variable(tf.random_normal([4096])), 55 'out': tf.Variable(tf.random_normal([n_classes])) 56 } 57 58 # 定義整個網絡 59 def alex_net(_X, _weights, _biases, _dropout): 60 # 向量轉為矩陣 61 _X = tf.reshape(_X, shape=[-1, 28, 28, 1]) 62 63 # 第一層卷積 64 # 卷積 65 conv1 = conv2d('conv1', _X, _weights['wc1'], _biases['bc1']) 66 # 下採樣 67 pool1 = max_pool('pool1', conv1, k=2) 68 # 歸一化 69 norm1 = norm('norm1', pool1, lsize=4) 70 71 # 第二層卷積 72 # 卷積 73 conv2 = conv2d('conv2', norm1, _weights['wc2'], _biases['bc2']) 74 # 下採樣 75 pool2 = max_pool('pool2', conv2, k=2) 76 # 歸一化 77 norm2 = norm('norm2', pool2, lsize=4) 78 79 # 第三層卷積 80 # 卷積 81 conv3 = conv2d('conv3', norm2, _weights['wc3'], _biases['bc3']) 82 # 歸一化 83 norm3 = norm('norm3', conv3, lsize=4) 84 85 # 第四層卷積 86 # 卷積 87 conv4 = conv2d('conv4', norm3, _weights['wc4'], _biases['bc4']) 88 # 歸一化 89 norm4 = norm('norm4', conv4, lsize=4) 90 91 # 第五層卷積 92 # 卷積 93 conv5 = conv2d('conv5', norm4, _weights['wc5'], _biases['bc5']) 94 # 下採樣 95 pool5 = max_pool('pool5', conv5, k=2) 96 # 歸一化 97 norm5 = norm('norm5', pool5, lsize=4) 98 99 # 全連接層1100 dense1 = tf.reshape(norm5, [-1, _weights['wd1'].get_shape().as_list()[0]])101 dense1 = tf.nn.relu(tf.matmul(dense1, _weights['wd1']) + _biases['bd1'], name='fc1')102 dense1 = tf.nn.dropout(dense1, _dropout)103 104 # 全連接層2105 dense2 = tf.reshape(dense1, [-1, _weights['wd2'].get_shape().as_list()[0]])106 dense2 = tf.nn.relu(tf.matmul(dense1, _weights['wd2']) + _biases['bd2'], name='fc2') # Relu activation107 dense2 = tf.nn.dropout(dense2, _dropout)108 109 # 網絡輸出層110 out = tf.matmul(dense2, _weights['out']) + _biases['out']111 return out112 113 # 構建模型114 pred = alex_net(x, weights, biases, keep_prob)115 116 # 定義損失函數和學習步驟117 cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits = pred, labels = y))118 optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)119 120 # 測試網絡121 correct_pred = tf.equal(tf.argmax(pred,1), tf.argmax(y,1))122 accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))123 124 # 初始化所有的共享變量125 init = tf.initialize_all_variables()126 127 # session會話開始訓練128 with tf.Session() as sess:129 sess.run(init)130 step = 1131 # Keep training until reach max iterations132 while step * batch_size < training_iters:133 batch_xs, batch_ys = mnist.train.next_batch(batch_size)134 # 獲取批數據135 sess.run(optimizer, feed_dict={x: batch_xs, y: batch_ys, keep_prob: dropout})136 if step % display_step == 0:137 # 計算精度138 acc = sess.run(accuracy, feed_dict={x: batch_xs, y: batch_ys, keep_prob: 1.})139 # 計算損失值140 loss = sess.run(cost, feed_dict={x: batch_xs, y: batch_ys, keep_prob: 1.})141 print ("Iter " + str(step*batch_size) + ", Minibatch Loss = " + "{:.6f}".format(loss) + ", Training Accuracy = " + "{:.5f}".format(acc))142 step += 1143 # 計算測試精度144 print ("Testing Accuracy:", sess.run(accuracy, feed_dict={x: mnist.test.images[:256], y: mnist.test.labels[:256], keep_prob: 1.}))
運行效果如下，
WARNING:tensorflow:From alex_mnist.py:5: read_data_sets (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.Instructions for updating:Please use alternatives such as official/mnist/dataset.py from tensorflow/models.WARNING:tensorflow:From /home/cnu105/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:260: maybe_download (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.Instructions for updating:Please write your own downloading logic.WARNING:tensorflow:From /home/cnu105/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/base.py:252: _internal_retry.<locals>.wrap.<locals>.wrapped_fn (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.Instructions for updating:Please use urllib or similar directly.Successfully downloaded train-images-idx3-ubyte.gz 9912422 bytes.WARNING:tensorflow:From /home/cnu105/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:262: extract_images (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.Instructions for updating:Please use tf.data to implement this functionality.Extracting /tmp/data/train-images-idx3-ubyte.gzSuccessfully downloaded train-labels-idx1-ubyte.gz 28881 bytes.WARNING:tensorflow:From /home/cnu105/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:267: extract_labels (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.Instructions for updating:Please use tf.data to implement this functionality.Extracting /tmp/data/train-labels-idx1-ubyte.gzWARNING:tensorflow:From /home/cnu105/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:110: dense_to_one_hot (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.Instructions for updating:Please use tf.one_hot on tensors.Successfully downloaded t10k-images-idx3-ubyte.gz 1648877 bytes.Extracting /tmp/data/t10k-images-idx3-ubyte.gzSuccessfully downloaded t10k-labels-idx1-ubyte.gz 4542 bytes.Extracting /tmp/data/t10k-labels-idx1-ubyte.gzWARNING:tensorflow:From /home/cnu105/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:290: DataSet.__init__ (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.Instructions for updating:Please use alternatives such as official/mnist/dataset.py from tensorflow/models.WARNING:tensorflow:From alex_mnist.py:117: softmax_cross_entropy_with_logits (from tensorflow.python.ops.nn_ops) is deprecated and will be removed in a future version.Instructions for updating:Future major versions of TensorFlow will allow gradients to flowinto the labels input on backprop by default.See @{tf.nn.softmax_cross_entropy_with_logits_v2}.WARNING:tensorflow:From /home/cnu105/anaconda3/lib/python3.6/site-packages/tensorflow/python/util/tf_should_use.py:118: initialize_all_variables (from tensorflow.python.ops.variables) is deprecated and will be removed after 2017-03-02.Instructions for updating:Use `tf.global_variables_initializer` instead.2019-06-02 11:17:06.612024: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA2019-06-02 11:17:11.746821: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1392] Found device 0 with properties: name: GeForce GTX TITAN X major: 5 minor: 2 memoryClockRate(GHz): 1.076pciBusID: 0000:04:00.0totalMemory: 11.93GiB freeMemory: 11.81GiB2019-06-02 11:17:11.987157: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1392] Found device 1 with properties: name: GeForce GTX TITAN X major: 5 minor: 2 memoryClockRate(GHz): 1.076pciBusID: 0000:05:00.0totalMemory: 11.93GiB freeMemory: 11.81GiB2019-06-02 11:17:12.242069: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1392] Found device 2 with properties: name: GeForce GTX TITAN X major: 5 minor: 2 memoryClockRate(GHz): 1.076pciBusID: 0000:08:00.0totalMemory: 11.93GiB freeMemory: 11.81GiB2019-06-02 11:17:12.494381: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1392] Found device 3 with properties: name: GeForce GTX TITAN X major: 5 minor: 2 memoryClockRate(GHz): 1.076pciBusID: 0000:09:00.0totalMemory: 11.93GiB freeMemory: 11.81GiB2019-06-02 11:17:12.769954: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:897] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero2019-06-02 11:17:12.771926: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1392] Found device 4 with properties: name: GeForce GTX TITAN X major: 5 minor: 2 memoryClockRate(GHz): 1.076pciBusID: 0000:83:00.0totalMemory: 11.93GiB freeMemory: 11.81GiB2019-06-02 11:17:13.045728: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:897] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero2019-06-02 11:17:13.047507: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1392] Found device 5 with properties: name: GeForce GTX TITAN X major: 5 minor: 2 memoryClockRate(GHz): 1.076pciBusID: 0000:84:00.0totalMemory: 11.93GiB freeMemory: 11.81GiB2019-06-02 11:17:13.330365: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:897] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero2019-06-02 11:17:13.332353: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1392] Found device 6 with properties: name: GeForce GTX TITAN X major: 5 minor: 2 memoryClockRate(GHz): 1.076pciBusID: 0000:87:00.0totalMemory: 11.93GiB freeMemory: 11.81GiB2019-06-02 11:17:13.647518: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:897] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero2019-06-02 11:17:13.649454: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1392] Found device 7 with properties: name: GeForce GTX TITAN X major: 5 minor: 2 memoryClockRate(GHz): 1.076pciBusID: 0000:88:00.0totalMemory: 11.93GiB freeMemory: 11.81GiB2019-06-02 11:17:13.656551: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1471] Adding visible gpu devices: 0, 1, 2, 3, 4, 5, 6, 72019-06-02 11:17:16.979112: I tensorflow/core/common_runtime/gpu/gpu_device.cc:952] Device interconnect StreamExecutor with strength 1 edge matrix:2019-06-02 11:17:16.979177: I tensorflow/core/common_runtime/gpu/gpu_device.cc:958] 0 1 2 3 4 5 6 7 2019-06-02 11:17:16.979185: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 0: N Y Y Y N N N N 2019-06-02 11:17:16.979191: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 1: Y N Y Y N N N N 2019-06-02 11:17:16.979195: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 2: Y Y N Y N N N N 2019-06-02 11:17:16.979200: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 3: Y Y Y N N N N N 2019-06-02 11:17:16.979204: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 4: N N N N N Y Y Y 2019-06-02 11:17:16.979208: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 5: N N N N Y N Y Y 2019-06-02 11:17:16.979216: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 6: N N N N Y Y N Y 2019-06-02 11:17:16.979222: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 7: N N N N Y Y Y N 2019-06-02 11:17:16.981134: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 11432 MB memory) -> physical GPU (device: 0, name: GeForce GTX TITAN X, pci bus id: 0000:04:00.0, compute capability: 5.2)2019-06-02 11:17:18.523499: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:1 with 11432 MB memory) -> physical GPU (device: 1, name: GeForce GTX TITAN X, pci bus id: 0000:05:00.0, compute capability: 5.2)2019-06-02 11:17:19.855094: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:2 with 11432 MB memory) -> physical GPU (device: 2, name: GeForce GTX TITAN X, pci bus id: 0000:08:00.0, compute capability: 5.2)2019-06-02 11:17:21.862306: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:3 with 11432 MB memory) -> physical GPU (device: 3, name: GeForce GTX TITAN X, pci bus id: 0000:09:00.0, compute capability: 5.2)2019-06-02 11:17:23.769646: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:4 with 11432 MB memory) -> physical GPU (device: 4, name: GeForce GTX TITAN X, pci bus id: 0000:83:00.0, compute capability: 5.2)2019-06-02 11:17:25.666274: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:5 with 11432 MB memory) -> physical GPU (device: 5, name: GeForce GTX TITAN X, pci bus id: 0000:84:00.0, compute capability: 5.2)2019-06-02 11:17:27.601951: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:6 with 11432 MB memory) -> physical GPU (device: 6, name: GeForce GTX TITAN X, pci bus id: 0000:87:00.0, compute capability: 5.2)2019-06-02 11:17:29.557999: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1084] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:7 with 11432 MB memory) -> physical GPU (device: 7, name: GeForce GTX TITAN X, pci bus id: 0000:88:00.0, compute capability: 5.2)Iter 1280, Minibatch Loss = 334336.312500, Training Accuracy = 0.53125Iter 2560, Minibatch Loss = 160650.203125, Training Accuracy = 0.65625Iter 3840, Minibatch Loss = 58704.535156, Training Accuracy = 0.84375Iter 5120, Minibatch Loss = 108148.156250, Training Accuracy = 0.81250Iter 6400, Minibatch Loss = 28172.527344, Training Accuracy = 0.89062Iter 7680, Minibatch Loss = 77572.265625, Training Accuracy = 0.81250Iter 8960, Minibatch Loss = 34307.992188, Training Accuracy = 0.90625Iter 10240, Minibatch Loss = 63918.812500, Training Accuracy = 0.84375Iter 11520, Minibatch Loss = 42227.984375, Training Accuracy = 0.89062Iter 12800, Minibatch Loss = 44791.882812, Training Accuracy = 0.87500Iter 14080, Minibatch Loss = 31170.875000, Training Accuracy = 0.90625Iter 15360, Minibatch Loss = 37148.425781, Training Accuracy = 0.87500Iter 16640, Minibatch Loss = 24845.406250, Training Accuracy = 0.92188Iter 17920, Minibatch Loss = 16903.435547, Training Accuracy = 0.93750Iter 19200, Minibatch Loss = 64891.562500, Training Accuracy = 0.92188Iter 20480, Minibatch Loss = 6705.311523, Training Accuracy = 0.96875Iter 21760, Minibatch Loss = 13014.283203, Training Accuracy = 0.95312Iter 23040, Minibatch Loss = 27697.880859, Training Accuracy = 0.93750Iter 24320, Minibatch Loss = 22059.644531, Training Accuracy = 0.89062Iter 25600, Minibatch Loss = 38557.429688, Training Accuracy = 0.92188Iter 26880, Minibatch Loss = 27364.382812, Training Accuracy = 0.93750Iter 28160, Minibatch Loss = 11737.021484, Training Accuracy = 0.92188Iter 29440, Minibatch Loss = 22346.703125, Training Accuracy = 0.92188Iter 30720, Minibatch Loss = 31779.984375, Training Accuracy = 0.87500Iter 32000, Minibatch Loss = 2139.180664, Training Accuracy = 0.98438Iter 33280, Minibatch Loss = 15727.681641, Training Accuracy = 0.92188Iter 34560, Minibatch Loss = 49101.921875, Training Accuracy = 0.89062Iter 35840, Minibatch Loss = 14567.428711, Training Accuracy = 0.92188Iter 37120, Minibatch Loss = 56209.976562, Training Accuracy = 0.92188Iter 38400, Minibatch Loss = 9376.914062, Training Accuracy = 0.96875Iter 39680, Minibatch Loss = 6285.625488, Training Accuracy = 0.96875Iter 40960, Minibatch Loss = 8804.068359, Training Accuracy = 0.93750Iter 42240, Minibatch Loss = 20884.656250, Training Accuracy = 0.95312Iter 43520, Minibatch Loss = 4721.957031, Training Accuracy = 0.98438Iter 44800, Minibatch Loss = 18102.578125, Training Accuracy = 0.89062Iter 46080, Minibatch Loss = 1833.863403, Training Accuracy = 0.98438Iter 47360, Minibatch Loss = 1440.751953, Training Accuracy = 0.98438Iter 48640, Minibatch Loss = 10380.244141, Training Accuracy = 0.92188Iter 49920, Minibatch Loss = 3477.957031, Training Accuracy = 0.98438Iter 51200, Minibatch Loss = 7075.423828, Training Accuracy = 0.98438Iter 52480, Minibatch Loss = 16556.628906, Training Accuracy = 0.95312Iter 53760, Minibatch Loss = 22188.033203, Training Accuracy = 0.90625Iter 55040, Minibatch Loss = 19870.359375, Training Accuracy = 0.95312Iter 56320, Minibatch Loss = 4704.218750, Training Accuracy = 0.98438Iter 57600, Minibatch Loss = 17331.410156, Training Accuracy = 0.89062Iter 58880, Minibatch Loss = 3667.264648, Training Accuracy = 0.95312Iter 60160, Minibatch Loss = 19282.742188, Training Accuracy = 0.93750Iter 61440, Minibatch Loss = 9652.192383, Training Accuracy = 0.93750Iter 62720, Minibatch Loss = 9680.341797, Training Accuracy = 0.96875Iter 64000, Minibatch Loss = 18480.486328, Training Accuracy = 0.96875Iter 65280, Minibatch Loss = 20012.876953, Training Accuracy = 0.92188Iter 66560, Minibatch Loss = 15355.311523, Training Accuracy = 0.93750Iter 67840, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 69120, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 70400, Minibatch Loss = 4498.256348, Training Accuracy = 0.95312Iter 71680, Minibatch Loss = 6279.625000, Training Accuracy = 0.95312Iter 72960, Minibatch Loss = 12970.482422, Training Accuracy = 0.92188Iter 74240, Minibatch Loss = 5218.539062, Training Accuracy = 0.96875Iter 75520, Minibatch Loss = 15303.201172, Training Accuracy = 0.93750Iter 76800, Minibatch Loss = 3257.265137, Training Accuracy = 0.95312Iter 78080, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 79360, Minibatch Loss = 14279.500000, Training Accuracy = 0.95312Iter 80640, Minibatch Loss = 8679.945312, Training Accuracy = 0.98438Iter 81920, Minibatch Loss = 7986.568359, Training Accuracy = 0.96875Iter 83200, Minibatch Loss = 3156.044434, Training Accuracy = 0.96875Iter 84480, Minibatch Loss = 13759.476562, Training Accuracy = 0.89062Iter 85760, Minibatch Loss = 6791.756836, Training Accuracy = 0.93750Iter 87040, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 88320, Minibatch Loss = 14531.399414, Training Accuracy = 0.93750Iter 89600, Minibatch Loss = 9443.058594, Training Accuracy = 0.95312Iter 90880, Minibatch Loss = 11772.400391, Training Accuracy = 0.93750Iter 92160, Minibatch Loss = 691.387207, Training Accuracy = 0.98438Iter 93440, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 94720, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 96000, Minibatch Loss = 5482.347168, Training Accuracy = 0.98438Iter 97280, Minibatch Loss = 4430.060059, Training Accuracy = 0.96875Iter 98560, Minibatch Loss = 3621.751465, Training Accuracy = 0.96875Iter 99840, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 101120, Minibatch Loss = 8754.013672, Training Accuracy = 0.95312Iter 102400, Minibatch Loss = 1073.010498, Training Accuracy = 0.98438Iter 103680, Minibatch Loss = 11556.099609, Training Accuracy = 0.93750Iter 104960, Minibatch Loss = 3953.391357, Training Accuracy = 0.96875Iter 106240, Minibatch Loss = 9589.357422, Training Accuracy = 0.93750Iter 107520, Minibatch Loss = 3694.890137, Training Accuracy = 0.98438Iter 108800, Minibatch Loss = 16642.496094, Training Accuracy = 0.95312Iter 110080, Minibatch Loss = 3514.834961, Training Accuracy = 0.96875Iter 111360, Minibatch Loss = 5243.551758, Training Accuracy = 0.92188Iter 112640, Minibatch Loss = 1790.374512, Training Accuracy = 0.98438Iter 113920, Minibatch Loss = 61.289795, Training Accuracy = 0.98438Iter 115200, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 116480, Minibatch Loss = 6023.192383, Training Accuracy = 0.93750Iter 117760, Minibatch Loss = 1192.194824, Training Accuracy = 0.98438Iter 119040, Minibatch Loss = 1352.831055, Training Accuracy = 0.96875Iter 120320, Minibatch Loss = 4922.876465, Training Accuracy = 0.92188Iter 121600, Minibatch Loss = 677.854614, Training Accuracy = 0.98438Iter 122880, Minibatch Loss = 6639.125977, Training Accuracy = 0.95312Iter 124160, Minibatch Loss = 1121.854980, Training Accuracy = 0.98438Iter 125440, Minibatch Loss = 2850.864502, Training Accuracy = 0.93750Iter 126720, Minibatch Loss = 24656.960938, Training Accuracy = 0.92188Iter 128000, Minibatch Loss = 7486.107910, Training Accuracy = 0.98438Iter 129280, Minibatch Loss = 9559.851562, Training Accuracy = 0.93750Iter 130560, Minibatch Loss = 8428.903320, Training Accuracy = 0.93750Iter 131840, Minibatch Loss = 6142.676758, Training Accuracy = 0.93750Iter 133120, Minibatch Loss = 8347.507812, Training Accuracy = 0.96875Iter 134400, Minibatch Loss = 3275.400146, Training Accuracy = 0.96875Iter 135680, Minibatch Loss = 1656.050293, Training Accuracy = 0.96875Iter 136960, Minibatch Loss = 8057.255371, Training Accuracy = 0.95312Iter 138240, Minibatch Loss = 4164.952637, Training Accuracy = 0.98438Iter 139520, Minibatch Loss = 6530.175781, Training Accuracy = 0.92188Iter 140800, Minibatch Loss = 3056.857178, Training Accuracy = 0.95312Iter 142080, Minibatch Loss = 3378.891113, Training Accuracy = 0.98438Iter 143360, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 144640, Minibatch Loss = 7268.273926, Training Accuracy = 0.96875Iter 145920, Minibatch Loss = 1075.168457, Training Accuracy = 0.98438Iter 147200, Minibatch Loss = 3567.811279, Training Accuracy = 0.96875Iter 148480, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 149760, Minibatch Loss = 6473.182617, Training Accuracy = 0.93750Iter 151040, Minibatch Loss = 992.911133, Training Accuracy = 0.95312Iter 152320, Minibatch Loss = 2331.686523, Training Accuracy = 0.96875Iter 153600, Minibatch Loss = 100.348145, Training Accuracy = 0.98438Iter 154880, Minibatch Loss = 1558.047974, Training Accuracy = 0.95312Iter 156160, Minibatch Loss = 2154.779053, Training Accuracy = 0.95312Iter 157440, Minibatch Loss = 330.975830, Training Accuracy = 0.98438Iter 158720, Minibatch Loss = 2498.369873, Training Accuracy = 0.98438Iter 160000, Minibatch Loss = 3539.295654, Training Accuracy = 0.93750Iter 161280, Minibatch Loss = 7858.209961, Training Accuracy = 0.96875Iter 162560, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 163840, Minibatch Loss = 4297.341309, Training Accuracy = 0.95312Iter 165120, Minibatch Loss = 212.943359, Training Accuracy = 0.98438Iter 166400, Minibatch Loss = 6788.069336, Training Accuracy = 0.92188Iter 167680, Minibatch Loss = 6468.097168, Training Accuracy = 0.96875Iter 168960, Minibatch Loss = 5101.724609, Training Accuracy = 0.95312Iter 170240, Minibatch Loss = 9081.995117, Training Accuracy = 0.90625Iter 171520, Minibatch Loss = 5971.581055, Training Accuracy = 0.93750Iter 172800, Minibatch Loss = 7561.061035, Training Accuracy = 0.92188Iter 174080, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 175360, Minibatch Loss = 506.290039, Training Accuracy = 0.96875Iter 176640, Minibatch Loss = 5510.062988, Training Accuracy = 0.93750Iter 177920, Minibatch Loss = 3553.265137, Training Accuracy = 0.95312Iter 179200, Minibatch Loss = 4029.658936, Training Accuracy = 0.92188Iter 180480, Minibatch Loss = 8109.407227, Training Accuracy = 0.95312Iter 181760, Minibatch Loss = 806.317383, Training Accuracy = 0.98438Iter 183040, Minibatch Loss = 3882.623779, Training Accuracy = 0.96875Iter 184320, Minibatch Loss = 10893.535156, Training Accuracy = 0.93750Iter 185600, Minibatch Loss = 2430.192383, Training Accuracy = 0.96875Iter 186880, Minibatch Loss = 2718.515869, Training Accuracy = 0.98438Iter 188160, Minibatch Loss = 5090.834961, Training Accuracy = 0.92188Iter 189440, Minibatch Loss = 3633.469238, Training Accuracy = 0.96875Iter 190720, Minibatch Loss = 0.000000, Training Accuracy = 1.00000Iter 192000, Minibatch Loss = 297.924072, Training Accuracy = 0.98438Iter 193280, Minibatch Loss = 2371.230713, Training Accuracy = 0.95312Iter 194560, Minibatch Loss = 3749.918213, Training Accuracy = 0.96875Iter 195840, Minibatch Loss = 2433.742676, Training Accuracy = 0.98438Iter 197120, Minibatch Loss = 921.659790, Training Accuracy = 0.96875Iter 198400, Minibatch Loss = 3149.487549, Training Accuracy = 0.98438Iter 199680, Minibatch Loss = 30.536133, Training Accuracy = 0.98438Testing Accuracy: 0.96484375
註：在實驗過程中，可以使用nvidia-smi命令了解GPU使用情況，如下所示。