我构建了一个多层感知机（MLP）并集成了一个称为轮廓系数函数的功能，并用它训练了网络。然而，在测试时，我需要删除一些权重（以10%为增量，从10%到100%），以研究这种调整对分类准确率的影响。
我期望得到10组测试准确率，每个百分比轮廓对应一组。
在标题“使用轮廓系数删除权重”下的函数（multilayer_perceptron_drop）会在测试时删除权重。我没有收到任何错误，但它返回的测试准确率为None。
# Multilayer Perceptron.from __future__ import print_function# Import MNIST datafrom tensorflow.examples.tutorials.mnist import input_datamnist = input_data.read_data_sets("/tmp/data/", one_hot=True)import timeimport numpy as npimport tensorflow as tfimport mathimport matplotlib.pyplot as pltfrom tensorflow.python.framework import ops# Parameterslearning_rate = 0.07training_epochs = 10batch_size = 64display_step = 1# Network Parametersn_1 = 100 # 1st layer number of neuronsn_2 = 100 # 2nd layer number of neuronsn_input = 784 # MNIST data input (img shape: 28*28)n_classes = 10 # MNIST total classes (0-9 digits)tic = time.time()# tf Graph inputtf.reset_default_graph()X = tf.placeholder(tf.float32, [None, n_input])Y = tf.placeholder(tf.float32, [None, n_classes])# Store layers weight & biasdef initialize_param(n_input, n_1, n_2, n_class): tf.set_random_seed(1) W1 = tf.get_variable("W1", shape = [n_input, n_1], initializer = tf.contrib.layers.xavier_initializer(seed = 1)) b1 = tf.get_variable("b1", shape = [n_1], initializer = tf.zeros_initializer()) W2 = tf.get_variable("W2", shape = [n_1, n_2], initializer = tf.contrib.layers.xavier_initializer(seed = 1)) b2 = tf.get_variable("b2", shape = [n_2], initializer = tf.zeros_initializer()) W3 = tf.get_variable("W3", shape = [n_2, n_class], initializer = tf.contrib.layers.xavier_initializer(seed = 1)) b3 = tf.get_variable("b3", shape = [n_class], initializer = tf.zeros_initializer()) parameters = {"W1": W1, "b1": b1, "W2": W2, "b2": b2, "W3": W3, "b3": b3} return parametersparameters = initialize_param(784, 100, 100, 10)# Create profile functiondef linear_func(n): return[np.float32(1.0 - 1.0 * i/n) for i in range(1, n + 1)]L = linear_func(100)# Create model with profile coefficientdef multilayer_perceptron(x): Z1 = tf.add(tf.matmul(x, parameters['W1']), parameters['b1']) # First fully connected layer with complete input channels A1 = tf.nn.relu(Z1) P1 = tf.multiply(L, A1) Z2 = tf.add(tf.matmul(P1, parameters['W2']), parameters['b2']) # Second fully connected layer with half input channels (0.5 neurons) A2 = tf.nn.relu(Z2) P2 = tf.multiply(L, A2) out_layer = tf.add(tf.matmul(P2, parameters['W3']), parameters['b3']) # Output fully connected layer with quater input channels (0.25 neuron for each class) return out_layer# Construct modellogits = multilayer_perceptron(X)# Define loss and optimizerloss_op = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits = logits, labels = Y))optimizer = tf.train.GradientDescentOptimizer(learning_rate = learning_rate).minimize(loss_op)# Initializing the variablesinit = tf.global_variables_initializer()with tf.Session() as sess: sess.run(init) # Training Loop cost = [] for epoch in range(training_epochs): avg_cost = 0. total_batch = int(mnist.train.num_examples/batch_size) # Loop over all batches for i in range(total_batch): batch_x, batch_y = mnist.train.next_batch(batch_size) # Run optimization op (backprop) and cost op (to get loss value) _, c = sess.run([optimizer, loss_op], feed_dict = {X: batch_x, Y: batch_y}) # Compute average loss avg_cost += c / total_batch cost.append(avg_cost) if i % 5000 == 0: pred = tf.nn.softmax(logits) # Apply softmax to logits correct_prediction = tf.equal(tf.argmax(pred, 1), tf.argmax(Y, 1)) accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float")) trian_accuracy = accuracy.eval({X: mnist.train.images, Y: mnist.train.labels}) # Display logs per epoch step if epoch % display_step == 0: print("Epoch:", '%03d' % (epoch + 1), "cost = {:.9f}".format(avg_cost)) # Create profile function def linear_func(n): return[np.float32(1.0 - 1.0 * i/n) for i in range(1, n + 1)] L = linear_func(100) def linear_profile(lp, n_1): p_L = tf.constant(L, shape = [1, 100]) L_11 = tf.constant(1.0, shape = [1, int(np.round((lp) * n_1))]) L_12 = tf.zeros(shape = [1, int(np.round((1 - lp) * n_1))]) L1 = tf.concat((L_11, L_12), axis = 1) p_L1 = tf.multiply(L1, p_L) return p_L1 pc = np.linspace(0.1, 1.0, 10) profile_1 = [] for i in pc: p_L1 = linear_profile(i, 100) profile = tf.stack(p_L1, axis = 0) profile_1.append(profile) profile_2 = tf.convert_to_tensor(profile_1, dtype=tf.float32) # Drop Weights with profile coefficients def multilayer_perceptron_drop(x): logist_1 = [] for j in range(len(profile_1)): Z_1 = tf.add(tf.matmul(x, parameters['W1']), parameters['b1']) A_1 = tf.nn.relu(Z_1) P_1 = tf.multiply(profile_2[j], A_1) Z_2 = tf.add(tf.matmul(A_1, parameters['W2']), parameters['b2']) A_2 = tf.nn.relu(Z_2) P_2 = tf.multiply(profile_2[j], A_2) out_layer = tf.add(tf.matmul(P_2, parameters['W3']), parameters['b3']) logist_1.append(out_layer) return logist_1 logist_1 = multilayer_perceptron_drop(X) #Retrieved model test_accuracy_2 = [] for k in range(len(logist_1)): pred_1 = tf.nn.softmax(logist_1[k]) correct_prediction_1 = tf.equal(tf.argmax(pred_1, 1), tf.argmax(Y, 1)) accuracy_1 = tf.reduce_mean(tf.cast(correct_prediction_1, "float")) test_accuracy_1 = accuracy_1.eval({X: mnist.test.images, Y: mnist.test.labels})*100 test_accuracy_2 = test_accuracy_2.append(test_accuracy_1) #test_accuracy_1 = sess.run(accuracy_1, feed_dict = {X: mnist.test.images, Y: mnist.test.labels}) sess.close() for l in range(len(pc)): print("Percentage_Profile:", '%03d' % (l + 1), "Test Accuracy = {}".format(test_accuracy_2)) #print('Test Accuracy: {}'.format(test_accuracy_2)) toc = time.time() print('Training Time is :' + str((toc - tic)/60) + 's') 输出:
Extracting /tmp/data/train-images-idx3-ubyte.gzExtracting /tmp/data/train-labels-idx1-ubyte.gzExtracting /tmp/data/t10k-images-idx3-ubyte.gzExtracting /tmp/data/t10k-labels-idx1-ubyte.gzEpoch: 001 cost = 0.463683842Epoch: 003 cost = 0.156443127Epoch: 005 cost = 0.108447251Epoch: 007 cost = 0.083334308Epoch: 009 cost = 0.064379380Percentage_Profile: 001 Test Accuracy = NonePercentage_Profile: 002 Test Accuracy = NonePercentage_Profile: 003 Test Accuracy = NonePercentage_Profile: 004 Test Accuracy = NonePercentage_Profile: 005 Test Accuracy = NonePercentage_Profile: 006 Test Accuracy = NonePercentage_Profile: 007 Test Accuracy = NonePercentage_Profile: 008 Test Accuracy = NonePercentage_Profile: 009 Test Accuracy = NonePercentage_Profile: 010 Test Accuracy = NoneTraining Time is :1.06416635116s回答：
更改这一行:
# 错误！`append`的结果是`None`，而不是列表test_accuracy_2 = test_accuracy_2.append(test_accuracy_1)改为…
# 正确。只需在列表中收集值test_accuracy_2.append(test_accuracy_1)