1 # GRADED FUNCTION: model 2 3 def model(X_train, Y_train, X_test, Y_test, learning_rate = 0.009, 4 num_epochs = 100, minibatch_size = 64, print_cost = True): 5 """ 6 Implements a three-layer ConvNet in Tensorflow: 7 CONV2D -> RELU -> MAXPOOL -> CONV2D -> RELU -> MAXPOOL -> FLATTEN -> FULLYCONNECTED 8 9 Arguments: 10 X_train -- training set, of shape (None, 64, 64, 3) 11 Y_train -- test set, of shape (None, n_y = 6) 12 X_test -- training set, of shape (None, 64, 64, 3) 13 Y_test -- test set, of shape (None, n_y = 6) 14 learning_rate -- learning rate of the optimization 15 num_epochs -- number of epochs of the optimization loop 16 minibatch_size -- size of a minibatch 17 print_cost -- True to print the cost every 100 epochs 18 19 Returns: 20 train_accuracy -- real number, accuracy on the train set (X_train) 21 test_accuracy -- real number, testing accuracy on the test set (X_test) 22 parameters -- parameters learnt by the model. They can then be used to predict. 23 """ 24 25 ops.reset_default_graph() # to be able to rerun the model without overwriting tf variables 26 tf.set_random_seed(1) # to keep results consistent (tensorflow seed) 27 seed = 3 # to keep results consistent (numpy seed) 28 (m, n_H0, n_W0, n_C0) = X_train.shape 29 n_y = Y_train.shape[1] 30 costs = [] # To keep track of the cost 31 32 # Create Placeholders of the correct shape 33 ### START CODE HERE ### (1 line) 34 X, Y = create_placeholders(n_H0, n_W0, n_C0, n_y) 35 ### END CODE HERE ### 36 37 # Initialize parameters 38 ### START CODE HERE ### (1 line) 39 parameters = initialize_parameters() #初始化filter 40 ### END CODE HERE ### 41 42 # Forward propagation: Build the forward propagation in the tensorflow graph 43 ### START CODE HERE ### (1 line) 44 Z3 = forward_propagation(X, parameters) 45 ### END CODE HERE ### 46 47 # Cost function: Add cost function to tensorflow graph 48 ### START CODE HERE ### (1 line) 49 cost = compute_cost(Z3, Y) # softmax -> average cost 50 ### END CODE HERE ### 51 52 # Backpropagation: Define the tensorflow optimizer. Use an AdamOptimizer that minimizes the cost. 53 ### START CODE HERE ### (1 line) 54 optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost) 55 ### END CODE HERE ### 56 57 # Initialize all the variables globally 58 init = tf.global_variables_initializer() 59 60 # Start the session to compute the tensorflow graph 61 with tf.Session() as sess: 62 63 # Run the initialization 64 sess.run(init) 65 66 # Do the training loop 67 for epoch in range(num_epochs): 68 69 minibatch_cost = 0. 70 num_minibatches = int(m / minibatch_size) # number of minibatches of size minibatch_size in the train set 71 seed = seed + 1 72 minibatches = random_mini_batches(X_train, Y_train, minibatch_size, seed) 73 74 for minibatch in minibatches: 75 76 # Select a minibatch 77 (minibatch_X, minibatch_Y) = minibatch 78 # IMPORTANT: The line that runs the graph on a minibatch. 79 # Run the session to execute the optimizer and the cost, the feedict should contain a minibatch for (X,Y). 80 ### START CODE HERE ### (1 line) 81 _ , temp_cost = sess.run([optimizer,cost], feed_dict={X: minibatch_X,Y: minibatch_Y}) 82 ### END CODE HERE ### 83 84 minibatch_cost += temp_cost / num_minibatches 85 86 87 # Print the cost every epoch 88 if print_cost == True and epoch % 5 == 0: 89 print ("Cost after epoch %i: %f" % (epoch, minibatch_cost)) 90 if print_cost == True and epoch % 1 == 0: 91 costs.append(minibatch_cost) 92 93 94 # plot the cost 95 plt.plot(np.squeeze(costs)) 96 plt.ylabel('cost') 97 plt.xlabel('iterations (per tens)') 98 plt.title("Learning rate =" + str(learning_rate)) 99 plt.show() 100 101 # Calculate the correct predictions 102 predict_op = tf.argmax(Z3, 1) 103 correct_prediction = tf.equal(predict_op, tf.argmax(Y, 1)) 104 105 # Calculate accuracy on the test set 106 accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float")) 107 print(accuracy) 108 train_accuracy = accuracy.eval({X: X_train, Y: Y_train}) 109 test_accuracy = accuracy.eval({X: X_test, Y: Y_test}) 110 print("Train Accuracy:", train_accuracy) 111 print("Test Accuracy:", test_accuracy) 112 113 return train_accuracy, test_accuracy, parameters
流程：placeholder for X and Y, parameters(weight of conv layer) initialization, forward_prop(X, parameters), compute_cost(Z, Y), backprop optimizer, global_variables.initializer, run init and optimizer, print the cost every epoch, caculate the crorrect predictions, caculate accuracy on the test set.
初始化：只需要初始conv layer的 weight，不用初始conv layer的 bias，不用初始fully connecyed layer的 weight 和 bias
完整版参见：https://www.cnblogs.com/CZiFan/p/9481110.html