DL之CNN：利用卷积神经网络算法(2→2,基于Keras的API-Functional)利用MNIST(手写数字图片识别)数据集实现多分类预测
目录
输出结果
设计思路
核心代码
输出结果
下边两张图对应查看，可知，数字0有965个是被准确识别到！
1.10.0Size of:- Training-set:55000- Validation-set:5000- Test-set:10000Epoch 1/1128/55000 [..............................] - ETA: 14:24 - loss: 2.3439 - acc: 0.0938256/55000 [..............................] - ETA: 14:05 - loss: 2.2695 - acc: 0.1016384/55000 [..............................] - ETA: 13:20 - loss: 2.2176 - acc: 0.1302512/55000 [..............................] - ETA: 13:30 - loss: 2.1608 - acc: 0.2109640/55000 [..............................] - ETA: 13:29 - loss: 2.0849 - acc: 0.2500768/55000 [..............................] - ETA: 13:23 - loss: 2.0309 - acc: 0.2734896/55000 [..............................] - ETA: 13:30 - loss: 1.9793 - acc: 0.29461024/55000 [..............................] - ETA: 13:23 - loss: 1.9105 - acc: 0.33691152/55000 [..............................] - ETA: 13:22 - loss: 1.8257 - acc: 0.3776……53760/55000 [============================>.] - ETA: 18s - loss: 0.2106 - acc: 0.932953888/55000 [============================>.] - ETA: 16s - loss: 0.2103 - acc: 0.933054016/55000 [============================>.] - ETA: 14s - loss: 0.2100 - acc: 0.933154144/55000 [============================>.] - ETA: 13s - loss: 0.2096 - acc: 0.933354272/55000 [============================>.] - ETA: 11s - loss: 0.2092 - acc: 0.933454400/55000 [============================>.] - ETA: 9s - loss: 0.2089 - acc: 0.9335 54528/55000 [============================>.] - ETA: 7s - loss: 0.2086 - acc: 0.933654656/55000 [============================>.] - ETA: 5s - loss: 0.2082 - acc: 0.933754784/55000 [============================>.] - ETA: 3s - loss: 0.2083 - acc: 0.933754912/55000 [============================>.] - ETA: 1s - loss: 0.2082 - acc: 0.933755000/55000 [==============================] - 837s 15ms/step - loss: 0.2080 - acc: 0.933832/10000 [..............................] - ETA: 21s160/10000 [..............................] - ETA: 8s 288/10000 [..............................] - ETA: 6s448/10000 [>.............................] - ETA: 5s576/10000 [>.............................] - ETA: 5s736/10000 [=>............................] - ETA: 4s864/10000 [=>............................] - ETA: 4s1024/10000 [==>...........................] - ETA: 4s1152/10000 [==>...........................] - ETA: 4s1312/10000 [==>...........................] - ETA: 4s1440/10000 [===>..........................] - ETA: 4s1600/10000 [===>..........................] - ETA: 3s1728/10000 [====>.........................] - ETA: 3s……3008/10000 [========>.....................] - ETA: 3s3168/10000 [========>.....................] - ETA: 3s3296/10000 [========>.....................] - ETA: 3s3456/10000 [=========>....................] - ETA: 2s……5248/10000 [==============>...............] - ETA: 2s5376/10000 [===============>..............] - ETA: 2s5536/10000 [===============>..............] - ETA: 2s5664/10000 [===============>..............] - ETA: 1s5792/10000 [================>.............] - ETA: 1s……7360/10000 [=====================>........] - ETA: 1s7488/10000 [=====================>........] - ETA: 1s7648/10000 [=====================>........] - ETA: 1s7776/10000 [======================>.......] - ETA: 1s7936/10000 [======================>.......] - ETA: 0s8064/10000 [=======================>......] - ETA: 0s8224/10000 [=======================>......] - ETA: 0s……9760/10000 [============================>.] - ETA: 0s9920/10000 [============================>.] - ETA: 0s10000/10000 [==============================] - 4s 449us/steploss 0.05686537345089018acc 0.982acc: 98.20%[[ 965 0 4 0 0 0 4 1 2 4][ 0 1128 3 0 0 0 0 1 3 0][ 0 0 1028 0 0 0 0 1 3 0][ 0 0 10 991 0 2 0 2 3 2][ 0 0 3 0 967 0 1 1 1 9][ 2 0 1 7 1 863 5 1 4 8][ 2 3 0 0 3 2 946 0 2 0][ 0 1 17 1 1 0 0 987 2 19][ 2 0 9 2 0 1 0 1 955 4][ 1 4 3 2 8 0 0 0 1 990]]_________________________________________________________________Layer (type) Output Shape Param # =================================================================input_1 (InputLayer) (None, 784)0 _________________________________________________________________reshape (Reshape) (None, 28, 28, 1) 0 _________________________________________________________________layer_conv1 (Conv2D) (None, 28, 28, 16) 416 _________________________________________________________________max_pooling2d (MaxPooling2D) (None, 14, 14, 16) 0 _________________________________________________________________layer_conv2 (Conv2D) (None, 14, 14, 36) 14436_________________________________________________________________max_pooling2d_1 (MaxPooling2 (None, 7, 7, 36)0 _________________________________________________________________flatten (Flatten) (None, 1764) 0 _________________________________________________________________dense (Dense)(None, 128)225920 _________________________________________________________________dense_1 (Dense) (None, 10)1290=================================================================Total params: 242,062Trainable params: 242,062Non-trainable params: 0_________________________________________________________________(5, 5, 1, 16)(1, 28, 28, 16)
设计思路
核心代码
后期更新……
path_model = 'Functional_model.keras' from tensorflow.python.keras.models import load_model model2_1 = load_model(path_model)model_weights_path = 'Functional_model_weights.keras'model2_1.save_weights(model_weights_path ) model2_1.load_weights(model_weights_path, by_name=True ) model2_1.load_weights(model_weights_path) result = model.evaluate(x=data.x_test,y=data.y_test)for name, value in zip(model.metrics_names, result):print(name, value)print("{0}: {1:.2%}".format(model.metrics_names[1], result[1]))y_pred = model.predict(x=data.x_test) cls_pred = np.argmax(y_pred, axis=1) plot_example_errors(cls_pred) plot_confusion_matrix(cls_pred)images = data.x_test[0:9] cls_true = data.y_test_cls[0:9] y_pred = model.predict(x=images)cls_pred = np.argmax(y_pred, axis=1) title = 'MNIST(Sequential Model): plot predicted example, resl VS predict'plot_images(title, images=images,cls_true=cls_true,cls_pred=cls_pred)
DL之CNN：利用卷积神经网络算法(2→2 基于Keras的API-Functional)利用MNIST(手写数字图片识别)数据集实现多分类预测
