Keras简单使用

在keras中建立模型

相对于自己写机器学习相关的函数，keras更能快速搭建模型，流程如下：

通过调用下面的函数创建模型
通过调用 model.compile(optimizer = "...", loss = "...", metrics = ["accuracy"])编译模型
通过调用 model.fit(x = ..., y = ..., epochs = ..., batch_size = ...)在训练集上训练模型
通过调用model.evaluate(x = ..., y = ...)在测试集上测试模型

如果你想查阅更多有关model.compile(), model.fit(), model.evaluate() 的信息和它们的参数, 请参考官方文档 Keras documentation.

代码如下：

 1 def model(input_shape):
 2     # Define the input placeholder as a tensor with shape input_shape. Think of this as your input image!
 3     X_input = Input(input_shape)
 4 
 5     # Zero-Padding: pads the border of X_input with zeroes
 6     X = ZeroPadding2D((3, 3))(X_input)
 7 
 8     # CONV -> BN -> RELU Block applied to X
 9     X = Conv2D(32, (7, 7), strides = (1, 1), name = 'conv0')(X)
10     X = BatchNormalization(axis = 3, name = 'bn0')(X)
11     X = Activation('relu')(X)
12 
13     # MAXPOOL
14     X = MaxPooling2D((2, 2), name='max_pool')(X)
15 
16     # FLATTEN X (means convert it to a vector) + FULLYCONNECTED
17     X = Flatten()(X)
18     X = Dense(1, activation='sigmoid', name='fc')(X)
19 
20     # Create model. This creates your Keras model instance, you'll use this instance to train/test the model.
21     model = Model(inputs = X_input, outputs = X, name='HappyModel')
22 
23 return model

 1 step 1:
 2 happyModel = HappyModel(X_train.shape[1:]) # 只保留一个例子
 3 
 4 step 2:
 5 happyModel.compile(optimizer = 'sgd', loss = 'binary_crossentropy', metrics = ['accuracy'])
 6 
 7 step 3：
 8 happyModel.fit(x = X_train,y = Y_train, epochs = 5, batch_size = 16)
 9 
10 step 4：
11 preds =  happyModel.evaluate(x = X_test, y = Y_test)
12 # preds[0] = Loss
13 # preds[1] = Test Accuracy

此时，模型参数均已确定，可用来测试自己的图片

测试自己的图片

1 1 img_path = 'your picture path'
2 2 img = image.load_img(img_path, target_size=(64, 64))
3 3 imshow(img)
4 4 
5 5 x = image.img_to_array(img)
6 6 x = np.expand_dims(x, axis=0)
7 7 x = preprocess_input(x)
8 8 
9 9 print(happyModel.predict(x))

一些有用的函数(持续更新)

happyModel.summary()：统计并打印如下内容

plot_model()画出流程图

1 plot_model(happyModel, to_file='HappyModel.png')
2 SVG(model_to_dot(happyModel).create(prog='dot', format='svg'))