比较乱，有空成型再整理

杂项

一维二维卷积

https://segmentfault.com/a/1190000012731665

模型layers各个API

https://blog.csdn.net/zhzhx1204/article/details/77406697

loss

https://blog.csdn.net/u011311291/article/details/79956195

优化器

https://blog.csdn.net/brucewong0516/article/details/78838124

https://www.csdn.net/gather_27/MtTacg4sOTU3OS1ibG9n.html

https://blog.csdn.net/BVL10101111/article/details/72616378

参数调用

之前都是
from keras.applications import VGG16

conv_base = VGG16(weights='imagenet',
                 include_top=False,
                 input_shape=(150, 150, 3)
                 )
现在是，用于演示
from keras.applications.vgg16 import VGG16

vgg16 = VGG16(weights='imagenet')

model输入都是Tensor，处理方法：

from keras.preprocessing import image
#from keras.applications.vgg16 import preprocess_input, decode_predictions
import numpy as np
img_path = 'D:\\Jupyter\\elephant.jpg'
img = image.load_img(img_path, target_size=(150, 150))
img_tensor = image.img_to_array(img)
img_tensor = np.expand_dims(img_tensor, axis=0)
#x = preprocess_input(x)
img_tensor /= 255.
print(model.predict(img_tensor))

from keras.preprocessing import image
from keras.applications.vgg16 import preprocess_input, decode_predictions
import numpy as np
img_path = 'D:\\Jupyter\\elephant.jpg'
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
preds = vgg16.predict(x)
print('Predict:', decode_predictions(preds, top=3))

Keras中文文档

https://keras-cn.readthedocs.io/en/latest/other/regularizers/

Keras的后端—-更高层级的构建模块

https://keras.io/zh/backend/

模型

keras系列︱Application中五款已训练模型、VGG16框架（Sequential式、Model式）解读（二）

http://www.cnblogs.com/Anita9002/p/8136535.html
VGG16结构：
https://blog.csdn.net/cai13160674275/article/details/71078554

模型官方下载：

https://github.com/fchollet/deep-learning-models/releases

model
model.layers,
model.get_layer(layer_name).output

model.perdict()
https://blog.csdn.net/zds13257177985/article/details/80638384

layer
layer.name   
layer.input   #img_tensor[1, 150, 150, 3]
layer.output #model有n层,n个特征图,每个特征图有对应n_features特征数
#[1，size，size，n_features(特征数)]

读入图像相关

ImageDataGenerator两种方法

Vgg16处理输入图像那里

#书 p143
img_path = 'elephant.jpg'
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
preds = model.predict(x)   
#对批量进行预处理（使其颜色标准化）
x = preprocess_input(x)
#x[1,224,224,3]
#将预测向量解读为人类可读形式（其实就是one-hot解码）
predict = model.predict(x)
print('Predicted:', decode_predictions(preds, top=3)[0])

输入numpy矩阵【n,height,weight,channels】，label【1，categories_num】

训练集，验证集，测试集

在测试数据上最终评估模型

test_generator = test_datagen.flow_from_directory(
        test_dir,
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')

test_loss, test_acc = model.evaluate_generator(test_generator, steps=50)
print('test acc:', test_acc)