tensorflow的容器Sequential 笔记心得

tensorflow的Sequential不能够改名字，源代码处仅为可读，可以修改为读写，layer = model.layers,layer.name。 Sequential可以添加模型，层，比较灵活。但是，同一个模型因为具有相同的名字，重复添加只保留第一次添加，尝试copy.deepcopy(模型)，只能得到相同名字的model，因此不能使用。可以直接添加drop层，另外，第二份代码使用了tensorboard。训练用到的方法是通过compile实现的。训练数据是fit，当然损失函数依据不同的问题类型选择。

import tensorflow as tf  #导入tensorflow库
mnist=tf.keras.datasets.mnist  
(x_train,y_train),(x_test,y_test)=mnist.load_data()  #加载数据

y_train

array([5, 0, 4, ..., 5, 6, 8], dtype=uint8)

import matplotlib.pyplot as plt
plt.imshow(x_train[0])

<matplotlib.image.AxesImage at 0x1b4dab17f88>

x_train=tf.keras.utils.normalize(x_train,axis=1)
x_test=tf.keras.utils.normalize(x_test,axis=1)

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(128,activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(130,activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(10,activation=tf.nn.softmax))
model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])

model.fit(x_train,y_train,epochs=5) #训练模型

Epoch 1/5
1875/1875 [==============================] - 11s 5ms/step - loss: 0.2621 - accuracy: 0.9227
Epoch 2/5
1875/1875 [==============================] - 9s 5ms/step - loss: 0.1040 - accuracy: 0.9686
Epoch 3/5
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0718 - accuracy: 0.9775
Epoch 4/5
1875/1875 [==============================] - 11s 6ms/step - loss: 0.0516 - accuracy: 0.9839
Epoch 5/5
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0407 - accuracy: 0.9865





<tensorflow.python.keras.callbacks.History at 0x1b4d877b888>

val_loss,val_acc=model.evaluate(x_test,y_test) #测试，获取准确率

313/313 [==============================] - 2s 4ms/step - loss: 0.0903 - accuracy: 0.9750

predictions = model.predict([x_test[5:8]])#识别测试集中第6到8张图片
print(predictions)

WARNING:tensorflow:Layers in a Sequential model should only have a single input tensor, but we receive a <class 'tuple'> input: (<tf.Tensor 'IteratorGetNext:0' shape=(None, 28, 28) dtype=float32>,)
Consider rewriting this model with the Functional API.
[[3.14021698e-10 9.99962211e-01 2.31917916e-08 5.51868453e-08
  1.11546058e-06 1.32857991e-09 2.55177746e-09 3.54971853e-05
  6.47773959e-07 4.19447389e-07]
 [7.13356707e-09 1.39042722e-05 3.48455131e-08 3.54035663e-08
  9.99842525e-01 7.90912527e-05 3.22929310e-07 6.94255186e-06
  1.20427085e-05 4.49937252e-05]
 [5.06938420e-08 3.42604551e-07 2.70455502e-07 1.93817541e-04
  6.74861003e-05 1.81618889e-05 1.80747389e-10 1.53591927e-07
  7.90399952e-07 9.99718845e-01]]

plt.imshow(x_test[6])

---------------------------------------------------------------------------------------

import tensorflow as tf  #导入tensorflow库
import os
import copy
mnist=tf.keras.datasets.mnist  
(x_train,y_train),(x_test,y_test)=mnist.load_data()  #加载数据

y_train
import matplotlib.pyplot as plt
plt.imshow(x_train[0])

x_train=tf.keras.utils.normalize(x_train,axis=1)
x_test=tf.keras.utils.normalize(x_test,axis=1)

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(128,activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(130,activation=tf.nn.relu))
model.add(tf.keras.layers.Dropout(0.2))
model.add(tf.keras.layers.Dense(10,activation=tf.nn.softmax))
#model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])

model2 = tf.keras.models.Sequential()
model2.add(model)
model2.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])


log_dir = os.path.join("logs")
# print(log_dir)
if not os.path.exists(log_dir):
    os.mkdir(log_dir)
# 定义TensorBoard对象.histogram_freq 如果设置为0，则不会计算直方图。
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir, histogram_freq=1)  
model2.fit(x_train, y_train, epochs=5, validation_data=(x_test, y_test), callbacks=[tensorboard_callback])  


model2.summary()
layer = model2.layers
layer[0].weights
model.summary()

#!/usr/bin/env python
# coding: utf-8
# In[1]:
import tensorflow as tf  #导入tensorflow库
import os
import copy
mnist=tf.keras.datasets.mnist  
(x_train,y_train),(x_test,y_test)=mnist.load_data()  #加载数据
# In[2]:
y_train
# In[3]:
import matplotlib.pyplot as plt
plt.imshow(x_train[0])
# In[4]:
x_train=tf.keras.utils.normalize(x_train,axis=1)
x_test=tf.keras.utils.normalize(x_test,axis=1)
# In[5]:
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(128,activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(130,activation=tf.nn.relu))
model.add(tf.keras.layers.Dropout(0.2))
model.add(tf.keras.layers.Dense(10,activation=tf.nn.softmax))
#model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])
# In[6]:
model2 = tf.keras.models.Sequential()
model2.add(model)
model2.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])
# In[7]:
log_dir = os.path.join("logs")
# print(log_dir)
if not os.path.exists(log_dir):
    os.mkdir(log_dir)
# 定义TensorBoard对象.histogram_freq 如果设置为0，则不会计算直方图。
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir, histogram_freq=1)  
model2.fit(x_train, y_train, epochs=5, validation_data=(x_test, y_test), callbacks=[tensorboard_callback])  
# In[8]:
layer = model2.layers
layer[0].weights
#In[9]:
model2.summary()
# In[10]:
model.summary()
# In[11]:
val_loss,val_acc=model2.evaluate(x_test,y_test) #测试，获取准确率
# In[12]:
predictions = model.predict([x_test[5:8]])#识别测试集中第6到8张图片
print(predictions)
# In[13]:
plt.imshow(x_test[6])