【升华】两小时学会使用tensorflow框架，tensorflow使用步骤（7步）

tensorflow：一个深度学习框架，有必要了解的是tensorflow 指的tensor：张量，　张量: TensorFlow 中的基本数据对象。，flow：流程图。

tensorflow 即指张量的工作流。那么tensorflow的框架应该是一张流程图。

import tensorflow as tf
mnist = tf.keras.datasets.mnist

(x_train, y_train),(x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0

model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(input_shape=(28, 28)),
  tf.keras.layers.Dense(128, activation='relu'),
  tf.keras.layers.Dropout(0.2),
  tf.keras.layers.Dense(10, activation='softmax')
])

model.compile(optimizer='adam',
  loss='sparse_categorical_crossentropy',
  metrics=['accuracy'])

model.fit(x_train, y_train, epochs=5)
model.evaluate(x_test, y_test)

tensorflow使用步骤（7步）

import packages （导包）
import data（导数据）
build model（构建模型）
compile（编译）
callbacks（优化调整，完善模型精确度）
fit（训练验证）
evaluate（评估）

1.导入包，这个很简单，导入的时候可以导入的更细致一点，比如from ... import ...

#### PACKAGE IMPORTS ####

# Run this cell first to import all required packages. Do not make any imports elsewhere in the notebook

import tensorflow as tf
from tensorflow.keras.preprocessing.image import load_img, img_to_array
from tensorflow.keras.models import Sequential, load_model
from tensorflow.keras.layers import Dense, Flatten, Conv2D, MaxPooling2D
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping
import os
import numpy as np
import pandas as pd

# If you would like to make further imports from tensorflow, add them here

2.导入数据，很多数据集都有自己的dataloader，导入需要训练的数据。

# Run this cell to import the Eurosat data

def load_eurosat_data():
    data_dir = 'data/'
    x_train = np.load(os.path.join(data_dir, 'x_train.npy'))
    y_train = np.load(os.path.join(data_dir, 'y_train.npy'))
    x_test  = np.load(os.path.join(data_dir, 'x_test.npy'))
    y_test  = np.load(os.path.join(data_dir, 'y_test.npy'))
    return (x_train, y_train), (x_test, y_test)

(x_train, y_train), (x_test, y_test) = load_eurosat_data()
x_train = x_train / 255.0
x_test = x_test / 255.0
# x_train=x_train[:500]  
# y_train=y_train[:500]

3.搭建模型

#### GRADED CELL ####

# Complete the following function. 
# Make sure to not change the function name or arguments.

def get_new_model(input_shape):
    """
    This function should build a Sequential model according to the above specification. Ensure the 
    weights are initialised by providing the input_shape argument in the first layer, given by the
    function argument.
    Your function should also compile the model with the Adam optimiser, sparse categorical cross
    entropy loss function, and a single accuracy metric.
    """
    model=Sequential([
        Conv2D(16,(3,3),activation='relu',padding='same',name='conv_1',input_shape=input_shape),
#这个16是说的通道数，可以理解为提取特征的数量，紧跟其后的（3,3）说的是卷积核的大小。
#再就是激活函数选择ReLU。填白的方式是same。然后给这个层取个名字‘conv_1’。最后给定输入的形状。
#当然这里还能加上初始化比如kernel_initializer='he_uniform',bias_initializer='ones'。
        Conv2D(8,(3,3),activation='relu',padding='same',name='conv_2'),
        MaxPooling2D((8,8),name='pool_1'),
#这里的池化层直接上了8,8的卷积核，给了个命名。
        Flatten(name='flatten'),
#注意，只有在卷积池化操作之后，你的数据通常不是一维的，你得把他拉平了才能用上全连接。
        Dense(32,activation='relu',name='dense_1'),
#32是神经元的数量
        Dense(10,activation='softmax',name='dense_2')
#我们的问题是十分类问题，所以我们最后一层一定是10个神经元，激活函数是softmax
    ])
#在这里我们直接把模型打包好，等会调用这个函数，就能直接训练。
    model.compile(
        optimizer=tf.keras.optimizers.Adam(lr=1e-3),# 我这里只设置了一个学习率，你也可以设置更多参数https://zhuanlan.zhihu.com/p/86261902
        loss=tf.keras.losses.SparseCategoricalCrossentropy(),#设置这个损失函数，是因为我们的标签是自然数，如果标签是独热码，就不要前面的sparse就行了。
        metrics=['accuracy']#我只用了正确率这一个指标，这个东西主要是为了得到模型训练中的记录和配合callback进行参数调整的操作。
    )
    return model

callbacks是你用来调整训练模型

#### GRADED CELL ####

# Complete the following functions. 
# Make sure to not change the function names or arguments.

def get_checkpoint_every_epoch():
    """
    This function should return a ModelCheckpoint object that:
    - saves the weights only at the end of every epoch
    - saves into a directory called 'checkpoints_every_epoch' inside the current working directory
    - generates filenames in that directory like 'checkpoint_XXX' where
      XXX is the epoch number formatted to have three digits, e.g. 001, 002, 003, etc.
    """
    checkpoint_path= 'checkpoints_every_epoch/checkpoint_{epoch:03d}'
    checkpoint=ModelCheckpoint(
        filepath=checkpoint_path,
        save_weights_only=True,#只保存参数
        save_freq='epoch',#每次epoch都保存，也可以写成frequency=1,
        verbose=1#会一直打印现在的情况
    )
    return checkpoint
    
    


def get_checkpoint_best_only():
    """
    This function should return a ModelCheckpoint object that:
    - saves only the weights that generate the highest validation (testing) accuracy
    - saves into a directory called 'checkpoints_best_only' inside the current working directory
    - generates a file called 'checkpoints_best_only/checkpoint' 
    """
    checkpoint_path= 'checkpoints_best_only/checkpoint'
    checkpoint=ModelCheckpoint(
        filepath=checkpoint_path,
        save_weights_only=True,
        monitor='val_accuracy',#把验证集上的正确率作为监控对象
        save_best_only=True,#以监控对象为指标，只保存最好的模型参数
        verbose=1
    )
    return checkpoint

训练

# Train model using the callbacks you just created

callbacks = [checkpoint_every_epoch, checkpoint_best_only, early_stopping]
model.fit(
x_train, y_train, #训练数据集
epochs=20, #训练次数20次
validation_data=(x_test, y_test), #验证集用测试集。还可以用validation_split=0.15来获取验证集。但是我感觉没有必要，反正你又不用验证集训练，你直接用测试集不就行了嘛。
callbacks=callbacks#最后加上callbacks
)

【升华】两小时学会使用tensorflow框架，tensorflow使用步骤（7步）

tensorflow使用步骤（7步）

猜你喜欢