TensorFlow2.0基本操作（五）

最新推荐文章于 2024-04-21 06:36:51 发布
最新推荐文章于 2024-04-21 06:36:51 发布
阅读量1.9k 收藏 3
点赞数
分类专栏： TensorFlow 文章标签： TensorFlow2.0
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/wanfuchun/article/details/102531430
版权

4 自定义网络

keras.Sequential(容器)：model.trainable_variables所有的权重参数，会自动调用model.call()

keras.layers.Layer（继承制）

keras.Model:母类，inherit:1，实现初始化参数，2，

5 模型保存与加载

1）仅保存训练参数

import  tensorflow as tf
from    tensorflow.keras import datasets, layers, optimizers, Sequential, metrics


def preprocess(x, y):
    """
    x is a simple image, not a batch
    """
    x = tf.cast(x, dtype=tf.float32) / 255.
    x = tf.reshape(x, [28*28])
    y = tf.cast(y, dtype=tf.int32)
    y = tf.one_hot(y, depth=10)
    return x,y


batchsz = 128
(x, y), (x_val, y_val) = datasets.mnist.load_data()
print('datasets:', x.shape, y.shape, x.min(), x.max())



db = tf.data.Dataset.from_tensor_slices((x,y))
db = db.map(preprocess).shuffle(60000).batch(batchsz)
ds_val = tf.data.Dataset.from_tensor_slices((x_val, y_val))
ds_val = ds_val.map(preprocess).batch(batchsz) 

sample = next(iter(db))
print(sample[0].shape, sample[1].shape)


network = Sequential([layers.Dense(256, activation='relu'),
                     layers.Dense(128, activation='relu'),
                     layers.Dense(64, activation='relu'),
                     layers.Dense(32, activation='relu'),
                     layers.Dense(10)])
network.build(input_shape=(None, 28*28))
network.summary()




network.compile(optimizer=optimizers.Adam(lr=0.01),
		loss=tf.losses.CategoricalCrossentropy(from_logits=True),
		metrics=['accuracy']
	)

# 模型训练
network.fit(db, epochs=3, validation_data=ds_val, validation_freq=2)

# 模型测试
network.evaluate(ds_val)

#保存参数
network.save_weights('weights.ckpt')
print('saved weights.')
# 删除网络
del network

# 重新搭建完全一样的网络
network = Sequential([layers.Dense(256, activation='relu'),
                     layers.Dense(128, activation='relu'),
                     layers.Dense(64, activation='relu'),
                     layers.Dense(32, activation='relu'),
                     layers.Dense(10)])
# 设置前向传播参数
network.compile(optimizer=optimizers.Adam(lr=0.01),
		loss=tf.losses.CategoricalCrossentropy(from_logits=True),
		metrics=['accuracy']
	)
# 加载保存好的参数
network.load_weights('weights.ckpt')
print('loaded weights!')
# 模型评估
network.evaluate(ds_val)


# 备注：最后训练的结果不完全一样，是因为，影响最终结果的不仅仅是这些保存的参数，还有一些其他的因素。
# 如果想要 完全一样，那么就要使用另外的一种保存方式。

datasets: (60000, 28, 28) (60000,) 0 255
(128, 784) (128, 10)
Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                multiple                  200960    
_________________________________________________________________
dense_1 (Dense)              multiple                  32896     
_________________________________________________________________
dense_2 (Dense)              multiple                  8256      
_________________________________________________________________
dense_3 (Dense)              multiple                  2080      
_________________________________________________________________
dense_4 (Dense)              multiple                  330       
=================================================================
Total params: 244,522
Trainable params: 244,522
Non-trainable params: 0
_________________________________________________________________
Epoch 1/3
469/469 [==============================] - 5s 12ms/step - loss: 0.2728 - accuracy: 0.8466
Epoch 2/3
469/469 [==============================] - 5s 11ms/step - loss: 0.1394 - accuracy: 0.9589 - val_loss: 0.2163 - val_accuracy: 0.9512
Epoch 3/3
469/469 [==============================] - 4s 10ms/step - loss: 0.1178 - accuracy: 0.9649
79/79 [==============================] - 0s 5ms/step - loss: 0.1110 - accuracy: 0.9674
saved weights.
loaded weights!
79/79 [==============================] - 1s 10ms/step - loss: 0.1110 - accuracy: 0.9634





[0.1110091062153607, 0.9674]

2） 保存整个网络

import  tensorflow as tf
from    tensorflow.keras import datasets, layers, optimizers, Sequential, metrics


def preprocess(x, y):
    """
    x is a simple image, not a batch
    """
    x = tf.cast(x, dtype=tf.float32) / 255.
    x = tf.reshape(x, [28*28])
    y = tf.cast(y, dtype=tf.int32)
    y = tf.one_hot(y, depth=10)
    return x,y


batchsz = 128
(x, y), (x_val, y_val) = datasets.mnist.load_data()
print('datasets:', x.shape, y.shape, x.min(), x.max())



db = tf.data.Dataset.from_tensor_slices((x,y))
db = db.map(preprocess).shuffle(60000).batch(batchsz)
ds_val = tf.data.Dataset.from_tensor_slices((x_val, y_val))
ds_val = ds_val.map(preprocess).batch(batchsz) 

sample = next(iter(db))
print(sample[0].shape, sample[1].shape)


network = Sequential([layers.Dense(256, activation='relu'),
                     layers.Dense(128, activation='relu'),
                     layers.Dense(64, activation='relu'),
                     layers.Dense(32, activation='relu'),
                     layers.Dense(10)])
network.build(input_shape=(None, 28*28))
network.summary()




network.compile(optimizer=optimizers.Adam(lr=0.01),
		loss=tf.losses.CategoricalCrossentropy(from_logits=True),
		metrics=['accuracy']
	)

network.fit(db, epochs=3, validation_data=ds_val, validation_freq=2)

# 创建完以后做一个evaluate
network.evaluate(ds_val)

# 接下来进行保存，保存整个model
network.save('model.h5')
print('saved total model.')
del network

print('loaded model from file.')
network = tf.keras.models.load_model('model.h5', compile=False)         # 这里就没有Sequential()这个方法了，直接从文件中恢复这个模型
network.compile(optimizer=optimizers.Adam(lr=0.01),
        loss=tf.losses.CategoricalCrossentropy(from_logits=True),
        metrics=['accuracy']
    )
x_val = tf.cast(x_val, dtype=tf.float32) / 255.
x_val = tf.reshape(x_val, [-1, 28*28])
y_val = tf.cast(y_val, dtype=tf.int32)
y_val = tf.one_hot(y_val, depth=10)
ds_val = tf.data.Dataset.from_tensor_slices((x_val, y_val)).batch(128)
network.evaluate(ds_val)          # 最后再做一个evaluate()


datasets: (60000, 28, 28) (60000,) 0 255
(128, 784) (128, 10)
Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                multiple                  200960    
_________________________________________________________________
dense_1 (Dense)              multiple                  32896     
_________________________________________________________________
dense_2 (Dense)              multiple                  8256      
_________________________________________________________________
dense_3 (Dense)              multiple                  2080      
_________________________________________________________________
dense_4 (Dense)              multiple                  330       
=================================================================
Total params: 244,522
Trainable params: 244,522
Non-trainable params: 0
_________________________________________________________________
Epoch 1/3
469/469 [==============================] - 5s 12ms/step - loss: 0.2801 - accuracy: 0.8401
Epoch 2/3
469/469 [==============================] - 5s 11ms/step - loss: 0.1357 - accuracy: 0.9582 - val_loss: 0.1496 - val_accuracy: 0.9562
Epoch 3/3
469/469 [==============================] - 4s 10ms/step - loss: 0.1084 - accuracy: 0.9689
79/79 [==============================] - 0s 6ms/step - loss: 0.1324 - accuracy: 0.9636
saved total model.
loaded model from file.
79/79 [==============================] - 0s 6ms/step - loss: 0.1324 - accuracy: 0.9584





[0.13236246153615014, 0.9636]

3） 第三种保存的方式，主要用于模型在工业的部署，这种方法更加的通用，可以供其它语言进行使用

在这里插入图片描述

5 Kera实战

CIFAR10自定义网络实战-1

32*32

要使用到一个自定义的网络层：My Dense Layer

import tensorflow as tf
# datasets数据集的管理, layers, optimizers优化器,sequential容器, metrics测试的度量器
from tensorflow.keras import datasets, layers, optimizers, Sequential, metrics
from tensorflow import keras
import os

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'


def preprocess(x, y):
    # [0~255] => [-1~1]
    x = 2 * tf.cast(x, dtype=tf.float32) / 255. - 1.
    y = tf.cast(y, dtype=tf.int32)
    return x, y


batchsz = 128
# [50k, 32, 32, 3], [10k, 1]
(x, y), (x_val, y_val) = datasets.cifar10.load_data()
y = tf.squeeze(y)          # 压缩掉张量里面的一维
y_val = tf.squeeze(y_val)
y = tf.one_hot(y, depth=10)    # [50k, 10]
y_val = tf.one_hot(y_val, depth=10)   # [10k, 10]
print('datasets:', x.shape, y.shape, x_val.shape, y_val.shape, x.min(), x.max())


train_db = tf.data.Dataset.from_tensor_slices((x, y))
train_db = train_db.map(preprocess).shuffle(10000).batch(batchsz)
test_db = tf.data.Dataset.from_tensor_slices((x_val, y_val))
test_db = test_db.map(preprocess).batch(batchsz)


# 这里生成一个sample，查看下它的shape
sample = next(iter(train_db))
print('batch:', sample[0].shape, sample[1].shape)


# 接下来新建一个这样的网络对象。

# （1）实现一个自定义的层
class MyDense(layers.Layer):         # 因为我们要自定义一个网络结构，因此我们要新建一个类，这个类要继承自layers.Layer这样一个母类
    # to replace standard layers.Dense()
    # 同样的我们要继承两个函数，自定义以下两个函数:
    # 1) 初始化的一个函数
    def __init__(self, inp_dim, outp_dim):      # 输入的维度，和输出的维度
        super(MyDense, self).__init__()

        # 新建一个kernel（核函数）
        self.kernel = self.add_variable('w', [inp_dim, outp_dim])    # kernel 等于这样的一个shape
        # self.bias = self.add_variable('b', [outp_dim])


    # 2） 前向逻辑的函数
    def __call__(self, inputs, training=None):

        x = inputs @ self.kernel      # 此处没有加bias
        return x


# （2） 实现一个自定义的网络
class MyNetwork(keras.Model):
    # 同样的道理，他也需要实现两个函数
    # 1）
    def __init__(self):
        super(MyNetwork, self).__init__()
        # 这里新建5层
        self.fc1 = MyDense(32*32*3, 256)
        self.fc2 = MyDense(256, 128)
        self.fc3 = MyDense(128, 64)
        self.fc4 = MyDense(64, 32)
        self.fc5 = MyDense(32, 10)


    # 2) 实现一个前向传播的逻辑
    def call(self,inputs, training=None):
        '''
        :param inputs:[b,32,32,3]
        :param training:
        :return:
        '''
        x = tf.reshape(inputs, [-1, 32*32*3])
        # [b, 32*32*3] => [b, 256]
        x = self.fc1(x)
        x = tf.nn.relu(x)
        # [b, 256] => [b, 128]
        x = self.fc2(x)
        x = tf.nn.relu(x)
        # [b, 128] => [b, 64]
        x = self.fc3(x)
        x = tf.nn.relu(x)
        # [b, 64] => [b, 32]
        x = self.fc4(x)
        x = tf.nn.relu(x)
        # [b, 32] => [b, 10]
        x = self.fc5(x)

        return x


# 接下来我们就将网络和loss装配起来就可以了。
network = MyNetwork()
network.compile(optimizer=optimizers.Adam(lr=1e-3),                                    # 开始装配
                loss=tf.losses.CategoricalCrossentropy(from_logits=True),              # 这里为了追求一个数据稳定性，一般都使用from_logits=True
                metrics=['accuracy'])
network.fit(train_db, epochs=15, validation_data=test_db, validation_freq=1)           # 做测试，频率为1

# 一般我们会选择保存一个参数，而不是整个网络，因为这样是一个轻量级的保存方式
network.evaluate(test_db)                                                              # 首先evaluate()来评估一下模型
network.save_weights('ckpt/weights.ckpt')                                              # 后缀名自己随便起，因为这个是他自己定义的，和我们怎么定义没有关系
del network                                                                            # 把这个网络删除掉
print('saved to ckpt/weights.ckpt')
# 到此为止模型已经保存起来了，保存起来以后我们需要额外的再创建一下
# 再创建模型的时候，因为以上只是单纯的保存了权值，因此还是需要以下步骤：
# 1）我们把网络创建的这一部分加载进来
network = MyNetwork()
network.compile(optimizer=optimizers.Adam(lr=1e-3),                                    # 开始装配
                loss=tf.losses.CategoricalCrossentropy(from_logits=True),              # 这里为了追求一个数据稳定性，一般都使用from_logits=True
                metrics=['accuracy'])
# 2）我们把权值给更新
network.load_weights('ckpt/weights.ckpt')
print('loaded weights from file.')
network.evaluate(test_db)                                                              # 做一下模型测试，对比之前和之后来看一下是否有变化

# 基本上对于cifar10来讲，如果不使用卷积神经网络或者深层次的RsNet的话，就很难达到70%的准确率
# 因为只保存了权值，因此最后的结果并不是完全一样，比如随机种子没有保存等等，但是差别不大

  File "<ipython-input-5-e010b1c928d6>", line 78
    :param inputs:[b,32,32,3]
    ^
IndentationError: unexpected indent

'''
# 本人此处是用pycharm运行的，copy到了这里，notebook同样可以运行

F:\Anaconda3\envs\gpu\python.exe H:/lesson13/keras_train.py
datasets: (50000, 32, 32, 3) (50000, 10) (10000, 32, 32, 3) (10000, 10) 0 255
batch: (128, 32, 32, 3) (128, 10)
Epoch 1/15

  1/391 [..............................] - ETA: 7:38 - loss: 2.3948 - accuracy: 0.0625
  5/391 [..............................] - ETA: 1:34 - loss: 2.2641 - accuracy: 0.1199
 10/391 [..............................] - ETA: 48s - loss: 2.1756 - accuracy: 0.1605 
 15/391 [>.............................] - ETA: 33s - loss: 2.1270 - accuracy: 0.1814
 20/391 [>.............................] - ETA: 26s - loss: 2.0827 - accuracy: 0.1958
 25/391 [>.............................] - ETA: 21s - loss: 2.0638 - accuracy: 0.2074
 30/391 [=>............................] - ETA: 18s - loss: 2.0296 - accuracy: 0.2171
 35/391 [=>............................] - ETA: 16s - loss: 2.0153 - accuracy: 0.2250
 40/391 [==>...........................] - ETA: 14s - loss: 2.0061 - accuracy: 0.2320
 45/391 [==>...........................] - ETA: 12s - loss: 1.9865 - accuracy: 0.2381
 50/391 [==>...........................] - ETA: 11s - loss: 1.9769 - accuracy: 0.2437
 55/391 [===>..........................] - ETA: 10s - loss: 1.9622 - accuracy: 0.2489
 60/391 [===>..........................] - ETA: 10s - loss: 1.9483 - accuracy: 0.2537
 65/391 [===>..........................] - ETA: 9s - loss: 1.9407 - accuracy: 0.2580 
 70/391 [====>.........................] - ETA: 9s - loss: 1.9314 - accuracy: 0.2617
 75/391 [====>.........................] - ETA: 8s - loss: 1.9235 - accuracy: 0.2651
 80/391 [=====>........................] - ETA: 8s - loss: 1.9141 - accuracy: 0.2682
 85/391 [=====>........................] - ETA: 7s - loss: 1.9072 - accuracy: 0.2712
 90/391 [=====>........................] - ETA: 7s - loss: 1.9008 - accuracy: 0.2740
 95/391 [======>.......................] - ETA: 6s - loss: 1.8936 - accuracy: 0.2765
100/391 [======>.......................] - ETA: 6s - loss: 1.8870 - accuracy: 0.2789
105/391 [=======>......................] - ETA: 6s - loss: 1.8826 - accuracy: 0.2813
110/391 [=======>......................] - ETA: 6s - loss: 1.8766 - accuracy: 0.2834
115/391 [=======>......................] - ETA: 5s - loss: 1.8684 - accuracy: 0.2853
120/391 [========>.....................] - ETA: 5s - loss: 1.8636 - accuracy: 0.2873
125/391 [========>.....................] - ETA: 5s - loss: 1.8589 - accuracy: 0.2891
130/391 [========>.....................] - ETA: 5s - loss: 1.8520 - accuracy: 0.2909
135/391 [=========>....................] - ETA: 5s - loss: 1.8474 - accuracy: 0.2926
140/391 [=========>....................] - ETA: 4s - loss: 1.8434 - accuracy: 0.2943
145/391 [==========>...................] - ETA: 4s - loss: 1.8354 - accuracy: 0.2958
150/391 [==========>...................] - ETA: 4s - loss: 1.8307 - accuracy: 0.2974
155/391 [==========>...................] - ETA: 4s - loss: 1.8252 - accuracy: 0.2989
160/391 [===========>..................] - ETA: 4s - loss: 1.8223 - accuracy: 0.3004
165/391 [===========>..................] - ETA: 4s - loss: 1.8162 - accuracy: 0.3019
170/391 [============>.................] - ETA: 4s - loss: 1.8116 - accuracy: 0.3033
175/391 [============>.................] - ETA: 3s - loss: 1.8080 - accuracy: 0.3047
180/391 [============>.................] - ETA: 3s - loss: 1.8042 - accuracy: 0.3060
185/391 [=============>................] - ETA: 3s - loss: 1.8024 - accuracy: 0.3074
190/391 [=============>................] - ETA: 3s - loss: 1.7984 - accuracy: 0.3087
195/391 [=============>................] - ETA: 3s - loss: 1.7957 - accuracy: 0.3099
200/391 [==============>...............] - ETA: 3s - loss: 1.7925 - accuracy: 0.3111
205/391 [==============>...............] - ETA: 3s - loss: 1.7900 - accuracy: 0.3123
210/391 [===============>..............] - ETA: 3s - loss: 1.7875 - accuracy: 0.3135
215/391 [===============>..............] - ETA: 2s - loss: 1.7851 - accuracy: 0.3146
220/391 [===============>..............] - ETA: 2s - loss: 1.7841 - accuracy: 0.3157
225/391 [================>.............] - ETA: 2s - loss: 1.7824 - accuracy: 0.3168
230/391 [================>.............] - ETA: 2s - loss: 1.7790 - accuracy: 0.3178
235/391 [=================>............] - ETA: 2s - loss: 1.7751 - accuracy: 0.3189
240/391 [=================>............] - ETA: 2s - loss: 1.7747 - accuracy: 0.3199
245/391 [=================>............] - ETA: 2s - loss: 1.7730 - accuracy: 0.3208
250/391 [==================>...........] - ETA: 2s - loss: 1.7699 - accuracy: 0.3218
255/391 [==================>...........] - ETA: 2s - loss: 1.7687 - accuracy: 0.3227
260/391 [==================>...........] - ETA: 2s - loss: 1.7655 - accuracy: 0.3236
265/391 [===================>..........] - ETA: 1s - loss: 1.7640 - accuracy: 0.3245
270/391 [===================>..........] - ETA: 1s - loss: 1.7619 - accuracy: 0.3254
275/391 [====================>.........] - ETA: 1s - loss: 1.7597 - accuracy: 0.3263
280/391 [====================>.........] - ETA: 1s - loss: 1.7585 - accuracy: 0.3271
285/391 [====================>.........] - ETA: 1s - loss: 1.7554 - accuracy: 0.3280
290/391 [=====================>........] - ETA: 1s - loss: 1.7533 - accuracy: 0.3288
295/391 [=====================>........] - ETA: 1s - loss: 1.7516 - accuracy: 0.3296
300/391 [======================>.......] - ETA: 1s - loss: 1.7502 - accuracy: 0.3304
305/391 [======================>.......] - ETA: 1s - loss: 1.7481 - accuracy: 0.3312
310/391 [======================>.......] - ETA: 1s - loss: 1.7457 - accuracy: 0.3319
317/391 [=======================>......] - ETA: 1s - loss: 1.7437 - accuracy: 0.3330
326/391 [========================>.....] - ETA: 0s - loss: 1.7424 - accuracy: 0.3342
335/391 [========================>.....] - ETA: 0s - loss: 1.7388 - accuracy: 0.3355
344/391 [=========================>....] - ETA: 0s - loss: 1.7346 - accuracy: 0.3367
353/391 [==========================>...] - ETA: 0s - loss: 1.7306 - accuracy: 0.3379
362/391 [==========================>...] - ETA: 0s - loss: 1.7273 - accuracy: 0.3391
370/391 [===========================>..] - ETA: 0s - loss: 1.7251 - accuracy: 0.3401
378/391 [============================>.] - ETA: 0s - loss: 1.7223 - accuracy: 0.3411
387/391 [============================>.] - ETA: 0s - loss: 1.7199 - accuracy: 0.3422
391/391 [==============================] - 6s 15ms/step - loss: 1.7180 - accuracy: 0.3428 - val_loss: 1.5808 - val_accuracy: 0.4397
Epoch 2/15

  1/391 [..............................] - ETA: 2:59 - loss: 1.6163 - accuracy: 0.5000
  6/391 [..............................] - ETA: 33s - loss: 1.5496 - accuracy: 0.4834 
 11/391 [..............................] - ETA: 19s - loss: 1.5743 - accuracy: 0.4759
 16/391 [>.............................] - ETA: 14s - loss: 1.5710 - accuracy: 0.4675
 21/391 [>.............................] - ETA: 12s - loss: 1.5458 - accuracy: 0.4636
 26/391 [>.............................] - ETA: 10s - loss: 1.5524 - accuracy: 0.4620
 31/391 [=>............................] - ETA: 9s - loss: 1.5481 - accuracy: 0.4616 
 36/391 [=>............................] - ETA: 8s - loss: 1.5489 - accuracy: 0.4606
 41/391 [==>...........................] - ETA: 7s - loss: 1.5539 - accuracy: 0.4597
 46/391 [==>...........................] - ETA: 7s - loss: 1.5477 - accuracy: 0.4588
 51/391 [==>...........................] - ETA: 6s - loss: 1.5553 - accuracy: 0.4583
 56/391 [===>..........................] - ETA: 6s - loss: 1.5553 - accuracy: 0.4579
304/391 [======================>.......] - ETA: 1s - loss: 0.7426 - accuracy: 0.7298
309/391 [======================>.......] - ETA: 1s - loss: 0.7415 - accuracy: 0.7299
315/391 [=======================>......] - ETA: 0s - loss: 0.7414 - accuracy: 0.7300
324/391 [=======================>......] - ETA: 0s - loss: 0.7399 - accuracy: 0.7301
333/391 [========================>.....] - ETA: 0s - loss: 0.7395 - accuracy: 0.7303
342/391 [=========================>....] - ETA: 0s - loss: 0.7389 - accuracy: 0.7304
350/391 [=========================>....] - ETA: 0s - loss: 0.7393 - accuracy: 0.7305
358/391 [==========================>...] - ETA: 0s - loss: 0.7379 - accuracy: 0.7307
367/391 [===========================>..] - ETA: 0s - loss: 0.7364 - accuracy: 0.7308
375/391 [===========================>..] - ETA: 0s - loss: 0.7352 - accuracy: 0.7309
383/391 [============================>.] - ETA: 0s - loss: 0.7361 - accuracy: 0.7311
390/391 [============================>.] - ETA: 0s - loss: 0.7351 - accuracy: 0.7312
391/391 [==============================] - 5s 13ms/step - loss: 0.7347 - accuracy: 0.7312 - val_loss: 2.0116 - val_accuracy: 0.4835

 1/79 [..............................] - ETA: 0s - loss: 1.7383 - accuracy: 0.4766
 9/79 [==>...........................] - ETA: 0s - loss: 1.9050 - accuracy: 0.4722
17/79 [=====>........................] - ETA: 0s - loss: 1.9247 - accuracy: 0.4793
25/79 [========>.....................] - ETA: 0s - loss: 1.9901 - accuracy: 0.4741
33/79 [===========>..................] - ETA: 0s - loss: 1.9831 - accuracy: 0.4846
41/79 [==============>...............] - ETA: 0s - loss: 1.9813 - accuracy: 0.4857
49/79 [=================>............] - ETA: 0s - loss: 1.9717 - accuracy: 0.4876
57/79 [====================>.........] - ETA: 0s - loss: 2.0079 - accuracy: 0.4840
66/79 [========================>.....] - ETA: 0s - loss: 2.0058 - accuracy: 0.4838
74/79 [===========================>..] - ETA: 0s - loss: 2.0027 - accuracy: 0.4848
79/79 [==============================] - 1s 7ms/step - loss: 2.0116 - accuracy: 0.4835
saved to ckpt/weights.ckpt
loaded weights from file.

 1/79 [..............................] - ETA: 21s - loss: 1.7383 - accuracy: 0.4766
 9/79 [==>...........................] - ETA: 2s - loss: 1.9050 - accuracy: 0.4498 
16/79 [=====>........................] - ETA: 1s - loss: 1.9337 - accuracy: 0.4591
24/79 [========>.....................] - ETA: 0s - loss: 1.9727 - accuracy: 0.4660
32/79 [===========>..................] - ETA: 0s - loss: 1.9818 - accuracy: 0.4690
40/79 [==============>...............] - ETA: 0s - loss: 1.9753 - accuracy: 0.4724
46/79 [================>.............] - ETA: 0s - loss: 1.9714 - accuracy: 0.4742
52/79 [==================>...........] - ETA: 0s - loss: 1.9880 - accuracy: 0.4757
60/79 [=====================>........] - ETA: 0s - loss: 2.0014 - accuracy: 0.4770
68/79 [========================>.....] - ETA: 0s - loss: 2.0076 - accuracy: 0.4778
76/79 [===========================>..] - ETA: 0s - loss: 2.0038 - accuracy: 0.4785
79/79 [==============================] - 1s 11ms/step - loss: 2.0116 - accuracy: 0.4787

Process finished with exit code 0
'''
```





这里就是目录下保存的文件

![在这里插入图片描述](https://img-blog.csdnimg.cn/20191013121056948.png)




```python

```
wanfuchun
关注
  • 0
    点赞
  • 踩
  • 3
    收藏
    觉得还不错? 一键收藏
  • 1
    评论
专栏目录
Tensorflow2.0课程代码库.zip
03-10
在压缩包中的“tensorflow2.0_course”文件夹里，你可能会找到以下类型的文件： 1. **Notebooks**：Jupyter Notebook 文件，通常包含教程、代码示例和解释。你可以通过运行这些笔记本来学习 TensorFlow 2.0 的基本...
TensorFlow2.0-Book:《 TensorFlow从零开始学》配书代码和资源
03-22
读者可以通过解压并运行这些代码，跟随书中讲解逐步理解TensorFlow的基本操作，如变量定义、张量运算、图构建、会话执行（在TF 2.0中已经不常用）以及模型训练等。 在深入学习TensorFlow 2.0时，你需要掌握以下几个...
tensorflow: 查看 tensor详细数值方法
12-24
问题 tensor详细数值 不能直接print打印： import tensorflow as tf x = tf.constant(1) print x 输出： Tensor("Const:0", shape=(), dtype=int32) 原因： print只能打印输出shape的信息，而要打印输出tensor的值，需要借助 tf.Session，tf.InteractiveSession。 因为我们在建立graph的时候，只建立 tensor 的 结构形状信息 ，并没有 执行 数据的操作。 解决方法 法一： import tensorflow as tf x = tf.const
tensorflow v2.0入门教程——02基本操作
wardseptember的博客
10-01 1388
个人博客 tensor tensor即张量，张量的数学含义是多维数组，标量（数字）可以看作是0维的张量。一维数组称为1维向量，二维数组称为矩阵。在神经网络学习中，几乎所有的数据都可以看作张量，如神经网络的权重，偏置等。 输出hello world import tensorflow as tf hello = tf.constant('hello world') # 创建一个常量 print(h...
nn.Sequential与tensorflow的Sequential对比
最新发布
Zsusan7的博客
04-21 346
nn.Sequential() 是 PyTorch 深度学习框架中的一个类，用于按顺序容器化模块。nn.Sequential 是一个有序的容器，它包含多个网络层，数据会按照在构造函数中传入顺序依次通过每个层。
【Tensorflow 2.0 正式版教程】Batch normalization实现
Apm__5的博客
10-17 1673
上一篇教程中讲解了如何实现自定义层，现在我们来实现一个非常特殊且重要的网络层：Batch normalization（批标准化层）。 该层的特殊之处是其在训练和测试阶段有着不同的行为，具体来说，该层需要计算出输入batch的均值和方差，然后对该batch的数据进行归一化，即减去均值再除以方差。训练阶段显然可以直接计算，但在测试阶段，每个样本间是没有关系的，即不应该存在batch的概念，那么只能依靠...
tensorflow2.0 datasets.shuffle(buffer_size).batch(batch_size)
aoyou19的博客
09-03 1120
一些个人理解，记录参考 // An highlighted block import tensorflow as tf (train_images, train_labels), (_, _) = tf.keras.datasets.mnist.load_data() print('读入初始训练图像, train_images.shape:', train_images.shape) train_images = train_images[0:5000, :, :] print('取前5000个训练图像
Tensorflow模型操作常用函数总结
charles_zhang_的博客
07-12 317
面将函数模型操作常用函数总结。如表所示： 表2 模型操作相关函数 函数 tf.train.Saver(var_list=None,reshape=False, sharded=False,max_to_keep=5, keep_checkpoint_every_n_hours=10000.0, name=None,r...
日月光华tensorflow2.0实战教程全部课件
12-19
在“日月光华tensorflow2.0实战教程”中，你将深入学习 TensorFlow 2.0 的核心概念和实践技巧。以下是该教程可能涵盖的一些关键知识点： 1. **Eager Execution**： - Eager Execution 是 TensorFlow 2.0 的一个...
TensorFlow2.0 GPU版安装
11-14
【TensorFlow2.0 GPU版安装】涉及到的关键技术点包括TensorFlow的基本概念、适用系统、GPU支持以及安装流程。TensorFlow是一个数据流编程系统，专为机器学习算法设计，广泛应用于谷歌内部产品开发和科研领域。它具备...
TensorFlow2.0入门学习笔记(2)——Matplotlib数据可视化
12-20
以下是一些基本的Matplotlib操作： - 创建图表：首先，我们需要导入matplotlib.pyplot模块，通常简称为plt，然后使用`plt.figure()`创建一个新的图表。 - 绘制线条：使用`plt.plot()`函数绘制线条，可以指定x轴和y...
TensorFlow官方文档里面，MNIST数据集 input_data
01-11
TensorFlow官方文档里面， MNIST数据集 input_data
tensorflow.keras.datasets.mnist加载的mnist数据集
04-14
通过tensorflow.keras.datasets.mnist.load_data()导入数据集时，由于需要google的服务器，国内无法访问资源。
简单粗暴 TensorFlow 2.0.pdf
09-16
print(tf.add(1, 2))`会在屏幕上输出3，展示了基本的加法操作。 **自动求导机制** 自动求导是TensorFlow的核心功能之一，它允许模型的梯度计算自动化，简化了训练过程。在TensorFlow 2.0中，Eager Execution默认...
TensorFlow2.0矩阵与向量的加减乘实例
09-18
在TensorFlow 2.0中，矩阵和向量的加减乘运算是深度学习和机器学习算法中的基本操作。这些运算对于构建神经网络和其他复杂的模型至关重要。以下将详细阐述如何在TensorFlow 2.0中执行这些操作。 1. **矩阵加法** ...
tensorflow2.0资料,tensorflow2.2,Python
09-10
1. **基础知识**：介绍TensorFlow 的安装、环境设置以及基本操作。 2. **模型构建**：讲解如何使用Keras API 构建各种神经网络模型，如卷积神经网络（CNN）、循环神经网络（RNN）和注意力机制等。 3. **训练与优化...
Tensorflow2.1实践----IMDB影评文本分类（Kears）
weixin_51072772的博客
10-22 594
来源 | 极链AI云（性价比最高的共享GPU算力平台，新人注册可领取198元大礼包，价值100+小时的GPU免费使用时长，领取地址：https://cloud.videojj.com/） 本文将文本形式的影评分为“正面”或“负面”影评。这是一个二元分类（又称为两类分类）的示例，也是一种重要且广泛适用的机器学习问题。 下载IMDB数据集 import tensorflow as tf from tensorflow import keras import numpy as np imdb = kera
解决 from tensorflow.keras import datasets,layers, optimizers报错的问题， SGD没有apply_gradients
热门推荐
qq_44644355的博客
05-05 1万+
首先我的tensorflow-gpu版本是2.0.0 解决方法： from tensorflow_core.python.keras import datasets, layers 问题描述： 我是条看视频的时候，有这样一句代码： from tensorflow.keras import datasets, layers 但是我自己输入的时候会报错，然后我看了一下代码所在环境下的包site-pac...
python实现前向传播（张量）
shengxing_stu的博客
12-03 451
tensorflow基础之前向传播（张量）
TensorFlow2.0基本api教程
06-06
好的，TensorFlow2.0是一个非常强大的深度学习框架，它提供了许多基本API，下面是一个简单的教程，介绍如何使用TensorFlow2.0中的基本API： ...以上是TensorFlow2.0基本API的简单介绍。希望对您有所帮助！

“相关推荐”对你有帮助么？

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论 1
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值