3-Tensorflow-demo_0802_复杂的Saver

最新推荐文章于 2020-09-26 08:31:41 发布
最新推荐文章于 2020-09-26 08:31:41 发布
阅读量163 收藏
点赞数
分类专栏： # 深度学习3-Tensorflow
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/HJZ11/article/details/103412940
版权 

import tensorflow as tf
import os
import numpy as np

"""
实现模型的断点继续训练。
"""
tf.set_random_seed(42)
np.random.seed(42)

def reg():
    with tf.Graph().as_default():
        # 一、构建模型图
        with tf.variable_scope('network'):
            # 1、构建输入的占位符
            input_x = tf.placeholder(
                dtype=tf.float32, shape=[None, 1], name='input_x'
            )
            input_y = tf.placeholder(
                dtype=tf.float32, shape=[None, 1], name='input_y'
            )

            # 2、构建变量
            w = tf.get_variable(
                name='w', shape=[1, 1], dtype=tf.float32,
                initializer=tf.random_normal_initializer(stddev=0.1)
            )
            b = tf.get_variable(
                name='b', shape=[1], dtype=tf.float32,
                initializer=tf.zeros_initializer()
            )

            # 3、正向传播，得到预测值。
            y_pred = tf.matmul(input_x, w) + b

        with tf.variable_scope('loss'):
            # 4、计算模型损失(MSE)
            loss = tf.reduce_mean(tf.square(input_y - y_pred))

        with tf.variable_scope('optimizer'):
            # 5、定义优化器。(含义：使用梯度下降的方式求解让损失函数最小的模型参数)
            optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01)
            train_opt = optimizer.minimize(loss=loss)
        print(input_x, input_y, loss)
        print(train_opt.name)
        # 构建持久化的对象
        """
        tf.train.Saver()(self,
               var_list=None,       # 给定具体持久化哪些变量，默认是持久化所有变量（参与模型训练的）
               reshape=False,
               sharded=False,
               max_to_keep=5,        # 指定最多保存最近的几份模型
               keep_checkpoint_every_n_hours=10000.0,
               name=None,
               restore_sequentially=False,
               saver_def=None,
               builder=None,
               defer_build=False,
               allow_empty=False,
               write_version=saver_pb2.SaverDef.V2,
               pad_step_number=False,
               save_relative_paths=False,
               filename=None):
        """
        saver = tf.train.Saver(max_to_keep=2)

        # 创建持久化的文件路径
        save_file = './models/ai20/11/model.ckpt'
        dirpath = os.path.dirname(save_file)
        if not os.path.exists(dirpath):
            os.makedirs(dirpath)
            print('成功创建文件夹:{}'.format(dirpath))

        # 二、构建会话
        with tf.Session() as sess:
            # a、变量初始化
            sess.run(tf.global_variables_initializer())
            # b、加载数据（训练的数据生成）
            N = 100
            train_x = np.linspace(0, 6, N) + np.random.normal(0, 2.0, N)
            train_y = train_x * 14 + 7 + np.random.normal(0, 5.0, N)
            train_x.shape = -1, 1
            train_y.shape = -1, 1
            print(train_x.shape, train_y.shape)
            # c、模型训练
            step = 1
            for e in range(1, 200):
                # 执行模型训练操作
                feed = {input_x: train_x, input_y: train_y}
                _, train_loss = sess.run([train_opt, loss], feed_dict=feed)
                print('Epoch:{} - Train Loss:{:.5f}'.format(e, train_loss))
                step += 1

            # 执行模型持久化
            # saver.save(sess, save_path=save_file, global_step=step)
            # print('模型成功保存至:{}'.format(save_file))


def restore_method1():
    """
    恢复模型第一种方式： 需要定义和 saver保存模型时候的一样的模型图，然后调用saver.restore()进行恢复
    :return:
    """
    with tf.Graph().as_default():
        # 一、构建模型图
        with tf.variable_scope('network'):
            # 1、构建输入的占位符
            input_x = tf.placeholder(
                dtype=tf.float32, shape=[None, 1], name='input_x'
            )
            input_y = tf.placeholder(
                dtype=tf.float32, shape=[None, 1], name='input_y'
            )

            # 2、构建变量
            w = tf.get_variable(
                name='w', shape=[1, 1], dtype=tf.float32,
                initializer=tf.random_normal_initializer(stddev=0.1)
            )
            b = tf.get_variable(
                name='b', shape=[1], dtype=tf.float32,
                initializer=tf.zeros_initializer()
            )

            # 3、正向传播，得到预测值。
            y_pred = tf.matmul(input_x, w) + b

        with tf.variable_scope('loss'):
            # 4、计算模型损失(MSE)
            loss = tf.reduce_mean(tf.square(input_y - y_pred))

        with tf.variable_scope('optimizer'):
            # 5、定义优化器。(含义：使用梯度下降的方式求解让损失函数最小的模型参数)
            optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01)
            train_opt = optimizer.minimize(loss=loss)

        # 构建持久化的对象
        saver = tf.train.Saver(max_to_keep=2)

        # 创建持久化的文件路径
        save_file = './models/ai20/11/model.ckpt'
        dirpath = os.path.dirname(save_file)
        if not os.path.exists(dirpath):
            os.makedirs(dirpath)
            print('成功创建文件夹:{}'.format(dirpath))

        # 二、构建会话
        with tf.Session() as sess:
            # 恢复模型
            # a、获取持久化的信息对象。
            ckpt = tf.train.get_checkpoint_state(dirpath)
            if ckpt and ckpt.model_checkpoint_path:
                # 恢复之前持久化的变量
                saver.restore(sess, ckpt.model_checkpoint_path)
                # 如果你有多个持久化文件，从多个持久化文件中恢复最后一个模型。
                saver.recover_last_checkpoints(ckpt.all_model_checkpoint_paths)
                print('方式1：加载持久化文件，继续训练！！')
            else:
                # a、变量初始化
                sess.run(tf.global_variables_initializer())
                print('没有持久化文件，从头开始训练!')

            # b、加载数据（训练的数据生成）
            N = 100
            train_x = np.linspace(0, 6, N) + np.random.normal(0, 2.0, N)
            train_y = train_x * 14 + 7 + np.random.normal(0, 5.0, N)
            train_x.shape = -1, 1
            train_y.shape = -1, 1
            print(train_x.shape, train_y.shape)
            # c、模型训练
            step = 1
            for e in range(1, 200):
                # 执行模型训练操作
                feed = {input_x: train_x, input_y: train_y}
                _, train_loss = sess.run([train_opt, loss], feed_dict=feed)
                print('Epoch:{} - Train Loss:{:.5f}'.format(e, train_loss))
                step += 1


def restore_method2():
    """
    恢复模型第2种方式： 直接从持久化的模型图文件中加载图，然后调用saver.restore()进行恢复变量
    :return:
    """
    # 二、构建会话
    with tf.Session() as sess:
        # 恢复模型
        # 创建持久化的文件路径
        save_file = './models/ai20/11/model.ckpt'
        dirpath = os.path.dirname(save_file)
        if not os.path.exists(dirpath):
            os.makedirs(dirpath)
            print('成功创建文件夹:{}'.format(dirpath))

        # a、获取持久化的信息对象。
        ckpt = tf.train.get_checkpoint_state(dirpath)
        if ckpt is None and ckpt.model_checkpoint_path is None:
            raise Exception('没有持久化文件!!')

        # b、加载模型图
        saver = tf.train.import_meta_graph(
            meta_graph_or_file='{}.meta'.format(ckpt.model_checkpoint_path)
        )
        # c、恢复模型
        saver.restore(sess, ckpt.model_checkpoint_path)
        print('方式2：加载持久化模型，继续训练!!')

        # d、加载数据（训练的数据生成）
        N = 100
        train_x = np.linspace(0, 6, N) + np.random.normal(0, 2.0, N)
        train_y = train_x * 14 + 7 + np.random.normal(0, 5.0, N)
        train_x.shape = -1, 1
        train_y.shape = -1, 1
        print(train_x.shape, train_y.shape)

        # e、获取断点训练的操作对象
        """
        Tensor("network/input_x:0", shape=(?, 1), dtype=float32) 
        Tensor("network/input_y:0", shape=(?, 1), dtype=float32) 
        Tensor("loss/Mean:0", shape=(), dtype=float32)
        optimizer/GradientDescent
        """
        input_x = tf.get_default_graph().get_tensor_by_name('network/input_x:0')
        input_y = tf.get_default_graph().get_tensor_by_name('network/input_y:0')
        loss = tf.get_default_graph().get_tensor_by_name('loss/Mean:0')
        train_opt = tf.get_default_graph().get_operation_by_name('optimizer/GradientDescent')

        # c、模型训练
        step = 1
        for e in range(1, 200):
            # 执行模型训练操作
            feed = {input_x: train_x, input_y: train_y}
            _, train_loss = sess.run([train_opt, loss], feed_dict=feed)
            print('Epoch:{} - Train Loss:{:.5f}'.format(e, train_loss))
            step += 1


if __name__ == '__main__':
    # reg()
    restore_method1()
    restore_method1()

方式1：加载持久化文件，继续训练！！
(100, 1) (100, 1)
Epoch:1 - Train Loss:22.61078
Epoch:2 - Train Loss:22.60567
Epoch:3 - Train Loss:22.60065
Epoch:4 - Train Loss:22.59573
Epoch:5 - Train Loss:22.59088
Epoch:6 - Train Loss:22.58613
Epoch:7 - Train Loss:22.58146
Epoch:8 - Train Loss:22.57687
Epoch:9 - Train Loss:22.57236
Epoch:10 - Train Loss:22.56793
Epoch:11 - Train Loss:22.56358
Epoch:12 - Train Loss:22.55930
Epoch:13 - Train Loss:22.55511
Epoch:14 - Train Loss:22.55098
Epoch:15 - Train Loss:22.54693
Epoch:16 - Train Loss:22.54294
Epoch:17 - Train Loss:22.53903
Epoch:18 - Train Loss:22.53518
Epoch:19 - Train Loss:22.53142
Epoch:20 - Train Loss:22.52770
Epoch:21 - Train Loss:22.52406
Epoch:22 - Train Loss:22.52048
Epoch:23 - Train Loss:22.51696
Epoch:24 - Train Loss:22.51351
Epoch:25 - Train Loss:22.51011
Epoch:26 - Train Loss:22.50677
Epoch:27 - Train Loss:22.50350
Epoch:28 - Train Loss:22.50028
Epoch:29 - Train Loss:22.49712
Epoch:30 - Train Loss:22.49402
Epoch:31 - Train Loss:22.49097
Epoch:32 - Train Loss:22.48797
Epoch:33 - Train Loss:22.48502
Epoch:34 - Train Loss:22.48212
Epoch:35 - Train Loss:22.47928
Epoch:36 - Train Loss:22.47649
Epoch:37 - Train Loss:22.47375
Epoch:38 - Train Loss:22.47105
Epoch:39 - Train Loss:22.46841
Epoch:40 - Train Loss:22.46581
Epoch:41 - Train Loss:22.46325
Epoch:42 - Train Loss:22.46074
Epoch:43 - Train Loss:22.45827
Epoch:44 - Train Loss:22.45585
Epoch:45 - Train Loss:22.45347
Epoch:46 - Train Loss:22.45113
Epoch:47 - Train Loss:22.44883
Epoch:48 - Train Loss:22.44658
Epoch:49 - Train Loss:22.44436
Epoch:50 - Train Loss:22.44218
Epoch:51 - Train Loss:22.44004
Epoch:52 - Train Loss:22.43794
Epoch:53 - Train Loss:22.43587
Epoch:54 - Train Loss:22.43385
Epoch:55 - Train Loss:22.43185
Epoch:56 - Train Loss:22.42989
Epoch:57 - Train Loss:22.42797
Epoch:58 - Train Loss:22.42608
Epoch:59 - Train Loss:22.42422
Epoch:60 - Train Loss:22.42240
Epoch:61 - Train Loss:22.42060
Epoch:62 - Train Loss:22.41885
Epoch:63 - Train Loss:22.41712
Epoch:64 - Train Loss:22.41542
Epoch:65 - Train Loss:22.41375
Epoch:66 - Train Loss:22.41211
Epoch:67 - Train Loss:22.41050
Epoch:68 - Train Loss:22.40891
Epoch:69 - Train Loss:22.40736
Epoch:70 - Train Loss:22.40583
Epoch:71 - Train Loss:22.40433
Epoch:72 - Train Loss:22.40286
Epoch:73 - Train Loss:22.40141
Epoch:74 - Train Loss:22.39999
Epoch:75 - Train Loss:22.39858
Epoch:76 - Train Loss:22.39721
Epoch:77 - Train Loss:22.39586
Epoch:78 - Train Loss:22.39454
Epoch:79 - Train Loss:22.39324
Epoch:80 - Train Loss:22.39196
Epoch:81 - Train Loss:22.39070
Epoch:82 - Train Loss:22.38946
Epoch:83 - Train Loss:22.38825
Epoch:84 - Train Loss:22.38706
Epoch:85 - Train Loss:22.38589
Epoch:86 - Train Loss:22.38474
Epoch:87 - Train Loss:22.38361
Epoch:88 - Train Loss:22.38250
Epoch:89 - Train Loss:22.38141
Epoch:90 - Train Loss:22.38034
Epoch:91 - Train Loss:22.37929
Epoch:92 - Train Loss:22.37825
Epoch:93 - Train Loss:22.37724
Epoch:94 - Train Loss:22.37624
Epoch:95 - Train Loss:22.37526
Epoch:96 - Train Loss:22.37430
Epoch:97 - Train Loss:22.37335
Epoch:98 - Train Loss:22.37242
Epoch:99 - Train Loss:22.37151
Epoch:100 - Train Loss:22.37061
Epoch:101 - Train Loss:22.36973
Epoch:102 - Train Loss:22.36886
Epoch:103 - Train Loss:22.36801
Epoch:104 - Train Loss:22.36718
Epoch:105 - Train Loss:22.36636
Epoch:106 - Train Loss:22.36555
Epoch:107 - Train Loss:22.36476
Epoch:108 - Train Loss:22.36398
Epoch:109 - Train Loss:22.36321
Epoch:110 - Train Loss:22.36246
Epoch:111 - Train Loss:22.36173
Epoch:112 - Train Loss:22.36100
Epoch:113 - Train Loss:22.36029
Epoch:114 - Train Loss:22.35958
Epoch:115 - Train Loss:22.35890
Epoch:116 - Train Loss:22.35823
Epoch:117 - Train Loss:22.35756
Epoch:118 - Train Loss:22.35691
Epoch:119 - Train Loss:22.35627
Epoch:120 - Train Loss:22.35564
Epoch:121 - Train Loss:22.35502
Epoch:122 - Train Loss:22.35441
Epoch:123 - Train Loss:22.35382
Epoch:124 - Train Loss:22.35323
Epoch:125 - Train Loss:22.35266
Epoch:126 - Train Loss:22.35209
Epoch:127 - Train Loss:22.35154
Epoch:128 - Train Loss:22.35099
Epoch:129 - Train Loss:22.35046
Epoch:130 - Train Loss:22.34992
Epoch:131 - Train Loss:22.34941
Epoch:132 - Train Loss:22.34890
Epoch:133 - Train Loss:22.34840
Epoch:134 - Train Loss:22.34791
Epoch:135 - Train Loss:22.34743
Epoch:136 - Train Loss:22.34696
Epoch:137 - Train Loss:22.34649
Epoch:138 - Train Loss:22.34604
Epoch:139 - Train Loss:22.34558
Epoch:140 - Train Loss:22.34514
Epoch:141 - Train Loss:22.34471
Epoch:142 - Train Loss:22.34428
Epoch:143 - Train Loss:22.34387
Epoch:144 - Train Loss:22.34346
Epoch:145 - Train Loss:22.34305
Epoch:146 - Train Loss:22.34265
Epoch:147 - Train Loss:22.34227
Epoch:148 - Train Loss:22.34188
Epoch:149 - Train Loss:22.34150
Epoch:150 - Train Loss:22.34114
Epoch:151 - Train Loss:22.34078
Epoch:152 - Train Loss:22.34042
Epoch:153 - Train Loss:22.34007
Epoch:154 - Train Loss:22.33972
Epoch:155 - Train Loss:22.33938
Epoch:156 - Train Loss:22.33905
Epoch:157 - Train Loss:22.33873
Epoch:158 - Train Loss:22.33841
Epoch:159 - Train Loss:22.33809
Epoch:160 - Train Loss:22.33778
Epoch:161 - Train Loss:22.33748
Epoch:162 - Train Loss:22.33718
Epoch:163 - Train Loss:22.33689
Epoch:164 - Train Loss:22.33660
Epoch:165 - Train Loss:22.33631
Epoch:166 - Train Loss:22.33603
Epoch:167 - Train Loss:22.33576
Epoch:168 - Train Loss:22.33549
Epoch:169 - Train Loss:22.33523
Epoch:170 - Train Loss:22.33497
Epoch:171 - Train Loss:22.33472
Epoch:172 - Train Loss:22.33447
Epoch:173 - Train Loss:22.33422
Epoch:174 - Train Loss:22.33398
Epoch:175 - Train Loss:22.33374
Epoch:176 - Train Loss:22.33351
Epoch:177 - Train Loss:22.33328
Epoch:178 - Train Loss:22.33306
Epoch:179 - Train Loss:22.33284
Epoch:180 - Train Loss:22.33262
Epoch:181 - Train Loss:22.33240
Epoch:182 - Train Loss:22.33220
Epoch:183 - Train Loss:22.33199
Epoch:184 - Train Loss:22.33179
Epoch:185 - Train Loss:22.33159
Epoch:186 - Train Loss:22.33139
Epoch:187 - Train Loss:22.33120
Epoch:188 - Train Loss:22.33102
Epoch:189 - Train Loss:22.33083
Epoch:190 - Train Loss:22.33065
Epoch:191 - Train Loss:22.33047
Epoch:192 - Train Loss:22.33029
Epoch:193 - Train Loss:22.33012
Epoch:194 - Train Loss:22.32995
Epoch:195 - Train Loss:22.32978
Epoch:196 - Train Loss:22.32962
Epoch:197 - Train Loss:22.32946
Epoch:198 - Train Loss:22.32931
Epoch:199 - Train Loss:22.32915
方式2加载持久化模型，继续训练!!
(100, 1) (100, 1)
Epoch:1 - Train Loss:19.94881
Epoch:2 - Train Loss:19.73273
Epoch:3 - Train Loss:19.63893
Epoch:4 - Train Loss:19.59705
Epoch:5 - Train Loss:19.57723
Epoch:6 - Train Loss:19.56680
Epoch:7 - Train Loss:19.56039
Epoch:8 - Train Loss:19.55572
Epoch:9 - Train Loss:19.55181
Epoch:10 - Train Loss:19.54827
Epoch:11 - Train Loss:19.54490
Epoch:12 - Train Loss:19.54164
Epoch:13 - Train Loss:19.53845
Epoch:14 - Train Loss:19.53533
Epoch:15 - Train Loss:19.53226
Epoch:16 - Train Loss:19.52924
Epoch:17 - Train Loss:19.52627
Epoch:18 - Train Loss:19.52335
Epoch:19 - Train Loss:19.52048
Epoch:20 - Train Loss:19.51765
Epoch:21 - Train Loss:19.51487
Epoch:22 - Train Loss:19.51213
Epoch:23 - Train Loss:19.50944
Epoch:24 - Train Loss:19.50679
Epoch:25 - Train Loss:19.50418
Epoch:26 - Train Loss:19.50162
Epoch:27 - Train Loss:19.49909
Epoch:28 - Train Loss:19.49660
Epoch:29 - Train Loss:19.49416
Epoch:30 - Train Loss:19.49177
Epoch:31 - Train Loss:19.48940
Epoch:32 - Train Loss:19.48707
Epoch:33 - Train Loss:19.48478
Epoch:34 - Train Loss:19.48252
Epoch:35 - Train Loss:19.48031
Epoch:36 - Train Loss:19.47813
Epoch:37 - Train Loss:19.47598
Epoch:38 - Train Loss:19.47387
Epoch:39 - Train Loss:19.47179
Epoch:40 - Train Loss:19.46975
Epoch:41 - Train Loss:19.46773
Epoch:42 - Train Loss:19.46576
Epoch:43 - Train Loss:19.46381
Epoch:44 - Train Loss:19.46189
Epoch:45 - Train Loss:19.46001
Epoch:46 - Train Loss:19.45816
Epoch:47 - Train Loss:19.45633
Epoch:48 - Train Loss:19.45453
Epoch:49 - Train Loss:19.45277
Epoch:50 - Train Loss:19.45103
Epoch:51 - Train Loss:19.44932
Epoch:52 - Train Loss:19.44764
Epoch:53 - Train Loss:19.44598
Epoch:54 - Train Loss:19.44435
Epoch:55 - Train Loss:19.44275
Epoch:56 - Train Loss:19.44117
Epoch:57 - Train Loss:19.43962
Epoch:58 - Train Loss:19.43810
Epoch:59 - Train Loss:19.43659
Epoch:60 - Train Loss:19.43511
Epoch:61 - Train Loss:19.43366
Epoch:62 - Train Loss:19.43223
Epoch:63 - Train Loss:19.43082
Epoch:64 - Train Loss:19.42944
Epoch:65 - Train Loss:19.42808
Epoch:66 - Train Loss:19.42673
Epoch:67 - Train Loss:19.42542
Epoch:68 - Train Loss:19.42412
Epoch:69 - Train Loss:19.42284
Epoch:70 - Train Loss:19.42158
Epoch:71 - Train Loss:19.42035
Epoch:72 - Train Loss:19.41913
Epoch:73 - Train Loss:19.41793
Epoch:74 - Train Loss:19.41676
Epoch:75 - Train Loss:19.41560
Epoch:76 - Train Loss:19.41446
Epoch:77 - Train Loss:19.41334
Epoch:78 - Train Loss:19.41223
Epoch:79 - Train Loss:19.41114
Epoch:80 - Train Loss:19.41008
Epoch:81 - Train Loss:19.40903
Epoch:82 - Train Loss:19.40799
Epoch:83 - Train Loss:19.40697
Epoch:84 - Train Loss:19.40597
Epoch:85 - Train Loss:19.40499
Epoch:86 - Train Loss:19.40402
Epoch:87 - Train Loss:19.40306
Epoch:88 - Train Loss:19.40212
Epoch:89 - Train Loss:19.40120
Epoch:90 - Train Loss:19.40029
Epoch:91 - Train Loss:19.39940
Epoch:92 - Train Loss:19.39852
Epoch:93 - Train Loss:19.39766
Epoch:94 - Train Loss:19.39680
Epoch:95 - Train Loss:19.39597
Epoch:96 - Train Loss:19.39514
Epoch:97 - Train Loss:19.39433
Epoch:98 - Train Loss:19.39353
Epoch:99 - Train Loss:19.39275
Epoch:100 - Train Loss:19.39198
Epoch:101 - Train Loss:19.39122
Epoch:102 - Train Loss:19.39047
Epoch:103 - Train Loss:19.38973
Epoch:104 - Train Loss:19.38901
Epoch:105 - Train Loss:19.38829
Epoch:106 - Train Loss:19.38759
Epoch:107 - Train Loss:19.38691
Epoch:108 - Train Loss:19.38623
Epoch:109 - Train Loss:19.38556
Epoch:110 - Train Loss:19.38490
Epoch:111 - Train Loss:19.38425
Epoch:112 - Train Loss:19.38362
Epoch:113 - Train Loss:19.38300
Epoch:114 - Train Loss:19.38238
Epoch:115 - Train Loss:19.38177
Epoch:116 - Train Loss:19.38118
Epoch:117 - Train Loss:19.38059
Epoch:118 - Train Loss:19.38002
Epoch:119 - Train Loss:19.37945
Epoch:120 - Train Loss:19.37889
Epoch:121 - Train Loss:19.37834
Epoch:122 - Train Loss:19.37780
Epoch:123 - Train Loss:19.37727
Epoch:124 - Train Loss:19.37675
Epoch:125 - Train Loss:19.37623
Epoch:126 - Train Loss:19.37572
Epoch:127 - Train Loss:19.37523
Epoch:128 - Train Loss:19.37473
Epoch:129 - Train Loss:19.37425
Epoch:130 - Train Loss:19.37378
Epoch:131 - Train Loss:19.37331
Epoch:132 - Train Loss:19.37285
Epoch:133 - Train Loss:19.37240
Epoch:134 - Train Loss:19.37195
Epoch:135 - Train Loss:19.37151
Epoch:136 - Train Loss:19.37108
Epoch:137 - Train Loss:19.37066
Epoch:138 - Train Loss:19.37024
Epoch:139 - Train Loss:19.36983
Epoch:140 - Train Loss:19.36943
Epoch:141 - Train Loss:19.36903
Epoch:142 - Train Loss:19.36864
Epoch:143 - Train Loss:19.36826
Epoch:144 - Train Loss:19.36788
Epoch:145 - Train Loss:19.36751
Epoch:146 - Train Loss:19.36714
Epoch:147 - Train Loss:19.36678
Epoch:148 - Train Loss:19.36642
Epoch:149 - Train Loss:19.36608
Epoch:150 - Train Loss:19.36573
Epoch:151 - Train Loss:19.36540
Epoch:152 - Train Loss:19.36506
Epoch:153 - Train Loss:19.36473
Epoch:154 - Train Loss:19.36441
Epoch:155 - Train Loss:19.36409
Epoch:156 - Train Loss:19.36378
Epoch:157 - Train Loss:19.36348
Epoch:158 - Train Loss:19.36318
Epoch:159 - Train Loss:19.36288
Epoch:160 - Train Loss:19.36259
Epoch:161 - Train Loss:19.36230
Epoch:162 - Train Loss:19.36202
Epoch:163 - Train Loss:19.36174
Epoch:164 - Train Loss:19.36147
Epoch:165 - Train Loss:19.36120
Epoch:166 - Train Loss:19.36094
Epoch:167 - Train Loss:19.36067
Epoch:168 - Train Loss:19.36042
Epoch:169 - Train Loss:19.36016
Epoch:170 - Train Loss:19.35992
Epoch:171 - Train Loss:19.35967
Epoch:172 - Train Loss:19.35943
Epoch:173 - Train Loss:19.35920
Epoch:174 - Train Loss:19.35896
Epoch:175 - Train Loss:19.35874
Epoch:176 - Train Loss:19.35851
Epoch:177 - Train Loss:19.35829
Epoch:178 - Train Loss:19.35807
Epoch:179 - Train Loss:19.35786
Epoch:180 - Train Loss:19.35764
Epoch:181 - Train Loss:19.35744
Epoch:182 - Train Loss:19.35723
Epoch:183 - Train Loss:19.35703
Epoch:184 - Train Loss:19.35684
Epoch:185 - Train Loss:19.35664
Epoch:186 - Train Loss:19.35645
Epoch:187 - Train Loss:19.35626
Epoch:188 - Train Loss:19.35608
Epoch:189 - Train Loss:19.35589
Epoch:190 - Train Loss:19.35571
Epoch:191 - Train Loss:19.35553
Epoch:192 - Train Loss:19.35536
Epoch:193 - Train Loss:19.35519
Epoch:194 - Train Loss:19.35502
Epoch:195 - Train Loss:19.35486
Epoch:196 - Train Loss:19.35469
Epoch:197 - Train Loss:19.35453
Epoch:198 - Train Loss:19.35438
Epoch:199 - Train Loss:19.35422

Process finished with exit code 0
HJZ11
关注
专栏目录
tensorflow_speech_recognition_demo-master_TensorFlow识别_DEMO_语音识别
09-29
在本项目"tensorflow_speech_recognition_demo-master"中，我们将探讨如何使用TensorFlow进行语音识别。TensorFlow是一个强大的开源机器学习框架，广泛应用于各种人工智能任务，包括自然语言处理（NLP）和语音识别。...
TensorFlow简单的音频识别，官方文档
rainhurt的专栏
04-08 4952
本教程将向您展示如何构建可以识别 10 个不同字词的基本语音识别网络。需要注意的是，真正的语音和音频识别系统要复杂得多，但就像用于识别图像的 MNIST，这个基本语音识别网络能够帮助您基本了解所涉及的技术。学完本教程后，您将获得一个模型，该模型会尝试将时长为 1 秒的音频片段归类为无声、未知字词、“yes”、“no”、“up”、“down”、“left”、“right”、“on”、“off”、“s...
tensorflow2之数学运算
allen的博客
05-20 2025
官方API 大部分数学运算：tf.math 随机数：tf.random 矩阵运算：tf.linalg 乘法：tf.matmul（实际上是tf.linalg.matmul），等同于@ 转置：tf.transpose() 生成对角阵：tf.linalg.diag：可定制对角线上一行/下一行等信息 获取对角阵元素：tf.linalg.diag_part 逆：tf.linalg.inv() 求迹：tf.linalg.trace() 范数：tf.linalg.norm() 行列式：tf.linalg.det()
tensorflow2自定义训练 【这里包含了如何绘制训练过程中参数的变化append ,plot,如何绘制数据集scatter】
csdnhuizhu的博客
06-13 1303
1.Tensorflow中的tensor是不可变无状态对象. 机器学习模型需要可改变状态，比如模型训练和模型预测的代码是相同的，但变量值随着时间而不同（希望尽量小的loss）,为了因对随着计算而改变的状态，可以利用Python的状态可变性。 2.Tensorflow,拥有内建可变状态操作，比使用底层Python状态表示更常见。 比如表示模型的权重，使用Tensorflow变量更方便高效。 变量是一个对象，这个对象存储着数值，当在Tensorflow计算中使用时，会隐式的读取这个存储的数值。 有一些操作（tf
TensorFlow教程之完整教程 2.6 卷积神经网络
weixin_33907511的博客
08-22 565
本文档为TensorFlow参考文档，本转载已得到TensorFlow中文社区授权。 卷积神经网络 注意:本教程适用于对Tensorflow有丰富经验的用户，并假定用户有机器学习相关领域的专业知识和经验。 概述 对CIFAR-10 数据集的分类是机器学习中一个公开的基准测试问题，其任务是对一组32x32RGB的图像进行分类，这些图像涵盖了10...
教程 | TensorFlow 2.0 教程 —— 机器学习基础 —— 图像分类
蓝三金的博客
08-23 4290
译自：TensorFlow 官方教程 本指南训练一个神经网络模型，对运动鞋和衬衫等服装图像进行分类。如果你不了解所有细节也没关系，这是一个完整的 TensorFlow 程序的快速概览。 本指南使用 tf.keras，一个高级API，用于在 TensorFlow 中构建和训练模型。 # TensorFlow 和 tf.keras import tensorflow as tf from ten...
wxPython-demo-4.0.6_wxpython官方demo_wxPython-demo_wxpython_DEMO_w
10-04
这个压缩包文件"wxPython-demo-4.0.6_wxpython官方demo_wxPython-demo_wxpython_DEMO_w"包含了wxPython的官方演示程序，是学习和探索wxPython功能的理想资源。 【wxPython Demo】是wxPython库附带的一系列示例和...
K-Studio-Demo_x64_1.2.4.0.zip
04-08
标题中的“K-Studio-Demo_x64_1.2.4.0.zip”表明这是一个名为K-Studio的演示版本，适用于64位操作系统，其版本号为1.2.4.0。这个软件可能是一个专业级的3D建模或编辑工具，提供对STL（标准三角形语言）文件的支持。 ...
微信JS-SDK Demo.rar_DEMO_微信_微信JS-SDK Demo_微信js sdk_微信扫一扫
09-21
微信JS-SDK Demo 摘录微信调用扫一扫功能的实现
PN_RFID-U_1500_TBEN-S2-2RFID-4DXP_DEMO_TURCK_rfid_
09-30
标题中的"PN_RFID-U_1500_TBEN-S2-2RFID-4DXP_DEMO_TURCK_rfid_"提到了一个与RFID（Radio Frequency Identification，射频识别）相关的技术产品，具体是PN RFID-U 1500系列。这个产品可能是一个集成在TBEN-S2-2RFID-...
基于tensorflow的语音唤醒实践
非专业人士的博客
09-26 2809
语音唤醒简单的来说就是一个分类任务，将样本分为唤醒词与非唤醒词，这次实践所完成的任务是对给出的n段音频，通过训练的模型给出其分类。中间通过数据预处理，模型搭建与训练，后处理三个步骤。文中代码均是在python3.7环境下
18.2:tensorflow分类模型mobilenetv2训练(数据增强,保存模型,衰减学习率,tensorboard)，预测图像(单张,批量预测)，导出为pb完整示例
u010397980的博客
12-09 2686
二、预测 经过一段时间的训练后在model_save文件夹下保存了下面的模型： 他想使用model.ckpt-250000模型对数据进行预测。 他写了个predict.py程序为： #coding:utf-8 import os, cv2 #os.environ["CUDA_VISIBLE_DEVICES"] = "-1" # use cpu import numpy as np...
Tensorflow object detection API调整输出的置信度
Bend_Function的博客
01-19 1336
Tensorflow object detection API 输出还是原图，没画框怎么办？ 有些时候改一下置信度就好了 在“vis_util.visualize_boxes_and_labels_on_image_array”中加入 min_score_thresh=0.2 0.2表示把20%以上的检测对象输出 ...
TensorFlow知识点备忘1
u014798590的博客
08-04 291
PS：本人初学，技术菜的抠脚，本博客仅对学习过程中遇到的知识点进行总结，如有错误，还望指出，谢谢。 1、Softmax层 功能： 实现对神经网络输出结果进行归一化处理，将各结果概率限制在（0-1）,并且概率和为1，同时对结果的置信度，转换为正值（使用以e为底的指数函数实现）。归一化公式见下图1 实现 1： out1 = tf.random.normal([1,3]) out2 = tf.nn.softmax(out1, axis=1) # 通过 softmax 函数转换为概率值 print('out1
Tensorflow中的深拷贝（deepcopy）
weixin_43354545的博客
11-25 4216
原文： https://stackoverflow.com/questions/47821221/deepcopy-in-tensorflow deepcopy = tf.Variable(tt.initialized_value()) tt是待拷贝的tensor
TensorFlow基础笔记(11) max_pool2D函数 深度学习
kyle1314608的博客
02-27 1364
# def max_pool2d(inputs, # kernel_size, # stride=2, # padding='VALID', # data_format=DATA_FORMAT_NHWC, # outputs_collections...
初学者都能看懂的95%置信区间
热门推荐
bitcarmanlee的博客
09-14 37万+
1.点估计与区间估计 首先我们看看点估计的含义： 是用样本统计量来估计总体参数，因为样本统计量为数轴上某一点值，估计的结果也以一个点的数值表示，所以称为点估计。点估计虽然给出了未知参数的估计值，但是未给出估计值的可靠程度，即估计值偏离未知参数真实值的程度。 接下来看下区间估计： 给定置信水平，根据估计值确定真实值可能出现的区间范围，该区间通常以估计值为中心，该区间则为置信区间。 2.中心...
Tensorflow2.0 实现 YOLOv3（三）：yolov3.py
cocofisher的博客
05-07 3545
文章目录文章说明传入参数YOLOv3decode 文章说明 本系列文章旨在对 Github 上 malin9402 提供的代码进行说明，在这篇文章中，我们会对 YOLOv3 项目中的 yolov3.py 文件进行说明。 如果只是想运行 Github 上的代码，可以参考对 YOLOv3 代码的说明一文。 传入参数 import numpy as np import tensorflow as tf ...
【YOLO】使用VOC数据集训练自己的YOLOv3模型（Keras/TensorFlow）
チン昶的博客
04-30 3万+
使用自己制作的VOC数据集训练YOLOv3模型，后台基于Keras和TensorFlow，详细记录了从数据集准备到训练模型和验证的整个流程。同时提出一些改进意见。
pytorch-fcn-easiest-demo-master_pytorch_fcn_demo_
最新发布
09-08
pytorch-fcn-easiest-demo-master_pytorch_fcn_demo_是一个基于PyTorch库的最简单的FCN（Fully Convolutional Networks）演示项目。 FCN是一种深度学习的语义分割模型，用于将图像中的每个像素分别分类到对应的语义...
评论
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