Keras:ModelCheckpoint和model.fit的verbose有什么差异？

1.软件环境⚙️
2.问题描述🔍
3.解决方法🐡
4.结论🤔

⚡插播一条老家自产的糖心苹果，多个品种，欢迎选购！有问题随时私信我⚡：🍎🍎来自雪域高原的馈赠——海拔2000米的大凉山高原生态糖心苹果，欢迎选购！！🍎🍎

大凉山高原生态糖心苹果

1.软件环境⚙️

Windows10 教育版64位
Python 3.6.3
Tensorflow-GPU 2.3.0
Keras 2.4.3
CUDA 10.1

2.问题描述🔍

我们知道在Keras框架中有多个verbose的设置，比如最常见的model.fit中的verbose：

  def fit(self,
          x=None,
          y=None,
          batch_size=None,
          epochs=1,
          verbose=1,
          callbacks=None,
          validation_split=0.,
          validation_data=None,
          shuffle=True,
          class_weight=None,
          sample_weight=None,
          initial_epoch=0,
          steps_per_epoch=None,
          validation_steps=None,
          validation_batch_size=None,
          validation_freq=1,
          max_queue_size=10,
          workers=1,
          use_multiprocessing=False):
    """Trains the model for a fixed number of epochs (iterations on a dataset).

    Arguments:c
        verbose: 0, 1, or 2. Verbosity mode.
            0 = silent, 1 = progress bar, 2 = one line per epoch.
            Note that the progress bar is not particularly useful when
            logged to a file, so verbose=2 is recommended when not running
            interactively (eg, in a production environment)."""
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

以及ModelCheckpoint中的verbose：

class EarlyStopping(Callback):
  def __init__(self,
               filepath,
               monitor='val_loss',
               verbose=0,
               save_best_only=False,
               save_weights_only=False,
               mode='auto',
               save_freq='epoch',
               options=None,
               **kwargs):
               
   """Arguments:
      verbose: verbosity mode, 0 or 1."""
1
2
3
4
5
6
7
8
9
10
11
12
13
14

可以看到verbose主要控制训练模型时，训练的loss、accuracy等评价指标的具体信息：

5/5 [==============================] - ETA: 0s - loss: 3.5273 - accuracy: 0.2000 - precision: 1.0000 - recall: 0.0625
Epoch 00001: val_accuracy improved from -inf to 0.50000, saving model to model_and_log\img_aug+mixup+Xception\Xception-01-3.5273-0.2000-3.3476-0.5000.h5
1
2

其实除了上面的verbose，callbacks中还有大量的verbose，比如LearningRateScheduler、EarlyStopping等等都有自己的verbose。
那为什么有多个verbose呢？相应的问题就来了：

每个verbose控制的都是同一个东西吗？
如果同时开启会不会相互覆盖设置？
有没有优先级？

在这里插入图片描述

3.解决方法🐡

本文以model.fit和ModelCheckpoint的verbose为例，展示了不同情况的verbose会产生什么影响，其它函数的verbose也是相同的结论。

3.1.只打开`model.fit`的`verbose`

Epoch 1/30
1/5 [=====>........................] - ETA: 0s - loss: 3.7069 - accuracy: 0.2500 - precision: 0.5000 - recall: 1.0000
2/5 [===========>..................] - ETA: 0s - loss: 3.6293 - accuracy: 0.2500 - precision: 0.7143 - recall: 0.8333
3/5 [=================>............] - ETA: 0s - loss: 3.6301 - accuracy: 0.1667 - precision: 0.6000 - recall: 0.7500
4/5 [=======================>......] - ETA: 0s - loss: 3.6179 - accuracy: 0.1250 - precision: 0.6000 - recall: 0.5000
5/5 [==============================] - ETA: 0s - loss: 3.5854 - accuracy: 0.1000 - precision: 0.6364 - recall: 0.4375
5/5 [==============================] - 5s 900ms/step - loss: 3.5854 - accuracy: 0.1000 - precision: 0.6364 - recall: 0.4375 - val_loss: 3.3600 - val_accuracy: 0.5000 - val_precision: 0.0000e+00 - val_recall: 0.0000e+00
Epoch 2/30
1/5 [=====>........................] - ETA: 0s - loss: 3.4214 - accuracy: 0.2500 - precision: 0.0000e+00 - recall: 0.0000e+00
2/5 [===========>..................] - ETA: 0s - loss: 3.4038 - accuracy: 0.1250 - precision: 0.5000 - recall: 0.1667        
3/5 [=================>............] - ETA: 0s - loss: 3.3367 - accuracy: 0.1667 - precision: 0.5000 - recall: 0.1111
4/5 [=======================>......] - ETA: 0s - loss: 3.2700 - accuracy: 0.1875 - precision: 0.5000 - recall: 0.0833
5/5 [==============================] - ETA: 0s - loss: 3.2022 - accuracy: 0.2000 - precision: 0.6667 - recall: 0.1250
5/5 [==============================] - 3s 669ms/step - loss: 3.2022 - accuracy: 0.2000 - precision: 0.6667 - recall: 0.1250 - val_loss: 2.8924 - val_accuracy: 0.5000 - val_precision: 0.0000e+00 - val_recall: 0.0000e+00
Epoch 3/30
1/5 [=====>........................] - ETA: 0s - loss: 2.8710 - accuracy: 0.2500 - precision: 0.0000e+00 - recall: 0.0000e+00
2/5 [===========>..................] - ETA: 0s - loss: 2.8412 - accuracy: 0.1250 - precision: 0.0000e+00 - recall: 0.0000e+00
3/5 [=================>............] - ETA: 0s - loss: 2.8127 - accuracy: 0.1667 - precision: 0.5000 - recall: 0.1000        
4/5 [=======================>......] - ETA: 0s - loss: 2.7665 - accuracy: 0.1875 - precision: 0.3333 - recall: 0.1818
5/5 [==============================] - ETA: 0s - loss: 2.7303 - accuracy: 0.1500 - precision: 0.2857 - recall: 0.1429
5/5 [==============================] - 3s 627ms/step - loss: 2.7303 - accuracy: 0.1500 - precision: 0.2857 - recall: 0.1429 - val_loss: 2.3946 - val_accuracy: 0.5000 - val_precision: 0.0000e+00 - val_recall: 0.0000e+00
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21

可以看到，这个时候只会以进度条的方式打印每一个Epoch的评价指标（包括默认的loss、 accuracy、precision、recall以及你自己设定的评价指标），不会打印模型的存储信息。

3.2.只打开`ModelCheckpoint`的`verbose`

可以看到，这个时候会根据ModelCheckpoint中设定的监控指标：

    checkpoint = ModelCheckpoint(model_x.__name__ + '-{epoch:02d}'
                                                    '-{loss:.4f}'
                                                    '-{accuracy:.4f}'
                                                    '-{val_loss:.4f}'
                                                    '-{val_accuracy:.4f}.h5',
                                 monitor='val_accuracy',
                                 verbose=1,
                                 mode='max',
                                 save_best_only=True)
1
2
3
4
5
6
7
8
9

只打印模型的存储信息，模型的评价指标（包括默认的loss、 accuracy、precision、recall以及你自己设定的评价指标）不会打印：

Epoch 00001: val_accuracy improved from -inf to 0.50000, saving model to model_and_log\img_aug+mixup+Xception\Xception-01-3.4481-0.2000-3.2658-0.5000.h5
Epoch 00002: val_accuracy did not improve from 0.50000
Epoch 00003: val_accuracy did not improve from 0.50000
1
2
3

3.3.同时打开`ModelCheckpoint`和`model.fit`的`verbose`

Epoch 1/30
1/5 [=====>........................] - ETA: 0s - loss: 3.5883 - accuracy: 0.5000 - precision: 0.0000e+00 - recall: 0.0000e+00
2/5 [===========>..................] - ETA: 0s - loss: 3.5933 - accuracy: 0.2500 - precision: 1.0000 - recall: 0.1667        
3/5 [=================>............] - ETA: 0s - loss: 3.5551 - accuracy: 0.2500 - precision: 0.5000 - recall: 0.1250
4/5 [=======================>......] - ETA: 0s - loss: 3.5576 - accuracy: 0.2500 - precision: 0.6667 - recall: 0.1667
5/5 [==============================] - ETA: 0s - loss: 3.5451 - accuracy: 0.2000 - precision: 0.8000 - recall: 0.2500
Epoch 00001: val_accuracy improved from -inf to 0.50000, saving model to model_and_log\img_aug+mixup+Xception\Xception-01-3.5451-0.2000-3.3282-0.5000.h5
5/5 [==============================] - 4s 889ms/step - loss: 3.5451 - accuracy: 0.2000 - precision: 0.8000 - recall: 0.2500 - val_loss: 3.3282 - val_accuracy: 0.5000 - val_precision: 0.0000e+00 - val_recall: 0.0000e+00
Epoch 2/30
1/5 [=====>........................] - ETA: 0s - loss: 3.3335 - accuracy: 0.2500 - precision: 0.0000e+00 - recall: 0.0000e+00
2/5 [===========>..................] - ETA: 0s - loss: 3.3294 - accuracy: 0.1250 - precision: 0.0000e+00 - recall: 0.0000e+00
3/5 [=================>............] - ETA: 0s - loss: 3.2696 - accuracy: 0.1667 - precision: 0.5000 - recall: 0.1111        
4/5 [=======================>......] - ETA: 0s - loss: 3.1904 - accuracy: 0.1875 - precision: 0.6667 - recall: 0.1667
5/5 [==============================] - ETA: 0s - loss: 3.1653 - accuracy: 0.1500 - precision: 0.8000 - recall: 0.2500
Epoch 00002: val_accuracy did not improve from 0.50000
5/5 [==============================] - 3s 599ms/step - loss: 3.1653 - accuracy: 0.1500 - precision: 0.8000 - recall: 0.2500 - val_loss: 2.8628 - val_accuracy: 0.5000 - val_precision: 0.5000 - val_recall: 1.0000
learning_rate= 0.00285
Epoch 3/30
1/5 [=====>........................] - ETA: 0s - loss: 2.8587 - accuracy: 0.2500 - precision: 0.5000 - recall: 0.5000
2/5 [===========>..................] - ETA: 0s - loss: 2.8461 - accuracy: 0.1250 - precision: 0.7500 - recall: 0.5000
3/5 [=================>............] - ETA: 0s - loss: 2.7913 - accuracy: 0.1667 - precision: 0.8000 - recall: 0.4000
4/5 [=======================>......] - ETA: 0s - loss: 2.7635 - accuracy: 0.2500 - precision: 0.6250 - recall: 0.4545
5/5 [==============================] - ETA: 0s - loss: 2.7230 - accuracy: 0.2500 - precision: 0.6364 - recall: 0.5000
Epoch 00003: val_accuracy did not improve from 0.50000
5/5 [==============================] - 3s 617ms/step - loss: 2.7230 - accuracy: 0.2500 - precision: 0.6364 - recall: 0.5000 - val_loss: 2.3710 - val_accuracy: 0.5000 - val_precision: 0.5000 - val_recall: 1.0000
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

可以看到，如果二者同时打开的话，模型的评价指标和存储信息均会打印，并不会产生影响，二者是完全独立的！

4.结论🤔

model.fit的verbose负责打印每一个Epoch的评价指标（包括默认的loss、 accuracy、precision、recall以及你自己设定的评价指标）
ModelCheckpoint的verbose负责打印模型的存储信息，如存储路径，存储模型的文件名
二者是完全独立的不会相互覆盖

渣男！都看到这里了，还不赶紧点赞，评论，收藏走一波？

相关阅读:
Ros1 学习12- tf坐标系广播与监听的编程实现
 内外“双驱”， NFT价值UPUP
手把手教你，Selenium 遇见伪元素该如何处理？
基于Socket编程下实现Linux-Linux、Linux-Windows udp通信
 九、MFC控件（一）
通过JavaScript、css、H5 实现简单的tab栏的交换和复用
 Typora收费了，于是乎我自己写了一个
 TCP三次握手四次分手不一样的解释
 自动化测试之路 —— Appium输入及模拟手势
 2、一起探讨MySQL 安装
原文地址：https://blog.csdn.net/qq_15969343/article/details/126330450

Keras:ModelCheckpoint和model.fit的verbose有什么差异？