创建一个Keras回调，每n个周期激活一次 [英] Creating a Keras Callback that activates every n epochs

问题描述

我想在Keras中创建自己的Callback，但我不太了解该怎么做.

I want to create my own Callback in Keras but don't really understand how to do it.

我想做的是创建一个回调，每个 n (例如 n = 10 )时代都调用一个函数.根据Keras文档( https://keras.io/callbacks/)，基类keras.callbacks.Callback 具有属性 params .

What I want to do is to create a Callback that every n (e.g. n=10) epochs calls a function. According to the Keras documentation (https://keras.io/callbacks/) the base class keras.callbacks.Callback has a property params.

此属性是否包括当前纪元?如果是这样，如何使用/调用它?

Does this property include the current epoch? And if so, how can it be used/called?

或者是否还有其他好的方法来每隔 n 个纪元调用一个函数?

Or is there any other good way to call a function every n epochs?

我想到了一个回调，当使用参数 period = 10 时，它的作用类似于 keras.callbacks.ModelCheckpoint .

I had a Callback in mind that works similar as keras.callbacks.ModelCheckpoint when using the argument period=10.

我们将非常感谢您的帮助.谢谢:)

Help would be really appreciated. Thank you :)

经过一番阅读后，我想到了这一点，它似乎可以正常工作(仍然需要对其进行正确的测试)

After quite some reading I came up with this and it seems to work (still needs to test it properly)

class Test1(callbacks.Callback):
    def on_epoch_end(self, epochs, logs={})
        if epochs == 10:
            print('abc') #a random function

callb = Test1()    
model = networks.compute_network(layers=layers, batch_size=batch_size, epochs=epochs, call_list=[callb])
# compute_network in my case loads all the data, trains the network, and then returns it

一个网站特别帮助我更好地了解了回调: https://keunwoochoi.wordpress.com/2016/07/16/keras-callbacks/

One site helped me in particular to understand Callbacks better: https://keunwoochoi.wordpress.com/2016/07/16/keras-callbacks/

推荐答案

正如@HMK所说， on_epoch_begin 和 on_epoch_end 方法提供了当前纪元.例如，要每隔10个时间应用一个函数，您可以像这样编辑自定义回调:

As @HMK said, on_epoch_begin and on_epoch_end methods provide the current epoch. For example, to apply a function every 10th epochs you can edit your custom callback like that:

class myCallback(keras.callbacks.Callback):
    def on_epoch_end(self, epoch, logs=None)
        if epoch % 10 == 0:
            your_func()

这篇关于创建一个Keras回调，每n个周期激活一次的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！