blocks|key|2689389|text|何时使用？|type|unstyled|depth|inlineStyleRanges|offset|length|style|BOLD|entityRanges|data|2689390|如果你使用的是compile，那么它肯定是在load_model()之后。毕竟，你需要一个模型来编译。(PS：load_model使用随模型一起保存的优化器自动编译模型)|CODE|2689391|+compile+做了什么？|2689392|Compile定义了损失函数、优化器和指标。就这样。|2689393|它与权重无关，您可以根据需要多次编译模型，而不会对预先训练的权重造成任何问题。|2689394|您需要一个编译的模型来训练(因为训练使用损失函数和优化器)。但是没有必要为了预测而编译一个模型。|2689395|是否需要多次使用编译？|2689396|仅在以下情况下：|2689397|2689398|您想要更改其中之一：|unordered-list-item|2689399|Loss+function+|2689400|优化器/学习率|2689401|指标|2689402|某些图层优化器的trainable属性|2689403|2689404|2689405|​|2689406|2689407|您加载(或创建)了尚未编译的模型。或者您的加载/保存方法没有考虑之前的compilation.|2689408|2689409|重新编译的后果：|2689410|如果您再次编译模型，您将丢失状态的优化器。|2689411|这意味着你的训练在一开始会受到一些影响，直到它调整了学习率、动量等。但绝对不会对权重造成损害(当然，除非你的初始学习率太高，以至于第一步训练就疯狂地改变了微调的权重)。|2689412|entityMap^0|0|5|0|7|7|M|C|1J|A|0|1|7|9|5|1|7|0|A|4|F|3|J|2|0|0|0|0|B|0|0|0|1|1|1|1|8|9|1|0|0|0|0|0|0|5|3|0|H|4|0|0|2C|0^^$0|@$1|2|3|4|5|6|7|1M|8|@$9|1N|A|1O|B|C]]|D|@]|E|$]]|$1|F|3|G|5|6|7|1P|8|@$9|1Q|A|1R|B|H]|$9|1S|A|1T|B|H]|$9|1U|A|1V|B|H]]|D|@]|E|$]]|$1|I|3|J|5|6|7|1W|8|@$9|1X|A|1Y|B|C]|$9|1Z|A|20|B|C]|$9|21|A|22|B|H]]|D|@]|E|$]]|$1|K|3|L|5|6|7|23|8|@$9|24|A|25|B|C]|$9|26|A|27|B|C]|$9|28|A|29|B|C]]|D|@]|E|$]]|$1|M|3|N|5|6|7|2A|8|@]|D|@]|E|$]]|$1|O|3|P|5|6|7|2B|8|@]|D|@]|E|$]]|$1|Q|3|R|5|6|7|2C|8|@$9|2D|A|2E|B|C]]|D|@]|E|$]]|$1|S|3|T|5|6|7|2F|8|@]|D|@]|E|$]]|$1|U|3|-4|5|6|7|2G|8|@]|D|@]|E|$]]|$1|V|3|W|5|X|7|2H|8|@]|D|@]|E|$]]|$1|Y|3|Z|5|X|7|2I|8|@]|D|@]|E|$]]|$1|10|3|11|5|X|7|2J|8|@]|D|@]|E|$]]|$1|12|3|13|5|X|7|2K|8|@]|D|@]|E|$]]|$1|14|3|15|5|X|7|2L|8|@$9|2M|A|2N|B|H]]|D|@]|E|$]]|$1|16|3|-4|5|6|7|2O|8|@]|D|@]|E|$]]|$1|17|3|-4|5|6|7|2P|8|@]|D|@]|E|$]]|$1|18|3|19|5|6|7|2Q|8|@]|D|@]|E|$]]|$1|1A|3|-4|5|6|7|2R|8|@]|D|@]|E|$]]|$1|1B|3|1C|5|X|7|2S|8|@]|D|@]|E|$]]|$1|1D|3|-4|5|6|7|2T|8|@]|D|@]|E|$]]|$1|1E|3|1F|5|6|7|2U|8|@$9|2V|A|2W|B|C]]|D|@]|E|$]]|$1|1G|3|1H|5|6|7|2X|8|@$9|2Y|A|2Z|B|C]]|D|@]|E|$]]|$1|1I|3|1J|5|6|7|30|8|@$9|31|A|32|B|C]]|D|@]|E|$]]|$1|1K|3|-4|5|6|7|33|8|@]|D|@]|E|$]]]|1L|$]]

When to use?

If you're using <code>compile</code>, surely it must be after <code>load_model()</code>. After all, you need a model to compile. (PS: <code>load_model</code> automatically compiles the model with the optimizer that was saved along with the model)

What does <code>compile</code> do?

Compile defines the loss function, the optimizer and the metrics. That's all. 

It has nothing to do with the weights and you can compile a model as many times as you want without causing any problem to pretrained weights. 

You need a compiled model to train (because training uses the loss function and the optimizer). But it's not necessary to compile a model for predicting. 

Do you need to use compile more than once?

Only if:

<ul>
<li>You want to change one of these: 

<ul>
<li>Loss function </li>
<li>Optimizer / Learning rate </li>
<li>Metrics </li>
<li>The <code>trainable</code> property of some layer </li>
</ul></li>
<li>You loaded (or created) a model that is not compiled yet. Or your load/save method didn't consider the previous compilation.</li>
</ul>

Consequences of compiling again:

If you compile a model again, you will lose the optimizer states. 

This means that your training will suffer a little at the beginning until it adjusts the learning rate, the momentums, etc. But there is absolutely no damage to the weights (unless, of course, your initial learning rate is so big that the first training step wildly changes the fine tuned weights).

blocks|key|23791|text|不要忘记，在更改层的trainable标志后，您还需要编译模型，例如，当您想要对模型进行微调时：|type|unstyled|depth|inlineStyleRanges|offset|length|style|CODE|entityRanges|data|23792|在没有顶层的情况下加载VGG模型所有图层(例如，将一些图层放到顶层)通过再次设置data|23793|un-freeze+trainable+=+True|unordered-list-item|23794|
|23795|compile+|23796|(不要忘记此步骤！)|ordered-list-item|23797|在一些data|23798|上编译和训练模型VGG的一些图层|23799|23800|entityMap^0|A|9|0|0|A|G|0|0|0|0|0|0|0^^$0|@$1|2|3|4|5|6|7|Y|8|@$9|Z|A|10|B|C]]|D|@]|E|$]]|$1|F|3|G|5|6|7|11|8|@]|D|@]|E|$]]|$1|H|3|I|5|J|7|12|8|@$9|13|A|14|B|C]]|D|@]|E|$]]|$1|K|3|L|5|6|7|15|8|@]|D|@]|E|$]]|$1|M|3|N|5|J|7|16|8|@]|D|@]|E|$]]|$1|O|3|P|5|Q|7|17|8|@]|D|@]|E|$]]|$1|R|3|S|5|Q|7|18|8|@]|D|@]|E|$]]|$1|T|3|U|5|6|7|19|8|@]|D|@]|E|$]]|$1|V|3|-4|5|6|7|1A|8|@]|D|@]|E|$]]|$1|W|3|-4|5|6|7|1B|8|@]|D|@]|E|$]]]|X|$]]

Don't forget that you also need to compile the model after changing the <code>trainable</code> flag of a layer, e.g. when you want to fine-tune a model like this:

<ol>
<li>load VGG model without top classifier</li>
<li>freeze all the layers (i.e. <code>trainable = False</code>)</li>
<li>add some layers to the top</li>
<li>compile and train the model on some data</li>
<li>un-freeze some of the layers of VGG by setting <code>trainable = True</code></li>
<li>compile the model again (DON'T FORGET THIS STEP!)</li>
<li>train the model on some data</li>
</ol>

When I start training a model, there is no model saved previously. I can use <code>model.compile()</code> safely. I have now saved the model in a <code>h5</code> file for further training using <code>checkpoint</code>.

Say, I want to train the model further. I am confused at this point: can I use <code>model.compile()</code> here? And should it be placed before or after the <code>model = load_model()</code> statement? If <code>model.compile()</code> reinitializes all the weights and biases, I should place it before <code>model = load_model()</code> statement.

After discovering some discussions, it seems to me that <code>model.compile()</code> is only needed when I have no model saved previously. Once I have saved the model, there is no need to use <code>model.compile()</code>. Is it true or false? And when I want to predict using the trained model, should I use <code>model.compile()</code> before predicting?

Does model.compile() initialize all the weights and biases in Keras (tensorflow backend)?

当我开始训练模型时，之前没有保存任何模型。我可以安全地使用model.compile()。现在，我已经将模型保存在h5文件中，以便使用checkpoint进行进一步的培训。比方说，我想进一步训练模型。在这一点上我很困惑:我可以在这里使用model.compile()吗？它应该放在model = load_model()...

问model.compile()是否初始化Keras (tensorflow后端)中的所有权重和偏差？
EN

社区

活动

资源

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问model.compile()是否初始化Keras (tensorflow后端)中的所有权重和偏差？EN

社区

活动

资源

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问model.compile()是否初始化Keras (tensorflow后端)中的所有权重和偏差？
EN