python - 如何初始化现有 Keras 模型的变量？

Question

我想从头开始重新训练 Keras 模型 Inception_v3。

模型在这里定义: https://github.com/keras-team/keras-applications/blob/master/keras_applications/inception_v3.py

我看了一些帖子，

列出的解决方案是:

1. 卡住层(这不是我想要的...)

   for layer in model.layers:
       layer.trainable = False
   

2. https://stackoverflow.com/a/51727616/7748163

通过检查初始化器来重置所有层:

def reset_weights(model):
    session = K.get_session()
    for layer in model.layers: 
        if hasattr(layer, 'kernel_initializer'):
            layer.kernel_initializer.run(session=session)
        if hasattr(layer, 'bias_initializer'):
            layer.bias_initializer.run(session=session)     

1. 使用tf.variables_initializer

     model = InceptionV3()
     for layer in model.layers:
         sess.run(tf.variables_initializer(layer.weights))
   

   引用:https://stackoverflow.com/a/56634827/7748163

2. 我认为最好的一个，但会引发错误。

   sess = tf.Session()
   for layer in model.layers: 
        for v in layer.__dict__:
            v_arg = getattr(layer,v)
            if hasattr(v_arg,'initializer'):
                initializer_method = getattr(v_arg, 'initializer')
                initializer_method.run(session=sess)
                print('reinitializing layer {}.{}'.format(layer.name, v))
   

但是，它们都不适用于 Inception_v3。

BatchNorm层的错误信息:

tensorflow.python.framework.errors_impl.FailedPreconditionError: Error while reading resource variable batch_normalization_9/moving_mean from Container: localhost. This could mean that the variable was uninitialized. Not found: Resource localhost/batch_normalization_9/moving_mean/N10tensorflow3VarE does not exist.
     [[{{node batch_normalization_9_1/AssignMovingAvg/ReadVariableOp}}]]
     [[metrics_1/categorical_accuracy/Identity/_469]]

那么，如何重新训练已有的Keras模型，并初始化变量呢？从 Keras 应用程序重新训练模型的最佳做法是什么？

进一步讨论:

https://github.com/keras-team/keras/issues/341

Answer 1

为什么不简单地不要求权重？

model = Inception_V3(..., weights=None,...)

https://github.com/keras-team/keras-applications/blob/master/keras_applications/inception_v3.py/#L100