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。该模型在这里定义： https ://github.com/keras-team/keras-applications/blob/master/keras_applications/inception_v3.py看了一些帖子，列出的解决方案是：冻结图层（这不是我想要的......）for layer in model.layers:    layer.trainable = Falsehttps://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)     采用tf.variables_initializer  model = InceptionV3()  for layer in model.layers:      sess.run(tf.variables_initializer(layer.weights))参考：https ://stackoverflow.com/a/56634827/7748163我认为最好的一个，但它引发了一个错误。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。错误信息适用于 BatchNorm 层：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，如何重新训练现有的 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实践是什么？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重量？

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

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