I have an existing TensorFlow model, and I want to add a new "parameter" (a tf.Variable
) to the model's list of parameters (such that it's trainable) and add it externally to the model's list of parameters / computational graph.
One approach that I tried, is to append the new parameters to the model's list of trainable weights, something like this (here new_parameter
is a tf.Variable
) -
model.layers[-1].trainable_weights.extend([new_parameter])
model.compile(....)
But I'm not sure if that's the best way to go about it. In PyTorch, we have nn.Parameter
instead of tf.Variable
, and we have