Cannot set dropout rate in custom AutoModel

[trbedwards]

Bug Description

I would like to create a custom AutoModel regressor with a 5% dropout rate in both the dense block and regression head:

import autokeras as ak
input_node = StructuredDataInput()
output_node = ak.Normalization()(input_node)
output_node = ak.DenseBlock(dropout=0.05)(output_node)
output_node = ak.RegressionHead(dropout=0.05)(output_node)
reg = ak.AutoModel(inputs=input_node, outputs=output_node, overwrite=True, max_trials=2)

However, this raises an error when trying to initialise the AutoModel:

AttributeError: 'float' object has no attribute 'get_config'

Setup Details

Include the details about the versions of:

• OS type and version: Linux, Ubuntu 18.04.3
• Python: 3.9.4
• autokeras: 1.0.16
• keras-tuner: 1.0.4
• scikit-learn: 0.24.2
• numpy: 1.19.5
• pandas: 1.2.4
• tensorflow: 2.5.0

Additional context

I'm attempting to build an AutoModel where the dropout is applied both at the training stage and the prediction stage, so I can get a distribution of predictions and incorporate bayesian uncertainty. Maybe this is a bit off-topic, but this is how I'm going to approach the problem, by following the suggestion from here and replacing each instance of Dropout with this:

class MonteCarloDropout(keras.layers.Dropout):
  def call(self, inputs):
    return super().call(inputs, training=True)

Anyway, if you know of an easier/better way to do this then please let me know