Keras：“指定为模型输入的张量不是输入张量。”如何将输出张量正确转换为输入张量？

Question

I am building a Network like below. model_A is a classification model, and its one-hot-encoding outputs combined with the original inputs becomes the inputs of model_B.

import keras
from keras.layers import Input, Dense
from keras.models import Model

inputs = Input(shape=(12,))

# ---------------------------------------
# model_A
x = Dense(64, activation='relu')(inputs)
x = Dense(64, activation='relu')(x)
predictions_A = Dense(3, activation='softmax')(x)
model_A = Model(inputs=inputs, outputs=predictions_A)

# ---------------------------------------
# model_B
inputs_B = keras.layers.concatenate([inputs, predictions_A])
x1 = Dense(64, activation='relu')(inputs_B)
x1 = Dense(64, activation='relu')(x1)
predictions_B = Dense(1, activation='sigmoid')(x1)
model_B = Model(inputs=inputs_B, outputs=predictions_B)

The model_A part works just fine. However, when I started to add the model_B, I got the following error:

workspace/git/tensorplay/venv/lib/python3.7/site-packages/keras/engine/network.py:180: UserWarning: Model inputs must come from `keras.layers.Input` (thus holding past layer metadata), they cannot be the output of a previous non-Input layer. Here, a tensor specified as input to your model was not an Input tensor, it was generated by layer concatenate_7.
Note that input tensors are instantiated via `tensor = keras.layers.Input(shape)`.
The tensor that caused the issue was: concatenate_7/concat:0
  str(x.name))

Any idea what's the properly handle the inputs for model_B? Thanks!

Answer 1

If you want the output of the first model, to be the input of the second model, while treating them as two separate models, then you should do it in the following way:

# Joint input layer for both model A and B
inputs = Input(shape=(12,))

# ---------------------------------------
# model_A
x = Dense(64, activation='relu')(inputs)
x = Dense(64, activation='relu')(x)
predictions_A = Dense(3, activation='softmax')(x)
model_A = Model(inputs=inputs, outputs=predictions_A)

# ---------------------------------------
# model_B
# The output of model A, will be provided as input to model B from this layer
# (Make sure to adjust the dimension to correspond to the output of model A)
input_B_out_A = Input(shape=(3,))
# Concatenating the two input layers
concat = keras.layers.concatenate([inputs, input_B_out_A])
x1 = Dense(64, activation='relu')(concat)
x1 = Dense(64, activation='relu')(x1)
predictions_B = Dense(1, activation='sigmoid')(x1)
# When creating the model, I am specifying that there are two inputs layers
# one that will get the joint input, and the other that will get the output
# from model A.
model_B = Model(inputs=[inputs, input_B_out_A], outputs=predictions_B)

---

However, if you want the models to be interconnected, you just need to change the following line:

model_B = Model(inputs=inputs, outputs=predictions_B)

Actually, the inputs of model_B is the same input layer as model_A, where you are just concatenating the input layer and the output of model_A. The whole code would be:

# Joint input layer for both model A and B
inputs = Input(shape=(12,))

# ---------------------------------------
# model_A
x = Dense(64, activation='relu')(inputs)
x = Dense(64, activation='relu')(x)
predictions_A = Dense(3, activation='softmax')(x)
model_A = Model(inputs=inputs, outputs=predictions_A)

# ---------------------------------------
# model_B
inputs_B = keras.layers.concatenate([inputs, predictions_A])
x1 = Dense(64, activation='relu')(inputs_B)
x1 = Dense(64, activation='relu')(x1)
predictions_B = Dense(1, activation='sigmoid')(x1)
model_B = Model(inputs=inputs, outputs=predictions_B)