keras - Graph disconnected: cannot obtain value for tensor KerasTensor

user_1492337 0
Viewed 2584

I've been trying to create a 7 columns (features) model with Keras functional API and map it to the a 6 classes output.

import tensorflow as tf
from tensorflow.keras import Model
from tensorflow.keras.layers import Input, Dense, Concatenate

input_message = Input(shape=(128,))
x = Dense(64, activation="relu")(input_message)
x = Dense(32, activation="relu")(x)
x = Dense(4, activation="relu")(x)
model_message = Model(inputs=input_message, outputs=x)

input_description = Input(shape=(128,))
x = Dense(64, activation="relu")(input_description)
x = Dense(32, activation="relu")(x)
x = Dense(4, activation="relu")(x)
model_description = Model(inputs=input_description, outputs=x)

input_errors = Input(shape=(2,))
x = Dense(1, activation="relu")(input_errors)
model_errors = Model(inputs=input_errors, outputs=x)

input_panics = Input(shape=(2,))
x = Dense(1, activation="relu")(input_panics)
model_panics = Model(inputs=input_panics, outputs=x)

input_images = Input(shape=(2,))
x = Dense(1, activation="relu")(input_images)
model_images = Model(inputs=input_images, outputs=x)

input_committer = Input(shape=(16,))
x = Dense(4, activation="relu")(input_description)
model_committer = Model(inputs=input_committer, outputs=x)

input_reporter = Input(shape=(6,))
x = Dense(1, activation="relu")(input_reporter)
model_reporter = Model(inputs=input_reporter, outputs=x)

combined = Concatenate()([model_message.output, model_description.output, model_errors.output, 
                         model_panics.output, model_images.output, model_committer.output, model_reporter.output])

z = Dense(6, activation='softmax')(combined)
model = Model(inputs=[model_message.input, model_description.input, model_errors.input, 
                     model_panics.input, model_images.input, model_committer.input, model_reporter.input], 
              outputs=z)

Sadly it was resulted in the below error:

ValueError: Graph disconnected: cannot obtain value for tensor KerasTensor(type_spec=TensorSpec(shape=(None, 128), dtype=tf.float32, name='input_23'), name='input_23', description="created by layer 'input_23'") at layer "dense_74". The following previous layers were accessed without issue: []

My feature list is as follows:

  1. message - text
  2. description - text
  3. has_errors - int represents a bool value
  4. has_panics - int represents a bool value
  5. has_images - int represents a bool value
  6. committer groups - categorical inputs
  7. reporter group - categorical inputs - 6 possible values overall

I've been trying to follow the below link: https://www.pyimagesearch.com/2019/02/04/keras-multiple-inputs-and-mixed-data/

so 2 questions:

  1. what yields the above error?
  2. Shouldn't I apply some kind of embedding for the text?

Thanks in advance

text-classification keras tensorflow python
2 Answers

You may want to look at the variable names that you have. Usually the GraphDisconnected error is caused by having overriding names in the program.

user_120979520

Why are you creating so many models when you actually want one?

import tensorflow as tf
from tensorflow.keras import Model
from tensorflow.keras.layers import Input, Dense, Concatenate

input_message = Input(shape=(128,))
x = Dense(64, activation="relu")(input_message)
x = Dense(32, activation="relu")(x)
x1 = Dense(4, activation="relu")(x)

input_description = Input(shape=(128,))
x = Dense(64, activation="relu")(input_description)
x = Dense(32, activation="relu")(x)
x2 = Dense(4, activation="relu")(x)

input_errors = Input(shape=(2,))
x3 = Dense(1, activation="relu")(input_errors)

input_panics = Input(shape=(2,))
x4 = Dense(1, activation="relu")(input_panics)

input_images = Input(shape=(2,))
x5 = Dense(1, activation="relu")(input_images)

input_committer = Input(shape=(16,))
x6 = Dense(4, activation="relu")(input_description)

input_reporter = Input(shape=(6,))
x7 = Dense(1, activation="relu")(input_reporter)

combined = Concatenate()([x1,x2,x3,x4,x5,x6,x7])

z = Dense(6, activation='softmax')(combined)
model = Model(inputs=[input_message,input_description,input_errors,input_panics,input_images,input_committer,input_reporter ], 
              outputs=z)

I haven't run it but this seems like it could work to me.

user_99408960
Related