Nsryjdtyk

I developed two keras models inside two unit tests. I converted the models into tensorflow graph (using https://github.com/amir-abdi/keras_to_tensorflow) to store on disk. When the tests are run separately, the models are loaded fine and work as they are expected to do. But when I run the test through unittest discover, I got the following error running the second test:

Tensor dense_2_target:0, specified in either feed_devices or fetch_devices was not found in the Graph.

I am wondering if it is a cause of any open resources or dependencies between the generated graphs? Any help is appreciated.

Here is the source code for the two models.

Model 1:

model = Sequential(name="Regressor")

model.add(Dense(10, input_dim=2, activation='relu'))

model.add(Dense(10, activation='relu'))

model.add(Dense(1, activation=None))

model.compile(loss='mean_absolute_error', optimizer='adam')

model.fit(X, y, verbose=0)

convert_to_tensorflow_graph() # as described in https://github.com/amir-abdi/keras_to_tensorflow.

Model 2:

model = Sequential(name="classifier")

model.add(Dense(10, input_dim=4, activation='relu'))

model.add(Dense(1, activation='sigmoid'))

model.compile(loss='binary_crossentropy', optimizer='adam')

model.fit(X, y, verbose=0)

convert_to_tensorflow_graph()

As you see, the structure of the models are not the same.