[FIXED] How to use trained model inside custom loss function in tensorflow?

Issue

I am trying to use an already NN-based Trained model inside the custom loss function in tensorflow. But I am getting an error while using this custom loss function inside another model. Can someone help me to figure out what’s the mistake that I am doing while designing this custom loss function.

The custom loss function code is the following

def custom_loss_function(y_true, y_pred):
    
    model1= tf.keras.models.load_model('Loss_DT_Model')
    test_pred = model1.predict(y_pred)
    
    test_pred_revert = tf.math.argmax(test_pred, axis=1)
    
    acc_matrix = tf.keras.metrics.Accuracy()
    acc_matrix.update_state(y_true, test_pred_revert)
    accuracy_score = acc_matrix.result()

    return tf.squeez(test_pred)

The error that I get at runtime is as follows

RuntimeError: in user code:

    C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\keras\engine\training.py:806 train_function  *
        return step_function(self, iterator)
    C:\Users\Hufsa Khan\Desktop\Loss_function_DT_code\xxxxxxxxxxx.py:184 custom_loss_function  *
        model1= load_model('Loss_DT_Model')
    C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\keras\saving\save.py:187 load_model  **
        return saved_model_load.load(filepath, compile, options)
    C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\keras\saving\saved_model\load.py:140 load
        sess = backend.get_session()  # Variables are initialized by this call.
    C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\keras\backend.py:627 get_session
        session = _get_session(op_input_list)
    C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\keras\backend.py:587 _get_session
        raise RuntimeError('Cannot get session inside Tensorflow graph function.')

    RuntimeError: Cannot get session inside Tensorflow graph function.

Solution

You could try loading the model outside of the loss function and only passing the weights. To pass an extra argument to the custom loss function, you can wrap it inside another function like it is explained here: https://medium.com/@Bloomore/how-to-write-a-custom-loss-function-with-additional-arguments-in-keras-5f193929f7a0

Answered By - Keine_Eule