ValueError: No gradients provided for any variable - Custom loss function

Question

I am trying to implement a custom loss function for the some time series, but receive the following error when I try to fit my model.

def mvmt_loss(y_true, y_pred):
    
    # Make sure float32 (for metric calculations)
    y_true = tf.cast(y_true, dtype=tf.float32)
    y_pred = tf.cast(y_pred, dtype=tf.float32)
    
    #extract the "next day's price" of tensor
    y_true_next = y_true[HORIZON:]
    y_pred_next = y_pred[HORIZON:]
    
    #extract the "today's price" of tensor
    y_true_tdy = y_true[:-HORIZON]
    y_pred_tdy = y_pred[:-HORIZON]
    
    #substract to get up/down movement of the two tensors
    y_true_diff = tf.subtract(y_true_next, y_true_tdy)
    y_pred_diff = tf.subtract(y_pred_next, y_pred_tdy)
    
    #create a standard tensor with zero value for comparison
    standard = tf.zeros_like(y_pred_diff)
    
    #compare with the standard; if true, UP; else DOWN
    y_true_move = tf.greater_equal(y_true_diff, standard)
    y_pred_move = tf.greater_equal(y_pred_diff, standard)
    
    #find indices where the directions are not the same
    condition = tf.not_equal(y_true_move, y_pred_move)
    indices_pred = tf.where(condition)*1
    
    pred_sum = tf.math.reduce_sum(indices_pred)    
    
    #extract the "next day's price" of tensor
    y_naive_next = y_pred[HORIZON:] + HORIZON*0.05
    
    #extract the "today's price" of tensor
    y_naive_tdy = y_pred[:-HORIZON]
    
    #substract to get up/down movement of the two tensors
    y_naive_diff = tf.subtract(y_naive_next, y_naive_tdy)
    
    #create a standard tensor with zero value for comparison
    standard = tf.zeros_like(y_naive_diff)
    
    #compare with the standard; if true, UP; else DOWN
    y_naive_move = tf.greater_equal(y_naive_diff, standard)
    
    #find indices where the directions are not the same
    condition = tf.not_equal(y_true_move, y_naive_move)
    indices_naive = tf.where(condition)*1
    
    naive_sum = tf.math.reduce_sum(indices_naive)
    
    mvmt_loss = tf.divide(naive_sum, pred_sum)
    
    loss = tf.cast(mvmt_loss, dtype=tf.float32)

    return loss  


---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Input In [59], in <module>
     13 SPY_Dense_MV.compile(loss=mvmt_loss,
     14                 optimizer=tf.keras.optimizers.Adam(),
     15                 metrics=['mae','mape'],)
     17 # Fit
---> 18 SPY_Dense_MV.fit(X_train, y_train,
     19             epochs=100,
     20             batch_size=128,
     21             verbose=0, # only print 1 line per epoch
     22             validation_data=(X_test, y_test),
     23             callbacks=[create_model_checkpoint(model_name=SPY_Dense_MV.name)])

File /Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/site-packages/keras/utils/traceback_utils.py:67, in filter_traceback.<locals>.error_handler(*args, **kwargs)
     65 except Exception as e:  # pylint: disable=broad-except
     66   filtered_tb = _process_traceback_frames(e.__traceback__)
---> 67   raise e.with_traceback(filtered_tb) from None
     68 finally:
     69   del filtered_tb

File /Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/site-packages/tensorflow/python/framework/func_graph.py:1147, in func_graph_from_py_func.<locals>.autograph_handler(*args, **kwargs)
   1145 except Exception as e:  # pylint:disable=broad-except
   1146   if hasattr(e, "ag_error_metadata"):
-> 1147     raise e.ag_error_metadata.to_exception(e)
   1148   else:
   1149     raise

ValueError: in user code:

    File "/Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/site-packages/keras/engine/training.py", line 1021, in train_function  *
        return step_function(self, iterator)
    File "/Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/site-packages/keras/engine/training.py", line 1010, in step_function  **
        outputs = model.distribute_strategy.run(run_step, args=(data,))
    File "/Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/site-packages/keras/engine/training.py", line 1000, in run_step  **
        outputs = model.train_step(data)
    File "/Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/site-packages/keras/engine/training.py", line 863, in train_step
        self.optimizer.minimize(loss, self.trainable_variables, tape=tape)
    File "/Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/site-packages/keras/optimizer_v2/optimizer_v2.py", line 532, in minimize
        return self.apply_gradients(grads_and_vars, name=name)
    File "/Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/site-packages/keras/optimizer_v2/optimizer_v2.py", line 633, in apply_gradients
        grads_and_vars = optimizer_utils.filter_empty_gradients(grads_and_vars)
    File "/Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/site-packages/keras/optimizer_v2/utils.py", line 73, in filter_empty_gradients
        raise ValueError(f"No gradients provided for any variable: {variable}. "

    ValueError: No gradients provided for any variable: (['dense_28/kernel:0', 'dense_28/bias:0', 'dense_29/kernel:0', 'dense_29/bias:0', 'dense_30/kernel:0', 'dense_30/bias:0', 'dense_31/kernel:0', 'dense_31/bias:0'],). Provided `grads_and_vars` is ((None, <tf.Variable 'dense_28/kernel:0' shape=(465, 128) dtype=float32>), (None, <tf.Variable 'dense_28/bias:0' shape=(128,) dtype=float32>), (None, <tf.Variable 'dense_29/kernel:0' shape=(128, 128) dtype=float32>), (None, <tf.Variable 'dense_29/bias:0' shape=(128,) dtype=float32>), (None, <tf.Variable 'dense_30/kernel:0' shape=(128, 128) dtype=float32>), (None, <tf.Variable 'dense_30/bias:0' shape=(128,) dtype=float32>), (None, <tf.Variable 'dense_31/kernel:0' shape=(128, 1) dtype=float32>), (None, <tf.Variable 'dense_31/bias:0' shape=(1,) dtype=float32>)).

If I run my custom loss function as a standalone, this is my output (FYI, I created some predicted values by using the 'mse' loss function):

mvmt_ratio = mvmt_loss(y_test,y_pred)
    
mvmt_ratio 

<tf.Tensor: shape=(), dtype=float32, numpy=0.98119444>

I'm not sure what the problem is because I have another custom loss function that produces similarly formatted output to the above and the model fits without issue.