[FIXED] Keras mean squared error not working over a 3-dimensional response

Issue

I have the following python code:

layer_width= 32

model= keras.Sequential()
# keras batch_size is the first dimension of data, while we are using the second dimension to create the mini-batches
model.add(keras.Input(shape= (batch_size, n_feat), batch_size= 72))
model.add(layers.Masking(mask_value= -1))
model.add(layers.LSTM(layer_width, activation= 'relu', recurrent_dropout= .5,))
model.add(layers.Dense(n_y))
model.compile(optimizer='adam', loss='mean_squared_error')

this works as long as n_y = 1 and y is limited to one dimension (except for batch_size and n_feat). I'm using the following code to feed one-dimensional target data to the network:

data= my_code.DataGenerator(train, y_train[:, :, 0:1], batch_size= batch_size)
[x.shape for x in next(iter(data))]
# [(72, 256, 22), (72, 256, 1)]

history= model.fit(data, epochs= 1)

this works without errors. However, y_train has actually 6 dimensions in its 3rd axis, and I wish to train the network to predict all of these. If I set n_y to 6 and I feed the whole of y_train to the data generator, I get this error:

----> 2 history= model.fit(data, epochs= 1)

File ~/.local/lib/python3.10/site-packages/keras/src/utils/traceback_utils.py:70, in filter_traceback.<locals>.error_handler(*args, **kwargs)
     67     filtered_tb = _process_traceback_frames(e.__traceback__)
     68     # To get the full stack trace, call:
     69     # `tf.debugging.disable_traceback_filtering()`
---> 70     raise e.with_traceback(filtered_tb) from None
     71 finally:
     72     del filtered_tb

File ~/.local/lib/python3.10/site-packages/tensorflow/python/eager/execute.py:60, in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
     53   # Convert any objects of type core_types.Tensor to Tensor.
     54   inputs = [
     55       tensor_conversion_registry.convert(t)
     56       if isinstance(t, core_types.Tensor)
     57       else t
     58       for t in inputs
     59   ]
---> 60   tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
     61                                       inputs, attrs, num_outputs)
     62 except core._NotOkStatusException as e:
     63   if name is not None:

InvalidArgumentError: Graph execution error:

Detected at node mean_squared_error/remove_squeezable_dimensions/Squeeze defined at (most recent call last):
  File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main

  File "/usr/lib/python3.10/runpy.py", line 86, in _run_code

  File "/home/carlo/.local/lib/python3.10/site-packages/ipykernel_launcher.py", line 17, in <module>

  File "/home/carlo/.local/lib/python3.10/site-packages/traitlets/config/application.py", line 1053, in launch_instance

  File "/home/carlo/.local/lib/python3.10/site-packages/ipykernel/kernelapp.py", line 737, in start

  File "/home/carlo/.local/lib/python3.10/site-packages/tornado/platform/asyncio.py", line 195, in start

  File "/usr/lib/python3.10/asyncio/base_events.py", line 603, in run_forever

  File "/usr/lib/python3.10/asyncio/base_events.py", line 1909, in _run_once

  File "/usr/lib/python3.10/asyncio/events.py", line 80, in _run

  File "/home/carlo/.local/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 524, in dispatch_queue

  File "/home/carlo/.local/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 513, in process_one

  File "/home/carlo/.local/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 418, in dispatch_shell

  File "/home/carlo/.local/lib/python3.10/site-packages/ipykernel/kernelbase.py", line 758, in execute_request

  File "/home/carlo/.local/lib/python3.10/site-packages/ipykernel/ipkernel.py", line 426, in do_execute

  File "/home/carlo/.local/lib/python3.10/site-packages/ipykernel/zmqshell.py", line 549, in run_cell

  File "/home/carlo/.local/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3046, in run_cell

  File "/home/carlo/.local/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3101, in _run_cell

  File "/home/carlo/.local/lib/python3.10/site-packages/IPython/core/async_helpers.py", line 129, in _pseudo_sync_runner

  File "/home/carlo/.local/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3306, in run_cell_async

  File "/home/carlo/.local/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3488, in run_ast_nodes

  File "/home/carlo/.local/lib/python3.10/site-packages/IPython/core/interactiveshell.py", line 3548, in run_code

  File "/tmp/ipykernel_5961/154309858.py", line 2, in <module>

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/utils/traceback_utils.py", line 65, in error_handler

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/engine/training.py", line 1783, in fit

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/engine/training.py", line 1377, in train_function

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/engine/training.py", line 1360, in step_function

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/engine/training.py", line 1349, in run_step

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/engine/training.py", line 1127, in train_step

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/engine/training.py", line 1185, in compute_loss

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/engine/compile_utils.py", line 277, in __call__

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/losses.py", line 143, in __call__

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/losses.py", line 263, in call

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/utils/losses_utils.py", line 200, in squeeze_or_expand_dimensions

  File "/home/carlo/.local/lib/python3.10/site-packages/keras/src/utils/losses_utils.py", line 139, in remove_squeezable_dimensions

Can not squeeze dim[2], expected a dimension of 1, got 6
     [[{{node mean_squared_error/remove_squeezable_dimensions/Squeeze}}]] [Op:__inference_train_function_29864]

why is that?

Solution

I am not sure why the code worked when y was one-dimensional, but I made it work with multidimensional y by setting True the pass_sequences argument of the LSTM layer. Otherwise that layer only passes the last values predicted of each series, and it would take a different data generator to make the training work (one with output shapes [(72, 256, 22), (72, 1, n_y)], note that I did not test this different data generator).

Answered By - carlo