Wrong dimensions for @tf.function decorator

[leonknight99]

Hello,

Thank you for all the work that had been put into this python library!

I have created my own dataset using the spektral Dataset format from top quark events where the graphs are either labelled as signal (1) or background (0). Each graph is fully connected with each node (objects in the event) having 6 features and each edge having a single feature (distance between the objects). The number of nodes in each graph varies graph to graph (event to event) ranging from a minimum of 8 to a maximum of 15.

Example of a graph:

>>>dataset[0]
Graph(n_nodes=8, n_node_features=6, n_edge_features=1, n_labels=1)
>>>dataset[0].a
<class 'scipy.sparse.csr.csr_matrix'>
(8, 8)
>>>dataset[0].x
<class 'numpy.ndarray'>
(8, 6)
>>>dataset[0].e
<class 'numpy.ndarray'>
(56, 1)
>>>dataset[0].y
<class 'numpy.int64'>

Before trying to make my own message passing layers I have first tried to use the QM9 example to test that my dataset is in the right format for spektral. Although it doesn't train properly to my data, it seems to be able to run through my dataset fine until I add back in the tf.function decorator for the train step. The tf.function returns a value error saying the python inputs are incompatible with the input signature. Just wondering if this implies that there is something wrong with my data or I have just included a format that the decorator doesn't like

WARNING:tensorflow:Model was constructed with shape (None, None, 6) for input KerasTensor(type_spec=TensorSpec(shape=(None, None, 6), dtype=tf.float32, name='X_in'), name='X_in', description="created by layer 'X_in'"), but it was called on an input with incompatible shape (None, 6).
Traceback (most recent call last):
  File "/Users/leonknight/PycharmProjects/FourTopQuarksTesting/spektralGNNv5.py", line 132, in <module>
    outs = train_step(*batch)
  File "/Users/leonknight/PycharmProjects/FourTopQuarksTesting/venv/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py", line 828, in __call__
    result = self._call(*args, **kwds)
  File "/Users/leonknight/PycharmProjects/FourTopQuarksTesting/venv/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py", line 888, in _call
    return self._stateless_fn(*args, **kwds)
  File "/Users/leonknight/PycharmProjects/FourTopQuarksTesting/venv/lib/python3.8/site-packages/tensorflow/python/eager/function.py", line 2941, in __call__
    filtered_flat_args) = self._maybe_define_function(args, kwargs)
  File "/Users/leonknight/PycharmProjects/FourTopQuarksTesting/venv/lib/python3.8/site-packages/tensorflow/python/eager/function.py", line 3314, in _maybe_define_function
    self._function_spec.canonicalize_function_inputs(*args, **kwargs)
  File "/Users/leonknight/PycharmProjects/FourTopQuarksTesting/venv/lib/python3.8/site-packages/tensorflow/python/eager/function.py", line 2703, in canonicalize_function_inputs
    inputs, flat_inputs, filtered_flat_inputs = _convert_inputs_to_signature(
  File "/Users/leonknight/PycharmProjects/FourTopQuarksTesting/venv/lib/python3.8/site-packages/tensorflow/python/eager/function.py", line 2805, in _convert_inputs_to_signature
    raise ValueError("Python inputs incompatible with input_signature:\n%s" %
ValueError: Python inputs incompatible with input_signature:
  inputs: (
    (array([[ 1.00000000e+00,  0.00000000e+00,  7.53844833e+01, -2.47192383e-02, -1.00000000e+00,  0.00000000e+00],
       [ 0.00000000e+00, -1.00000000e+00,  1.17000000e+02,  1.52109375e+01,  0.00000000e+00,  1.16333008e-01],
       [ 0.00000000e+00,  1.00000000e+00,  1.03000000e+02,  1.29296875e+01,  0.00000000e+00,  8.85253906e-01],
       ...,
       [ 0.00000000e+00, -1.00000000e+00,  5.96875000e+01,  1.13281250e+01,  0.00000000e+00,  1.65283203e-01],
       [ 0.00000000e+00, -1.00000000e+00,  5.48750000e+01,  9.30468750e+00,  0.00000000e+00,  3.05175781e-01],
       [ 0.00000000e+00, -1.00000000e+00,  3.17968750e+01,  5.36328125e+00,  0.00000000e+00,  4.32617188e-01]]), <tensorflow.python.framework.sparse_tensor.SparseTensor object at 0x7fc4bb51b580>, array([[3.29374356],
       [0.01978895],
       [4.68834923],
       ...,
       [0.57218011],
       [2.84891948],
       [0.62896705]]), array([ 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  1,  1,  1,  1,  2,  2,  2,  2,  2,  2,  2,  2,  2,  3,  3,  3,  3,  3,  3,  3,  3,  3,  4,  4,  4,  4,  4,  4,  4,  4,  4,  4,
        4,  4,  5,  5,  5,  5,  5,  5,  5,  5,  5,  6,  6,  6,  6,  6,  6,  6,  6,  6,  6,  7,  7,  7,  7,  7,  7,  7,  7,  7,  7,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  9,  9,  9,  9,  9,  9,
        9,  9,  9, 10, 10, 10, 10, 10, 10, 10, 10, 10, 11, 11, 11, 11, 11, 11, 11, 11, 11, 12, 12, 12, 12, 12, 12, 12, 12, 13, 13, 13, 13, 13, 13, 13, 13, 13, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
       15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 16, 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 17, 17, 17, 17, 17, 18, 18, 18, 18, 18, 18, 18, 18, 18, 19, 19, 19, 19, 19, 19, 19, 19, 19, 19,
       19, 19, 19, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 21, 21, 21, 21, 21, 21, 21, 21, 22, 22, 22, 22, 22, 22, 22, 22, 22, 22, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 23, 24, 24, 24, 24,
       24, 24, 24, 24, 24, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 29, 29, 29,
       29, 29, 29, 29, 29, 29, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 31, 31, 31, 31, 31, 31, 31, 31, 31])),
    [0 1 1 0 1 0 0 1 1 0 0 1 0 0 0 1 0 0 0 1 1 0 0 1 0 1 0 0 0 0 1 0])
  input_signature: (
    (TensorSpec(shape=(None, 6), dtype=tf.float64, name=None), SparseTensorSpec(TensorShape([None, None]), tf.float64), TensorSpec(shape=(None, 1), dtype=tf.float64, name=None), TensorSpec(shape=(None,), dtype=tf.int64, name=None)),
    TensorSpec(shape=(None, 1), dtype=tf.int64, name=None))

Just wondering if I have made an obvious mistake somewhere with the formats of the input arrays

Thank you,
Leon

[danielegrattarola]

Yes, this is a common problem that has been addressed in a very recent commit to the develop branch.
The quick fix for now is to convert your labels from simple scalars to arrays of shape (1, ), so basically something like y = np.array([1]).

It should work after then, and when I push the new version to master and PyPi it will work even if your labels are scalars.

Cheers

[leonknight99]

Nice, that did the trick!
Thanks again