tf.keras.to_categorical TypeError 在 model.fit 期间

Question

tf.keras.to_categorical seems incompatible with model fitting. If I use it in a model, I can run the model before training to predict some values (and verify that the shape of the output is correct), but fitting tells me that there's an error in my user code. I've tried on both stable and nightly.

I've simplified my model down to:

# Minimal example
class TestModel1(tf.keras.Model):
    def call(self, x):
        return tf.keras.utils.to_categorical(x)

# Fixed width and a trainable layer
class TestModel2(tf.keras.Model):
    def __init__(self):
        super(TestModel2, self).__init__()
        self.d1 = tf.keras.layers.Dense(2, activation='relu')
    def call(self, x):
        x = tf.keras.utils.to_categorical(x, num_classes=5)
        return self.d1(x)

Both models run fine, but run into the same error when fitting.

>>> x = [1, 2, 3, 4]
>>> y = [0, 1, 0, 1]
>>> TestModel1()(x)
array([[0., 1., 0., 0., 0.],
       [0., 0., 1., 0., 0.],
       [0., 0., 0., 1., 0.],
       [0., 0., 0., 0., 1.]], dtype=float32)
>>> TestModel2()(x)
<tf.Tensor: shape=(4, 2), dtype=float32, numpy=
array([[0.1654321 , 0.        ],
       [0.        , 0.07433152],
       [0.87672186, 0.        ],
       [0.25229335, 0.        ]], dtype=float32)>

>>> loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
>>> model.compile(optimizer='adam', loss=loss_fn, metrics=['accuracy'])
>>> model.fit(x, y, epochs=1)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py", line 108, in _method_wrapper
    return method(self, *args, **kwargs)
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py", line 1098, in fit
    tmp_logs = train_function(iterator)
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py", line 780, in __call__
    result = self._call(*args, **kwds)
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py", line 823, in _call
    self._initialize(args, kwds, add_initializers_to=initializers)
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py", line 696, in _initialize
    self._stateful_fn._get_concrete_function_internal_garbage_collected(  # pylint: disable=protected-access
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/eager/function.py", line 2855, in _get_concrete_function_internal_garbage_collected
    graph_function, _, _ = self._maybe_define_function(args, kwargs)
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/eager/function.py", line 3213, in _maybe_define_function
    graph_function = self._create_graph_function(args, kwargs)
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/eager/function.py", line 3065, in _create_graph_function
    func_graph_module.func_graph_from_py_func(
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/framework/func_graph.py", line 986, in func_graph_from_py_func
    func_outputs = python_func(*func_args, **func_kwargs)
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py", line 600, in wrapped_fn
    return weak_wrapped_fn().__wrapped__(*args, **kwds)
  File "/Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/framework/func_graph.py", line 973, in wrapper
    raise e.ag_error_metadata.to_exception(e)
TypeError: in user code:

    /Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py:806 train_function  *
        return step_function(self, iterator)
    /Users/user/golorry/tf/my_model.py:30 call  *
        return tf.keras.utils.to_categorical(x)
    /Users/user/.virtualenvs/tf/lib/python3.8/site-packages/tensorflow/python/keras/utils/np_utils.py:69 to_categorical  **
        y = np.array(y, dtype='int')

    TypeError: __array__() takes 1 positional argument but 2 were given

Is this intended or a bug?

Answer 1

I suggest u tf.one_hot

class TestModel1(tf.keras.Model):
    def __init__(self):
        super(TestModel1, self).__init__()
    def call(self, x):
        return tf.one_hot(x, depth=5)

class TestModel2(tf.keras.Model):
    def __init__(self):
        super(TestModel2, self).__init__()
        self.d1 = tf.keras.layers.Dense(2, activation='relu')
    def call(self, x):
        x = tf.one_hot(x, depth=5)
        return self.d1(x)

here a running notebook https://colab.research.google.com/drive/1udRnIdYGO0iBE3PfLBCJ3zT61a4qGWd2?usp=sharing