Error When Using CRF with tf.keras

Converted call: <bound method CRF.call of <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990>> args: (<tf.Tensor 'add_24/Identity:0' shape=(None, None, 128) dtype=float32>,) kwargs: {}

Not whitelisted: <method-wrapper 'call' of method object at 0x7fc5d70bf690>: default rule Not whitelisted: <class 'keras_contrib.layers.crf.CRF'>: default rule Not whitelisted: <bound method CRF.call of <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990>>: default rule Cache hit for entity <bound method CRF.call of <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990>> key <code object call at 0x7fc4c161d8a0, file "/home/tonychenxyz/anaconda3/envs/tf210/lib/python3.7/site-packages/tf_keras_contrib-2.0.8-py3.7.egg/keras_contrib/layers/crf.py", line 287> subkey (<tensorflow.python.autograph.core.converter.ConversionOptions object at 0x7fc5bc09eb10>, frozenset()): _ConvertedEntityFactoryInfo(tfcall in tmpp0p91xwy) Defaults of <function create_converted_entity_factory..create_converted_entity..tf__call at 0x7fc5bc082320> : (None,) KW defaults of <function create_converted_entity_factory..create_converted_entity..tfcall at 0x7fc5bc082320> : None Calling <function create_converted_entity_factory..create_converted_entity..tf__call at 0x7fc5bc082320> with self: <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990> X: Tensor("add_24/Identity:0", shape=(None, None, 128), dtype=float32) mask: None

Converted call: <bound method CRF.viterbi_decoding of <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990>> args: (<tf.Tensor 'add_24/Identity:0' shape=(None, None, 128) dtype=float32>, None) kwargs: None

Not whitelisted: <method-wrapper 'call' of method object at 0x7fc5d76dc9b0>: default rule Not whitelisted: <class 'keras_contrib.layers.crf.CRF'>: default rule Not whitelisted: <bound method CRF.viterbi_decoding of <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990>>: default rule Entity <bound method CRF.viterbi_decoding of <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990>> is not cached for key <code object viterbi_decoding at 0x7fc4c16174b0, file "/home/tonychenxyz/anaconda3/envs/tf210/lib/python3.7/site-packages/tf_keras_contrib-2.0.8-py3.7.egg/keras_contrib/layers/crf.py", line 558> subkey (<tensorflow.python.autograph.core.converter.ConversionOptions object at 0x7fc5bc09ea10>, frozenset()) Converting <bound method CRF.viterbi_decoding of <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990>> Source code of <bound method CRF.viterbi_decoding of <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990>>:

from future import absolute_import from future import division def viterbi_decoding(self, X, mask=None): input_energy = self.activation(K.dot(X, self.kernel) + self.bias) if self.use_boundary: input_energy = self.add_boundary_energy( input_energy, mask, self.left_boundary, self.right_boundary)

argmin_tables = self.recursion(input_energy, mask, return_logZ=False)
argmin_tables = K.cast(argmin_tables, 'int32')

# backward to find best path, `initial_best_idx` can be any,
# as all elements in the last argmin_table are the same
argmin_tables = K.reverse(argmin_tables, 1)
# matrix instead of vector is required by tf `K.rnn`
initial_best_idx = [K.expand_dims(argmin_tables[:, 0, 0])]
if K.backend() == 'theano':
    from theano import tensor as T
    initial_best_idx = [T.unbroadcast(initial_best_idx[0], 1)]

def gather_each_row(params, indices):
    n = K.shape(indices)[0]
    if K.backend() == 'theano':
        from theano import tensor as T
        return params[T.arange(n), indices]
    elif K.backend() == 'tensorflow':
        import tensorflow as tf
        indices = K.transpose(K.stack([tf.range(n), indices]))
        return tf.gather_nd(params, indices)
    else:
        raise NotImplementedError

def find_path(argmin_table, best_idx):
    next_best_idx = gather_each_row(argmin_table, best_idx[0][:, 0])
    next_best_idx = K.expand_dims(next_best_idx)
    if K.backend() == 'theano':
        from theano import tensor as T
        next_best_idx = T.unbroadcast(next_best_idx, 1)
    return next_best_idx, [next_best_idx]

_, best_paths, _ = K.rnn(find_path, argmin_tables, initial_best_idx,
input_length=K.int_shape(X)[1], unroll=self.unroll)
best_paths = K.reverse(best_paths, 1)
best_paths = K.squeeze(best_paths, 2)

return K.one_hot(best_paths, self.units)

Error transforming entity <bound method CRF.viterbi_decoding of <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990>> WARNING: AutoGraph could not transform <bound method CRF.viterbi_decoding of <keras_contrib.layers.crf.CRF object at 0x7fc4937fb990>> and will run it as-is. Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output. Cause: LIVE_VARS_IN Converted call: <function dot at 0x7fc62c801cb0> args: (<tf.Tensor 'add_24/Identity:0' shape=(None, None, 128) dtype=float32>, <tf.Variable 'crf/kernel:0' shape=(128, 11) dtype=float32>) kwargs: None

Converted call: <function zeros_like at 0x7fc62c801680> args: (<tf.Tensor 'crf/Reshape_5:0' shape=(None, None, 11) dtype=float32>,) kwargs: None

Converted call: <function in_train_phase at 0x7fc62c80a950> args: (<tf.Tensor 'crf/zeros_like_2:0' shape=(None, None, 11) dtype=float32>, <tf.Tensor 'crf/one_hot:0' shape=(None, None, 11) dtype=float32>) kwargs: None

keras-team / keras-contrib

Error When Using CRF with tf.keras #540