Keras callbacks Tensorboard multi-sortie Erreur

Question

Je n'arrive pas à comprendre quelle est l'origine de cette erreur pour le modèle Keras à sorties multiples lors de l'utilisation de callbacks.TensorBoard.

tbCallBack = keras.callbacks.TensorBoard(log_dir = logdir, histogram_freq = 1, write_graph = 1, write_images = 0, write_grads = 1) 
###No errror when not using callbacks 
regr.fit( Ax_train, [Ay_train_p, Ay_train_s], validation_data= (Ax_test, [Ay_test_p, Ay_test_s]), epochs = 500, batch_size = 10, verbose = 1) 
###No errror when not using validation_data 
regr.fit( Ax_train, [Ay_train_p, Ay_train_s], epochs = 500, batch_size = 10, verbose = 1, callbacks=[tbCallBack]) 
###Error Occurred 
regr.fit(Ax_train, [Ay_train_p, Ay_train_s], validation_data= (Ax_test, [Ay_test_p, Ay_test_s]), epochs = 500, batch_size = 10, verbose = 1, callbacks=[tbCallBack]) 

Erreur

Epoch 1/500 
1280/1663 [======================>.......] - ETA: 0s - loss: 1.6230 - output_power_loss: 0.9627 - output_slack_loss: 0.66032017-07-11 03:17:27.964542: W tensorflow/core/framework/op_kernel.cc:1148] Invalid argument: Shape [-1] has negative dimensions 
2017-07-11 03:17:27.964589: E tensorflow/core/common_runtime/executor.cc:644] Executor failed to create kernel. Invalid argument: Shape [-1] has negative dimensions 
    [[Node: output_slack_sample_weights = Placeholder[dtype=DT_FLOAT, shape=[?], _device="/job:localhost/replica:0/task:0/cpu:0"]()]] 
2017-07-11 03:17:27.970690: W tensorflow/core/framework/op_kernel.cc:1148] Invalid argument: Shape [-1] has negative dimensions 
2017-07-11 03:17:27.970735: E tensorflow/core/common_runtime/executor.cc:644] Executor failed to create kernel. Invalid argument: Shape [-1] has negative dimensions 
    [[Node: output_power_sample_weights = Placeholder[dtype=DT_FLOAT, shape=[?], _device="/job:localhost/replica:0/task:0/cpu:0"]()]] 
2017-07-11 03:17:27.972004: W tensorflow/core/framework/op_kernel.cc:1148] Invalid argument: Shape [-1] has negative dimensions 
2017-07-11 03:17:27.972026: E tensorflow/core/common_runtime/executor.cc:644] Executor failed to create kernel. Invalid argument: Shape [-1] has negative dimensions 
    [[Node: output_power_sample_weights = Placeholder[dtype=DT_FLOAT, shape=[?], _device="/job:localhost/replica:0/task:0/cpu:0"]()]] 
Traceback (most recent call last): 
    File "tf_keras.py", line 183, in <module> 
    regr.fit(Ax_train, [Ay_train_p, Ay_train_s], validation_data= (Ax_test, [Ay_test_p, Ay_test_s]), epochs = 500, batch_size = 10, verbose = 1, callbacks=[tbCallBack]) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/keras/engine/training.py", line 1507, in fit 
    initial_epoch=initial_epoch) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/keras/engine/training.py", line 1176, in _fit_loop 
    callbacks.on_epoch_end(epoch, epoch_logs) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/keras/callbacks.py", line 77, in on_epoch_end 
    callback.on_epoch_end(epoch, logs) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/keras/callbacks.py", line 768, in on_epoch_end 
    result = self.sess.run([self.merged], feed_dict=feed_dict) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 789, in run 
    run_metadata_ptr) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 997, in _run 
    feed_dict_string, options, run_metadata) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1132, in _do_run 
    target_list, options, run_metadata) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1152, in _do_call 
    raise type(e)(node_def, op, message) 
tensorflow.python.framework.errors_impl.InvalidArgumentError: Shape [-1] has negative dimensions 
    [[Node: output_power_sample_weights = Placeholder[dtype=DT_FLOAT, shape=[?], _device="/job:localhost/replica:0/task:0/cpu:0"]()]] 

Caused by op u'output_power_sample_weights', defined at: 
    File "tf_keras.py", line 180, in <module> 
    regr = nn_model() 
    File "tf_keras.py", line 177, in nn_model 
    model.compile(optimizer = 'adam', loss ={'output_power': 'mean_squared_error', 'output_slack': 'binary_crossentropy'}) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/keras/engine/training.py", line 870, in compile 
    name=name + '_sample_weights')) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/keras/backend/tensorflow_backend.py", line 431, in placeholder 
    x = tf.placeholder(dtype, shape=shape, name=name) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/tensorflow/python/ops/array_ops.py", line 1530, in placeholder 
    return gen_array_ops._placeholder(dtype=dtype, shape=shape, name=name) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/tensorflow/python/ops/gen_array_ops.py", line 1954, in _placeholder 
    name=name) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 767, in apply_op 
    op_def=op_def) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2506, in create_op 
    original_op=self._default_original_op, op_def=op_def) 
    File "/home/roy/keras_tf/local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1269, in __init__ 
    self._traceback = _extract_stack() 

InvalidArgumentError (see above for traceback): Shape [-1] has negative dimensions 
    [[Node: output_power_sample_weights = Placeholder[dtype=DT_FLOAT, shape=[?], _device="/job:localhost/replica:0/task:0/cpu:0"]()]] 

Que signifie "Forme [-1] a des dimensions négatives" signifie? J'avais également essayé avec chaque sortie avec des callbacks.Tensorboard et il n'y avait aucune erreur s'est produite. Recherchez aussi avec "Node: output_power_sample_weights" mais pas de résultats.

regr.summary()

____________________________________________________________________________________________________ 
Layer (type)      Output Shape   Param #  Connected to      
==================================================================================================== 
main_input (InputLayer)   (None, 5)    0            
____________________________________________________________________________________________________ 
Hidden (Dense)     (None, 5)    30   main_input[0][0]     
____________________________________________________________________________________________________ 
output_power (Dense)    (None, 1)    6   Hidden[0][0]      
____________________________________________________________________________________________________ 
output_slack (Dense)    (None, 1)    6   Hidden[0][0]      
==================================================================================================== 
Total params: 42 
Trainable params: 42 
Non-trainable params: 0 

Answer 1

Le "Shape [-1] a des dimensions négatives" message d'erreur est un (trompeur) erreur affiché lorsque vous ne nourrissez pas tf.placeholder() qui a une partie définie forme. Ce message d'erreur a été remplacé par un message plus informatif dans la version nocturne de TensorFlow.

L'erreur et la trace de la pile suggère que cet appel échoue Session.run() (source):

File "/home/roy/keras_tf/local/lib/python2.7/site-packages/keras/callbacks.py", line 768, in on_epoch_end 
    result = self.sess.run([self.merged], feed_dict=feed_dict) 

En regardant the implementation, il semble que ce n'est appelée lorsque le rappel TensorBoard est installé et self.validation_data is not None. De l'implémentation, il semble que ce rappel est incompatible avec les modèles qui dépendent des espaces réservés qui ne sont pas one of (i) model.inputs, (ii) model.targets, ou (iii) model.sample_weights. Il pourrait être préférable d'ouvrir un problème sur le Keras GitHub issue tracker à ce sujet.