chingyaoc/VQG-tensorflow

Tensorflow version?

anandmishra22 opened this issue · 3 comments

Does code is not compatible with tensorflow latest version? Which version do you use to make it work?

Did you find out which version he used?

@anandmishra22 @eladwass Did any of you get the model to train? I updated the code to make it compatible with Tensorflow 1.0 but it still won't run.

Will work at tf 1.3.0+ with some changes.

main.py
attrs = conf.__dict__['__flags']
pp(attrs )
>>>
pp(conf)

question_gen.py
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logit_words, onehot_labels) >>>
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits=logit_words, labels=onehot_labels)

labels = tf.expand_dims(1, self.question[:, i]) >>>
labels = tf.expand_dims(self.question[:, i], 1)

tf.pack >>>
tf.stack

state = tf.zeros([1, self.lstm.state_size]) >>>
state = self.lstm.zero_state(1,dtype=np.float32)

And don't forget to use https://github.com/machrisaa/tensorflow-vgg/blob/master/vgg19.py instead of embedded one.