How to solve AttributeError: module 'tensorflow.contrib.learn' has no attribute 'estimators' about chatbot-retrieval HOT 4 OPEN

I did the following change, and some how it worked for me.
first, add this import command in "udc_predict.py"
from tensorflow.contrib.learn.python.learn.estimators import *

change the following line from
estimator._targets_info = tf.contrib.learn.estimators.tensor_signature.TensorSignature(tf.constant(0, shape=[1,1]))
to
estimator._targets_info = tensor_signature.TensorSignature(tf.constant(0, shape=[1,1]))

Also at the bottom, change the print command from
print("{}: {:g}".format(r, prob[0,0]))
to
print("{}: {:g}".format(r, next(prob)[0]))

Hope it helps. I am using Tensorflow Version 1.8, Python 3.6.5

from chatbot-retrieval.

i cannot find any useful solution _ORZ

from chatbot-retrieval.

having the similar issue in google colab? tf 2.0

feature_columns = tf.contrib.learn.infer_real_valued_columns_from_input_fn
module 'tensorflow' has no attribute 'contrib'

from chatbot-retrieval.

1. the error is in function 'create_model_fn' in udc_model.py,
   move "batch_size = targets.get_shape().as_list()[0]" after "if mode == tf.contrib.learn.ModeKeys.EVAL:" can fix this problem.

the reason is when in PREDICT mode, model_fn deal the param 'targets' as None.

2. "prob = estimator.predict(input_fn=input_fn, as_iterable=False)" in cdc_predict.py,
   the param 'as_iterable' default value is True, change to False,
   the next line "prob[0,0]" can execute correct.

from chatbot-retrieval.