AttributeError: module 'keras.backend' has no attribute '_BACKEND'

Question

I am following a book on building chat bots and continue running into this error when attempting to start interactive learning.

The full error is this:

Traceback (most recent call last): File "train_initialize.py", line 18, in agent = Agent("horoscope_domain.yml", policies = [MemoizationPolicy(), KerasPolicy()]) File "C:\Users\Max\AppData\Local\Programs\Python\Python37\lib\site-packages\rasa_core\policies\keras_policy.py", line 31, in init if KerasPolicy.is_using_tensorflow() and not graph: File "C:\Users\Max\AppData\Local\Programs\Python\Python37\lib\site-packages\rasa_core\policies\keras_policy.py", line 48, in is_using_tensorflow return keras.backend._BACKEND == "tensorflow" AttributeError: module 'keras.backend' has no attribute '_BACKEND'

my code looks like this '''

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from __future__ import unicode_literals

from rasa_core import utils
import tensorflow.keras.backend
from rasa_core.agent import Agent
from rasa_core.policies.keras_policy import KerasPolicy
from rasa_core.policies.memoization import MemoizationPolicy
from rasa_core.policies.sklearn_policy import SklearnPolicy

if __name__ == '__main__':
    utils.configure_colored_logging(loglevel="DEBUG")

    training_data_file = './data/stories.md'
    model_path = './models/dialogue'
    agent = Agent("horoscope_domain.yml", policies = [MemoizationPolicy(), KerasPolicy()])

    training_data = agent.load_data(training_data_file)

    agent.train(training_data, augmentation_factor = 50, epochs = 500, batch_size = 10, validation_split = 0.2)
    agent.persist(model_path)

'''

score 0 · Accepted Answer · answered Nov 12 '19 at 23:06

0

Looks like outdated API code; open the files in the error trace, and replace ._BACKEND w/ .backend():

# In "C:\Users\Max\AppData\Local\Programs\Python\Python37\lib\site-packages
# \rasa_core\policies\keras_policy.py", line 48:

# return keras.backend._BACKEND == "tensorflow" # <-- DELETE
return keras.backend.backend() == "tensorflow"  # <-- PASTE

answered Nov 12 '19 at 23:06

OverLordGoldDragon

1
9
53
101

That worked, now i'm getting this error though ** Traceback (most recent call last): File "train_initialize.py", line 18, in agent = Agent("horoscope_domain.yml", policies = [MemoizationPolicy(), KerasPolicy()]) File "C:\Users\Max\AppData\Local\Programs\Python\Python37\lib\site-packages\rasa_core\policies\keras_policy.py", line 33, in __init__ self.graph = keras.backend.tf.get_default_graph() AttributeError: module 'keras.backend' has no attribute 'tf' ** – Max_Meadowcroft Nov 12 '19 at 23:13
@Max_Meadowcroft Whatever this "rasa" API you're using is, I suggest using something else - it appears ridden with bugs. Notify its creator of these issues - and I doubt they'll be stopping here. The question, as originally asked, is resolved - you can ask a new one if need be. But due to the error's simplicity, I'll answer regardless: remove `keras.backend.` entirely from the line of code in the error trace - it should just be `tf.get_default_graph()` – OverLordGoldDragon Nov 12 '19 at 23:23
@OverLordGoldDragon Can you please take a look into this question? https://stackoverflow.com/q/58900947/5904928 I am struggling for hours, couldn't find an answer. – Aaditya Ura Nov 17 '19 at 13:47

AttributeError: module 'keras.backend' has no attribute '_BACKEND'

1 Answers1