module 'tensorflow' has no attribute 'get_default_graph' - i don't want any graph

Question

for my master degree, I'm trying to create a simple neuronal network. But there are some errors in my code, so the programm stops and doesn't create a trained model.

I couldn't figure out what the error message wants to tell me and what I need to change in my code. Hence I need your help. I googled the error, but neither understood, nor could I solve my error in any way with the proposed ideas of other posts.

Can anyone explain me why tensorflow wants to create a graph and how it is possible that the framework doesn't know the needed function for it? Do I just have to install a package for the visualisation? Is it Possible to ignore this error?

I don't need any graphics. But does the computer need it for the classification and calculation with a ml-algorithm?

Please excuse my poor English and my unawareness of Tensorflow either.

Thanks in advance!

I've installed the newest tensorflow version 2.0.0-beta1, as well as the latest keras version.

Moreover, I've tried to create some graphs to show the classification process. Doesn't work.

I also activated the step-by-step debugging mode to find out my problem. It seems the error appears inside the evaluate_model function in which I create, train and evaluate a neuronal network.

The error occurs during the model creation process (model = Sequantial()).

# -*- coding: utf-8 -*-
"""
Created on Wed Apr  3 16:26:14 2019

@author: mattdoe
"""

from data_preprocessor_db import data_storage # validation data
from sklearn.model_selection import StratifiedKFold
from sklearn.metrics import confusion_matrix
from keras.utils import to_categorical
from keras.models import Sequential
from keras.layers import Dense
from keras.utils import normalize
from numpy import mean
from numpy import std
from numpy import array



# create and evaluate a single multi-layer-perzeptron
def evaluate_model(Train, Test, Target_Train, Target_Test):
    # define model
    model = Sequential()
    # input layer automatically created
    model.add(Dense(9, input_dim=9, kernel_initializer='normal', activation='relu')) # 1st hidden layer
    model.add(Dense(9, kernel_initializer='normal', activation='relu')) # 2nd hidden layer
    model.add(Dense(9, activation='softmax')) #output layer

    # create model
    model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

    # fit model
    model.fit(Train, to_categorical(Target_Train), epochs=50, verbose=0)

    # evaluate the model
    test_loss, test_acc = model.evaluate(Test, to_categorical(Target_Test), verbose=0)

    # as well: create a confussion matrix
    predicted = model.predict(Test)
    conf_mat = confusion_matrix(Target_Test, predicted)

    return model, test_acc, conf_mat



# for seperation of data_storage
# Link_ID = []
Input, Output = list(), list()

# list all results of k-fold cross-validation
scores, members, matrix = list(), list(), list()

# seperate data_storage in Input and Output data
for items in data_storage:
    # Link_ID = items[0] # identifier not needed
    Input.append([items[1], items[2], items[3], items[4], items[5], items[6], items[7], items[8], items[9]]) # Input: all characteristics
    Output.append(items[10]) # Output: scenario_class 1 to 8

# change to numpy_array (scalar index array)
Input = array(Input)
Output = array(Output)

# normalize Data
Input = normalize(Input)
# Output = normalize(Output) not needed; categorical number

# prepare k-fold cross-validation
kfold = StratifiedKFold(n_splits=15, random_state=1, shuffle=True)

for train_ix, test_ix in kfold.split(Input, Output):
    # select samples
    Train, Target_Train = Input[train_ix], Output[train_ix]
    Test, Target_Test = Input[test_ix], Output[test_ix]

    # evaluate model
    model, test_acc, conf_mat = evaluate_model(Train, Test, Target_Train, Target_Test)

    # display each evalution result
    print('>%.3f' % test_acc)

    # add result to list
    scores.append(test_acc)
    members.append(model)
    matrix.append(conf_mat)

# summarize expected performance
print('Estimated Accuracy %.3f (%.3f)' % (mean(scores), std(scores)))
# as well in confursion_matrix
print ('Confussion Matrix %' %(mean(matrix)))



# save model // trained neuronal network
model.save('neuronal_network_1.h5')

This Traceback is shown in Spyder:

Traceback (most recent call last):

  File "<ipython-input-12-25afb095a816>", line 1, in <module>
    runfile('C:/Workspace/Master-Thesis/Programm/MapValidationML/ml_neuronal_network_1.py', wdir='C:/Workspace/Master-Thesis/Programm/MapValidationML')

  File "C:\ProgramData\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 786, in runfile
    execfile(filename, namespace)

  File "C:\ProgramData\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 110, in execfile
    exec(compile(f.read(), filename, 'exec'), namespace)

  File "C:/Workspace/Master-Thesis/Programm/MapValidationML/ml_neuronal_network_1.py", line 77, in <module>
    model, test_acc, conf_mat = evaluate_model(Train, Test, Target_Train, Target_Test)

  File "C:/Workspace/Master-Thesis/Programm/MapValidationML/ml_neuronal_network_1.py", line 24, in evaluate_model
    model = Sequential()

  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\engine\sequential.py", line 87, in __init__
    super(Sequential, self).__init__(name=name)

  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\legacy\interfaces.py", line 91, in wrapper
    return func(*args, **kwargs)

  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\engine\network.py", line 96, in __init__
    self._init_subclassed_network(**kwargs)

  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\engine\network.py", line 294, in _init_subclassed_network
    self._base_init(name=name)

  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\engine\network.py", line 109, in _base_init
    name = prefix + '_' + str(K.get_uid(prefix))

  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py", line 74, in get_uid
    graph = tf.get_default_graph()

AttributeError: module 'tensorflow' has no attribute 'get_default_graph'

Answer 1

Change the imported module.Hope this method can solve your mistake.

from tensorflow.keras.utils import to_categorical
from tensorflow.keras.models  import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.utils import normalize

Answer 2

If you are using tf 2.0 beta make sure that all your keras imports are tensorflow.keras... any keras imports will pickup the standard keras package that assumes tensorflow 1.4.

i.e. use:

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, ...