Keras: AttributeError: 'Model' object has no attribute 'stateful_metric_names'

[AziziShekoofeh]

Hi,

I just upgrade my Keras to latest version and getting a error related to keras/engine/training on the code that used to work fine beforehand (my previous version was 2.1.0). I am not sure if it's some kind of deprectation or etc.

Traceback (most recent call last):
  File "template_model_1.py", line 118, in <module>
    acc, ll = test_MC_dropout(test_model, X_test, Y_test)
  File "/data/home/shekoofeh/Project/Dropout-BB/BBalpha-TeUS/BBalpha_dropout.py", line 72, in test_MC_dropout
    pred = model.predict(X, verbose=1)  # N x K x D
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1842, in predict
    verbose=verbose, steps=steps)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1292, in _predict_loop
    stateful_metrics=self.stateful_metric_names)
AttributeError: 'Model' object has no attribute 'stateful_metric_names'

It's happening on a brach of Dropout + BB-alpha for detecting adversarial examples. Unfortunately, I coudn't share my own version due to disclusure of the project. Thanks for being understanding.

I found Add Stateful (Global) Metrics #9200, but I couldn't track down changes. Any idea how to solve the issue?

Thanks,
Shek

• [checked] Check that you are up-to-date with the master branch of Keras. You can update with:
  pip install git+git://github.com/keras-team/keras.git --upgrade --no-deps

• [checked] If running on TensorFlow, check that you are up-to-date with the latest version. The installation instructions can be found here.

[Dref360]

Have you compiled your model?
The field is initialized there : https://github.com/keras-team/keras/blob/master/keras/engine/training.py#L855

[AziziShekoofeh]

@Dref360 Thanks for your answer. Yes.
I used the exact same code before updating my Keras version, and It was working fine. Nothing has changed in the code.

[ehrlinger]

I'm also seeing this error on Keras v2.1.4 only.

For more information, I'm trying to load the model from files:

```{python}
from keras.models import model_from_json

load json and create model

json_file = open(LSTM_MODEL, 'r')
loaded_model_json = json_file.read()
loaded_model = model_from_json(loaded_model_json)

load weights into new model

loaded_model.load_weights(MODEL_WEIGHTS)

The error occurs in the predict_proba call, which should not need compiling.

score = loaded_model.predict_proba(seq_array,verbose=1)
print(score.shape)

print(score)

AttributeError Traceback (most recent call last)
in ()
----> 1 score = loaded_model.predict_proba(seq_array,verbose=1)
2 print(score.shape)
3 print(score)

~/lib/conda/envs/azureml_7753abb5a4b4d3655662bae7f7e3b6fe/lib/python3.5/site-packages/keras/models.py in predict_proba(self, x, batch_size, verbose, steps)
1110 A Numpy array of probability predictions.
1111 """
-> 1112 preds = self.predict(x, batch_size, verbose, steps=steps)
1113 if preds.min() < 0. or preds.max() > 1.:
1114 warnings.warn('Network returning invalid probability values. '

~/lib/conda/envs/azureml_7753abb5a4b4d3655662bae7f7e3b6fe/lib/python3.5/site-packages/keras/models.py in predict(self, x, batch_size, verbose, steps)
1023 self.build()
1024 return self.model.predict(x, batch_size=batch_size, verbose=verbose,
-> 1025 steps=steps)
1026
1027 def predict_on_batch(self, x):

~/lib/conda/envs/azureml_7753abb5a4b4d3655662bae7f7e3b6fe/lib/python3.5/site-packages/keras/engine/training.py in predict(self, x, batch_size, verbose, steps)
1840 f = self.predict_function
1841 return self._predict_loop(f, ins, batch_size=batch_size,
-> 1842 verbose=verbose, steps=steps)
1843
1844 def train_on_batch(self, x, y,

~/lib/conda/envs/azureml_7753abb5a4b4d3655662bae7f7e3b6fe/lib/python3.5/site-packages/keras/engine/training.py in _predict_loop(self, f, ins, batch_size, verbose, steps)
1290 else:
1291 progbar = Progbar(target=num_samples,
-> 1292 stateful_metrics=self.stateful_metric_names)
1293
1294 indices_for_conversion_to_dense = []

AttributeError: 'Model' object has no attribute 'stateful_metric_names'

On Keras 2.1.3, this returned the expected output.

score = loaded_model.predict_proba(seq_array,verbose=1)
print(score.shape)
print(score)
15631/15631 [==============================] - 11s 698us/step
(15631, 1)
[[3.7404552e-05]
[3.7972197e-05]
[3.8670321e-05]
...
[9.9917173e-01]
[9.9919158e-01]
[9.9921858e-01]]

```

[Dref360]

Quick fix for both of you and I'll do something more permanent
just do a dummy model.compile

loaded_model = model_from_json(loaded_model_json)
# load weights into new model
loaded_model.load_weights(MODEL_WEIGHTS)
loaded_model.compile('sgd','mse')

This won't affect your results for prediction in any way.

[ehrlinger]

works for me. Thanks.

[AziziShekoofeh]

@Dref360 thank you very much. It seems, it solved my issue as well.

[alyato]

@Dref360 thanks. so nice.

[tavishm]

this is my code

• `

1. from keras.preprocessing.image import load_img
   from keras.preprocessing.image import ImageDataGenerator
   import csv, re, os
   import numpy as np
   from keras.preprocessing.image import img_to_array
   from keras.applications.imagenet_utils import decode_predictions
   import matplotlib.pyplot as plt
   import cv2
   import keras
   from keras import Model
   from keras.layers import Dense, Dropout
   import numpy as np
   from keras.utils import plot_model
   from keras.applications import inception_v3
   os.environ['CUDA_VISIBLE_DEVICES'] = '0'
   from utils import layer_utils

densenet_model = keras.applications.densenet.DenseNet201(weights=None, include_top=False, input_shape=(None,None,3))

all_layers = layer_utils.get_layer_names_bw_2_layers(densenet_model, 'input_1', 'pool 1')

input_img = densenet_model.get_layer('input_1').output
encoded = densenet_model.get_layer('pool1')(input_img)
decoded = densenet_model.get_layer('input_1')(encoded)

model = Model(input_img, decoded)

model.compile(loss="binary_crossentropy", optimizer='adadelta')

mypath = 'images/'

encoder = Model(input_img, encoded)
encoded_input = keras.layers.Input(shape=(1024,1024,3))
decoder_layer = model.layers[-1]
decoder = Model(encoded_input, decoder_layer(encoded_input))
train_datagen = ImageDataGenerator(rescale=1. / 255,
fill_mode='constant', cval=0.,
zoom_range=0,
rotation_range=0,
width_shift_range=0,
height_shift_range=0,
shear_range=0,
validation_split=0.1
)
train_generator = train_datagen.flow_from_directory( 'organized_data22/',batch_size=128, class_mode='input', subset='training')
test_generator = train_datagen.flow_from_directory( 'organized_data22/', batch_size=128, class_mode='input', subset='validation')

for s in range(1000):
w = s+1
print('Iteration - '+str(w))
model.fit_generator(train_generator,
epochs=1,
validation_data=test_generator,
)

    model.save('models/inception_animal_classify_new_test_train_'+str(i)+'.hf5')
    encoded_imgs = encoder.predict(x_test)
    decoded_imgs = decoder.predict(encoded_imgs)
    for i in range(n):
            # display original
            ax = plt.subplot(2, n, i + 1)

plt.imshow(x_test[i].reshape(28, 28))
plt.gray()
ax.get_xaxis().set_visible(False)
ax.get_yaxis().set_visible(False)

            # display reconstruction
            ax = plt.subplot(2, n, i + 1 + n)
            plt.imshow(decoded_imgs[i].reshape(28, 28))
            plt.gray()
            ax.get_xaxis().set_visible(False)
            ax.get_yaxis().set_visible(False)
    plt.show()

showing error
Traceback (most recent call last):
File "densenet_autoencoder.py", line 28, in
model = Model(input_img, decoded)
File "/usr/local/lib/python3.6/dist-packages/keras/legacy/interfaces.py", line 91, in wrapper
return func(args, *kwargs)
File "/usr/local/lib/python3.6/dist-packages/keras/engine/network.py", line 91, in __init__
self._init_graph_network(args, *kwargs)
File "/usr/local/lib/python3.6/dist-packages/keras/engine/network.py", line 183, in _init_graph_network
'The tensor that caused the issue was: ' +
AttributeError: 'Model' object has no attribute 'name'