Unable to construct simple LSTM layer

[dipanjannag]

Using latest Keras. Code used:

import keras.backend as K
from keras.layers.core import Dense
from keras.layers.recurrent import LSTM
from keras.models import Sequential

model = Sequential()
model.add(LSTM(10, input_shape=(2,3)))
model.add(Dense(1, activation='relu'))

Getting the following error, Although I'm able to put any non-recurrent layer as input layer.

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-3-611aef6e5d2e> in <module>()
      1 model = Sequential()
----> 2 model.add(LSTM(10, input_shape=(2,3)))
      3 model.add(Dense(1, activation='relu'))

/home/user/anaconda2/lib/python2.7/site-packages/Keras-2.1.6-py2.7.egg/keras/engine/sequential.pyc in add(self, layer)
    164                     # and create the node connecting the current layer
    165                     # to the input layer we just created.
--> 166                     layer(x)
    167                     set_inputs = True
    168                 else:

/home/user/anaconda2/lib/python2.7/site-packages/Keras-2.1.6-py2.7.egg/keras/layers/recurrent.pyc in __call__(self, inputs, initial_state, constants, **kwargs)
    498 
    499         if initial_state is None and constants is None:
--> 500             return super(RNN, self).__call__(inputs, **kwargs)
    501 
    502         # If any of `initial_state` or `constants` are specified and are Keras

/home/user/anaconda2/lib/python2.7/site-packages/Keras-2.1.6-py2.7.egg/keras/engine/base_layer.pyc in __call__(self, inputs, **kwargs)
    452             # Actually call the layer,
    453             # collecting output(s), mask(s), and shape(s).
--> 454             output = self.call(inputs, **kwargs)
    455             output_mask = self.compute_mask(inputs, previous_mask)
    456 

/home/user/anaconda2/lib/python2.7/site-packages/Keras-2.1.6-py2.7.egg/keras/layers/recurrent.pyc in call(self, inputs, mask, training, initial_state)
   2110                                       mask=mask,
   2111                                       training=training,
-> 2112                                       initial_state=initial_state)
   2113 
   2114     @property

/home/user/anaconda2/lib/python2.7/site-packages/Keras-2.1.6-py2.7.egg/keras/layers/recurrent.pyc in call(self, inputs, mask, training, initial_state, constants)
    607                                              mask=mask,
    608                                              unroll=self.unroll,
--> 609                                              input_length=timesteps)
    610         if self.stateful:
    611             updates = []

/home/user/anaconda2/lib/python2.7/site-packages/Keras-2.1.6-py2.7.egg/keras/backend/tensorflow_backend.pyc in rnn(step_function, inputs, initial_states, go_backwards, mask, constants, unroll, input_length)
   2927             parallel_iterations=32,
   2928             swap_memory=True,
-> 2929             maximum_iterations=input_length)
   2930         last_time = final_outputs[0]
   2931         output_ta = final_outputs[1]

TypeError: while_loop() got an unexpected keyword argument 'maximum_iterations'

EDIT: It seems keras master branch does not work with tensorflow 1.4.1. Is that an expected behaviour?
https://travis-ci.org/deeiip/keras/jobs/382610473#L1917

[ryubidragonfire]

I have the same issue. Hope that can get help soon.

[izuro]

Had the same issue with Keras 2.2.0 & tensorflow-gpu 1.8.0 . Reverted back to 2.1.6 and model fits without error.
pip uninstall keras
pip install keras==2.1.6

[dipanjannag]

@izuro seems to be working. But the it seems version 2.2.0 is not compaitable with tensorflow 1.4.1

[romass12]

I am getting this error : if isinstance(identifier, tf.train.Optimizer):
NameError: name 'tf' is not defined
TF : 1.4.0
Keras : 2.1.6

MODEL

import os
import importlib
import warnings
import tensorflow as tf
import keras
warnings.filterwarnings(action='ignore', category=DeprecationWarning)
def set_keras_backend(backend):
    from keras import backend as K

    if K.backend() != backend:
        os.environ['KERAS_BACKEND'] = backend
        importlib.reload(K)
        assert K.backend() == backend
set_keras_backend("tensorflow")

from keras.models import load_model
def model(embedding_matrix,embed,len_distinct,len_label,epochs,batch_size,weight,vector_size):
    import tensorflow as tf
    model = Sequential()
    model.add(Embedding(len_distinct, vector_size, input_length=maxlen,weights=[embedding_matrix],trainable=True))
    model.add(Bidirectional(LSTM(100,return_sequences=False,dropout=0.2))) # was false
    model.add(Dense(len_label, activation=tf.nn.softmax))
    adam=tf.train.AdamOptimizer(learning_rate=0.0001)
    model.compile(loss='categorical_crossentropy', optimizer=adam, metrics=['accuracy',f1score,sensitivity,precision])

[praveenraghuvanshi1512]

I am getting this error : if isinstance(identifier, tf.train.Optimizer):
NameError: name 'tf' is not defined
TF : 1.4.0
Keras : 2.1.6

MODEL

import os
import importlib
import warnings
import tensorflow as tf
import keras
warnings.filterwarnings(action='ignore', category=DeprecationWarning)
def set_keras_backend(backend):
    from keras import backend as K

    if K.backend() != backend:
        os.environ['KERAS_BACKEND'] = backend
        importlib.reload(K)
        assert K.backend() == backend
set_keras_backend("tensorflow")

from keras.models import load_model
def model(embedding_matrix,embed,len_distinct,len_label,epochs,batch_size,weight,vector_size):
    import tensorflow as tf
    model = Sequential()
    model.add(Embedding(len_distinct, vector_size, input_length=maxlen,weights=[embedding_matrix],trainable=True))
    model.add(Bidirectional(LSTM(100,return_sequences=False,dropout=0.2))) # was false
    model.add(Dense(len_label, activation=tf.nn.softmax))
    adam=tf.train.AdamOptimizer(learning_rate=0.0001)
    model.compile(loss='categorical_crossentropy', optimizer=adam, metrics=['accuracy',f1score,sensitivity,precision])

I am also getting the same error "name 'tf' is not defined" for a simple network.
Tried 'import tensorflow as tf' without success.
TF: 1.8.0
Keras: 2.1.6

[praveenraghuvanshi1512]

I am getting this error : if isinstance(identifier, tf.train.Optimizer):
NameError: name 'tf' is not defined
TF : 1.4.0
Keras : 2.1.6
MODEL

import os
import importlib
import warnings
import tensorflow as tf
import keras
warnings.filterwarnings(action='ignore', category=DeprecationWarning)
def set_keras_backend(backend):
    from keras import backend as K

    if K.backend() != backend:
        os.environ['KERAS_BACKEND'] = backend
        importlib.reload(K)
        assert K.backend() == backend
set_keras_backend("tensorflow")

from keras.models import load_model
def model(embedding_matrix,embed,len_distinct,len_label,epochs,batch_size,weight,vector_size):
    import tensorflow as tf
    model = Sequential()
    model.add(Embedding(len_distinct, vector_size, input_length=maxlen,weights=[embedding_matrix],trainable=True))
    model.add(Bidirectional(LSTM(100,return_sequences=False,dropout=0.2))) # was false
    model.add(Dense(len_label, activation=tf.nn.softmax))
    adam=tf.train.AdamOptimizer(learning_rate=0.0001)
    model.compile(loss='categorical_crossentropy', optimizer=adam, metrics=['accuracy',f1score,sensitivity,precision])

I am also getting the same error "name 'tf' is not defined" for a simple network.
Tried 'import tensorflow as tf' without success.
TF: 1.8.0
Keras: 2.1.6

Upgraded tensforflow to latest(2.0.0) and replaced import statement from 'from keras.models import Sequential' to 'from tensorflow.keras.models import Sequential' with tensorlfow prefix and everything worked... 😊