ImportError: cannot import name 'CuDNNLSTM' from 'keras.layers'

[muammar]

I tried using SMILES-X with a recent environment, and the example notebook is failing with the following error:

---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-4-84592a845c88> in <module>
      1 import pandas as pd
      2 
----> 3 from SMILESX import main, inference
      4 get_ipython().run_line_magic('matplotlib', 'inline')

~/git/SMILES-X/SMILESX/main.py in <module>
     24 from sklearn.metrics import r2_score
     25 
---> 26 from SMILESX import utils, token, augm, model
     27 
     28 np.set_printoptions(precision=3)

~/git/SMILES-X/SMILESX/model.py in <module>
      4 from keras.layers import Embedding
      5 from keras.layers.wrappers import Bidirectional
----> 6 from keras.layers import CuDNNLSTM, TimeDistributed
      7 
      8 from keras.engine.topology import Layer

ImportError: cannot import name 'CuDNNLSTM' from 'keras.layers' (/home/elkhatim/miniconda3/envs/py38/lib/python3.8/site-packages/keras/layers/__init__.py)

This is due to the latest version of Keras deprecating CuDNNLSTM. As I am under a conda environment with Python3.8, I tried installing the version 2.3.0 of Keras but not possible:

conda install -c conda-forge keras==2.3

UnsatisfiableError: The following specifications were found
to be incompatible with the existing python installation in your environment:

Specifications:

  - keras==2.3.0 -> python[version='>=2.7,<2.8.0a0|>=3.6,<3.7.0a0|>=3.7,<3.8.0a0']

Your python: python=3.8

It seems the only way to use SMILES-X would be to install an environment with Python3.6 and use the Keras version posted above, which becomes very inconvenient on my setup to access my jupyterlab instance.

[muammar]

More advances, after installing a conda environment with python3.6, and trying to run the example I am getting this error:

---------------------------------------------------------------------------
InvalidArgumentError                      Traceback (most recent call last)
~/miniconda3/envs/py36/lib/python3.6/site-packages/tensorflow/python/client/session.py in _do_call(self, fn, *args)
   1355     try:
-> 1356       return fn(*args)
   1357     except errors.OpError as e:

~/miniconda3/envs/py36/lib/python3.6/site-packages/tensorflow/python/client/session.py in _run_fn(feed_dict, fetch_list, target_list, options, run_metadata)
   1338       # Ensure any changes to the graph are reflected in the runtime.
-> 1339       self._extend_graph()
   1340       return self._call_tf_sessionrun(

~/miniconda3/envs/py36/lib/python3.6/site-packages/tensorflow/python/client/session.py in _extend_graph(self)
   1373     with self._graph._session_run_lock():  # pylint: disable=protected-access
-> 1374       tf_session.ExtendSession(self._session)
   1375 

InvalidArgumentError: No OpKernel was registered to support Op 'CudnnRNN' used by {{node bidirectional_1/CudnnRNN}}with these attrs: [rnn_mode="lstm", seed2=0, is_training=true, dropout=0, seed=87654321, T=DT_FLOAT, input_mode="linear_input", direction="unidirectional"]
Registered devices: [CPU, XLA_CPU]
Registered kernels:
  <no registered kernels>

	 [[bidirectional_1/CudnnRNN]]

During handling of the above exception, another exception occurred:

InvalidArgumentError                      Traceback (most recent call last)
<ipython-input-44-ad7b10e344ac> in <module>
     18           alpha_ref = 22,
     19           patience = 50,
---> 20           n_epochs = 100)

~/git/SMILES-X/SMILESX/main.py in Main(data, data_name, bayopt_bounds, data_units, k_fold_number, augmentation, outdir, bayopt_n_epochs, bayopt_n_rounds, bayopt_it_factor, bayopt_on, lstmunits_ref, denseunits_ref, embedding_ref, batch_size_ref, alpha_ref, n_gpus, bridge_type, patience, n_epochs)
    304                                                             exact_feval = False,
    305                                                             normalize_Y = True,
--> 306                                                             num_cores = multiprocessing.cpu_count()-1)
    307             print("Optimization:\n")
    308             Bayes_opt.run_optimization(max_iter=bayopt_n_rounds)

~/miniconda3/envs/py36/lib/python3.6/site-packages/GPyOpt/methods/bayesian_optimization.py in __init__(self, f, domain, constraints, cost_withGradients, model_type, X, Y, initial_design_numdata, initial_design_type, acquisition_type, normalize_Y, exact_feval, acquisition_optimizer_type, model_update_interval, evaluator_type, batch_size, num_cores, verbosity, verbosity_model, maximize, de_duplication, **kwargs)
    117         self.initial_design_type  = initial_design_type
    118         self.initial_design_numdata = initial_design_numdata
--> 119         self._init_design_chooser()
    120 
    121         # --- CHOOSE the model type. If an instance of a GPyOpt model is passed (possibly user defined), it is used.

~/miniconda3/envs/py36/lib/python3.6/site-packages/GPyOpt/methods/bayesian_optimization.py in _init_design_chooser(self)
    193         if self.X is None:
    194             self.X = initial_design(self.initial_design_type, self.space, self.initial_design_numdata)
--> 195             self.Y, _ = self.objective.evaluate(self.X)
    196         # Case 2
    197         elif self.X is not None and self.Y is None:

~/miniconda3/envs/py36/lib/python3.6/site-packages/GPyOpt/core/task/objective.py in evaluate(self, x)
     48 
     49         if self.n_procs == 1:
---> 50             f_evals, cost_evals = self._eval_func(x)
     51         else:
     52             try:

~/miniconda3/envs/py36/lib/python3.6/site-packages/GPyOpt/core/task/objective.py in _eval_func(self, x)
     72         for i in range(x.shape[0]):
     73             st_time    = time.time()
---> 74             rlt = self.func(np.atleast_2d(x[i]))
     75             f_evals     = np.vstack([f_evals,rlt])
     76             cost_evals += [time.time()-st_time]

~/git/SMILES-X/SMILESX/main.py in create_mod(params)
    285                                                     shuffle = True,
    286                                                     initial_epoch = 0,
--> 287                                                     verbose = 0)
    288 
    289                 best_epoch = np.argmin(history.history['val_loss'])

~/miniconda3/envs/py36/lib/python3.6/site-packages/keras/legacy/interfaces.py in wrapper(*args, **kwargs)
     89                 warnings.warn('Update your `' + object_name + '` call to the ' +
     90                               'Keras 2 API: ' + signature, stacklevel=2)
---> 91             return func(*args, **kwargs)
     92         wrapper._original_function = func
     93         return wrapper

~/miniconda3/envs/py36/lib/python3.6/site-packages/keras/engine/training.py in fit_generator(self, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, validation_freq, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)
   1730             use_multiprocessing=use_multiprocessing,
   1731             shuffle=shuffle,
-> 1732             initial_epoch=initial_epoch)
   1733 
   1734     @interfaces.legacy_generator_methods_support

~/miniconda3/envs/py36/lib/python3.6/site-packages/keras/engine/training_generator.py in fit_generator(model, generator, steps_per_epoch, epochs, verbose, callbacks, validation_data, validation_steps, validation_freq, class_weight, max_queue_size, workers, use_multiprocessing, shuffle, initial_epoch)
     40 
     41     do_validation = bool(validation_data)
---> 42     model._make_train_function()
     43     if do_validation:
     44         model._make_test_function()

~/miniconda3/envs/py36/lib/python3.6/site-packages/keras/engine/training.py in _make_train_function(self)
    331                     updates=updates + metrics_updates,
    332                     name='train_function',
--> 333                     **self._function_kwargs)
    334 
    335     def _make_test_function(self):

~/miniconda3/envs/py36/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py in function(inputs, outputs, updates, **kwargs)
   3004 def function(inputs, outputs, updates=None, **kwargs):
   3005     if _is_tf_1():
-> 3006         v1_variable_initialization()
   3007     return tf_keras_backend.function(inputs, outputs,
   3008                                      updates=updates,

~/miniconda3/envs/py36/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py in v1_variable_initialization()
    418 
    419 def v1_variable_initialization():
--> 420     session = get_session()
    421     with session.graph.as_default():
    422         variables = tf.global_variables()

~/miniconda3/envs/py36/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py in get_session()
    383             '`get_session` is not available when '
    384             'TensorFlow is executing eagerly.')
--> 385     return tf_keras_backend.get_session()
    386 
    387 

~/miniconda3/envs/py36/lib/python3.6/site-packages/tensorflow/python/keras/backend.py in get_session(op_input_list)
    460   if not _MANUAL_VAR_INIT:
    461     with session.graph.as_default():
--> 462       _initialize_variables(session)
    463   return session
    464 

~/miniconda3/envs/py36/lib/python3.6/site-packages/tensorflow/python/keras/backend.py in _initialize_variables(session)
    877     # marked as initialized.
    878     is_initialized = session.run(
--> 879         [variables_module.is_variable_initialized(v) for v in candidate_vars])
    880     uninitialized_vars = []
    881     for flag, v in zip(is_initialized, candidate_vars):

~/miniconda3/envs/py36/lib/python3.6/site-packages/tensorflow/python/client/session.py in run(self, fetches, feed_dict, options, run_metadata)
    948     try:
    949       result = self._run(None, fetches, feed_dict, options_ptr,
--> 950                          run_metadata_ptr)
    951       if run_metadata:
    952         proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

~/miniconda3/envs/py36/lib/python3.6/site-packages/tensorflow/python/client/session.py in _run(self, handle, fetches, feed_dict, options, run_metadata)
   1171     if final_fetches or final_targets or (handle and feed_dict_tensor):
   1172       results = self._do_run(handle, final_targets, final_fetches,
-> 1173                              feed_dict_tensor, options, run_metadata)
   1174     else:
   1175       results = []

~/miniconda3/envs/py36/lib/python3.6/site-packages/tensorflow/python/client/session.py in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)
   1348     if handle is None:
   1349       return self._do_call(_run_fn, feeds, fetches, targets, options,
-> 1350                            run_metadata)
   1351     else:
   1352       return self._do_call(_prun_fn, handle, feeds, fetches)

~/miniconda3/envs/py36/lib/python3.6/site-packages/tensorflow/python/client/session.py in _do_call(self, fn, *args)
   1368           pass
   1369       message = error_interpolation.interpolate(message, self._graph)
-> 1370       raise type(e)(node_def, op, message)
   1371 
   1372   def _extend_graph(self):

InvalidArgumentError: No OpKernel was registered to support Op 'CudnnRNN' used by node bidirectional_1/CudnnRNN (defined at /home/elkhatim/miniconda3/envs/py36/lib/python3.6/site-packages/keras/layers/cudnn_recurrent.py:517) with these attrs: [rnn_mode="lstm", seed2=0, is_training=true, dropout=0, seed=87654321, T=DT_FLOAT, input_mode="linear_input", direction="unidirectional"]
Registered devices: [CPU, XLA_CPU]
Registered kernels:
  <no registered kernels>

	 [[bidirectional_1/CudnnRNN]]

Errors may have originated from an input operation.
Input Source operations connected to node bidirectional_1/CudnnRNN:
 bidirectional_1/ExpandDims_2 (defined at /home/elkhatim/miniconda3/envs/py36/lib/python3.6/site-packages/keras/layers/cudnn_recurrent.py:488)	
 bidirectional_1/transpose (defined at /home/elkhatim/miniconda3/envs/py36/lib/python3.6/site-packages/keras/layers/cudnn_recurrent.py:484)	
 bidirectional_1/ExpandDims_1 (defined at /home/elkhatim/miniconda3/envs/py36/lib/python3.6/site-packages/keras/layers/cudnn_recurrent.py:487)	
 bidirectional_1/concat (defined at /home/elkhatim/miniconda3/envs/py36/lib/python3.6/site-packages/keras/layers/cudnn_recurrent.py:60)

I would love trying to help with this but I am just fluent in PyTorch, and have not used keras too much. Any ideas how to solve this issue?

Best,