ImportError: cannot import name 'create_features'

[sekhar14]

Rasa version: 0.14.1

Rasa X version (if used & relevant):

Python version: 3.6

Operating system (windows, osx, ...): ubuntu 16.04

Issue:
create_features not found in bert extract_features.py

Error (including full traceback):

Traceback (most recent call last):
  File "/usr/lib/python3.5/runpy.py", line 184, in _run_module_as_main
    "__main__", mod_spec)
  File "/usr/lib/python3.5/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "/home/gpu21080ti/tensorflow1.3.3.5/tensorcode/rasa/rasa_nlu/train.py", line 184, in <module>
    num_threads=cmdline_args.num_threads)
  File "/home/gpu21080ti/tensorflow1.3.3.5/tensorcode/rasa/rasa_nlu/train.py", line 148, in do_train
    trainer = Trainer(cfg, component_builder)
  File "/home/gpu21080ti/tensorflow1.3.3.5/tensorcode/rasa/rasa_nlu/model.py", line 152, in __init__
    components.validate_requirements(cfg.component_names)
  File "/home/gpu21080ti/tensorflow1.3.3.5/tensorcode/rasa/rasa_nlu/components.py", line 49, in validate_requirements
    from rasa_nlu import registry
  File "/home/gpu21080ti/tensorflow1.3.3.5/tensorcode/rasa/rasa_nlu/registry.py", line 36, in <module>
    from rasa_nlu.featurizers.bert_featurizer import BertFeaturizer
  File "/home/gpu21080ti/tensorflow1.3.3.5/tensorcode/rasa/rasa_nlu/featurizers/bert_featurizer.py", line 15, in <module>
    from bert.extract_features import create_features, model_fn_builder
ImportError: cannot import name 'create_features'

Command or request that led to error:

 python -m rasa_nlu.train -c sample_configs/config_bert.yml --data testData.json

Content of configuration file (config.yml) (if relevant):

language: "en"

pipeline:
- name: "intent_featurizer_bert"
  model_dir: "./uncased_L-12_H-768_A-12"
- name: "intent_classifier_sklearn"

Content of domain file (domain.yml) (if relevant):

[DavidForlen]

Did you find a solution to this problem ?

[msamogh]

Thanks for raising this issue, @Ghostvv will get back to you about it soon.

[Ghostvv]

what problem?

[DavidForlen]

In order to replace intent_featurizer_spacy component by rasa_components.featurizers.bert_featurizer.BertFeaturizer in my pipeline, I took theses files from the bert_embeddings branch (https://github.com/RasaHQ/rasa/tree/bert_embeddings) :

• rasa_nlu/featurizers/bert_featurizer.py
• rasa_nlu/utils/bert_utils.py

I have installed the bert-tensorflow package on docker.
The packages that may help to understand why I'm facing this issue :

$ docker-compose exec medibot pip freeze | grep rasa
rasa-core==0.13.8
rasa-core-sdk==0.12.2
rasa-nlu==0.14.6

$ docker-compose exec medibot pip freeze | grep bert
bert-tensorflow==1.0.1

And now it seems that the rasa_nlu/featurizers/bert_featurizer.py file tries to import the create_features function that python (and I) can't found.

The error is almost the same as the one reported by @sekhar14 :

docker-compose exec medibot make -C ./medibot train-nlu
make: Entering directory '/usr/src/aliae/medibot'
python3 -m rasa_nlu.train -c nlu_config_fr_bert_feature.yml --data data/fr_init_data_set_train.json -o models --fixed_model_name nlu --project bert_feature_on_FR --verbose
/usr/local/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
Traceback (most recent call last):
  File "/usr/local/lib/python3.6/site-packages/rasa_nlu/registry.py", line 109, in get_component_class
    return utils.class_from_module_path(component_name)
  File "/usr/local/lib/python3.6/site-packages/rasa_nlu/utils/__init__.py", line 176, in class_from_module_path
    m = importlib.import_module(module_name)
  File "/usr/local/lib/python3.6/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 994, in _gcd_import
  File "<frozen importlib._bootstrap>", line 971, in _find_and_load
  File "<frozen importlib._bootstrap>", line 955, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 665, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 678, in exec_module
  File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
  File "/usr/src/aliae/medibot/rasa_components/featurizers/bert_featurizer.py", line 15, in <module>
    from bert.extract_features import create_features, model_fn_builder
ImportError: cannot import name 'create_features'

I can import model_fn_builder (I tried to put the import of this function first, to test it), meaning that it can find the module and his functions, but not create_features, who doesn't even appear in the code of https://github.com/google-research/bert/blob/master/extract_features.py.

The first part of my pipeline file :

language: "fr"

pipeline:
- name: "nlp_spacy"
- name: "tokenizer_spacy"
- name: "rasa_components.featurizers.bert_featurizer.BertFeaturizer"
  model_dir: "multi_cased_L-12_H-768_A-12/"

[Ghostvv]

you should use this fork: https://github.com/RasaHQ/bert

[DavidForlen]

Thanks a lot ! It runs correctly now (even if it takes time on CPUs, I will try on GPUs).

But I couldn't install the bert package from this branch neatly. How should I install it without having to take the bert directory from the branch and putting it directly in my working directory ?
I tried docker-compose exec medibot pip install git+https://github.com/RasaHQ/bert, and pip tells me Successfully built bert. But then rerunning the training shows me : ModuleNotFoundError: No module named 'bert'.