Rasa-nlu failing to install with pip install rasa-nlu

Python Help
1 
pip install rasa-nlu
Collecting rasa-nlu
  Using cached rasa_nlu-0.15.1-py3-none-any.whl (147 kB)
Collecting cloudpickle~=0.6.1 (from rasa-nlu)
  Using cached cloudpickle-0.6.1-py2.py3-none-any.whl (14 kB)
Collecting gevent~=1.3 (from rasa-nlu)
  Using cached gevent-1.5.0.tar.gz (5.3 MB)
  Installing build dependencies ... done
  Getting requirements to build wheel ... error
  error: subprocess-exited-with-error

  × Getting requirements to build wheel did not run successfully.
  │ exit code: 1
  ╰─> [318 lines of output]

.
.
.
      Error compiling Cython file:
      ------------------------------------------------------------
      ...
              if events & ~(libev.EV__IOFDSET | libev.EV_READ | libev.EV_WRITE):
                  raise ValueError('illegal event mask: %r' % events)
              # All the vfd_functions are no-ops on POSIX
              cdef int vfd = libev.vfd_open(fd)
              libev.ev_io_init(&self._watcher, <void *>gevent_callback_io, vfd, events)
              self.__watcher = <libev.ev_watcher*>&self._watcher
                               ^
      ------------------------------------------------------------

      src\gevent\libev\corecext.pyx:1027:25: Cannot convert 'ev_watcher *' to Python object

      Error compiling Cython file:
      ------------------------------------------------------------
      ...
                  raise ValueError('illegal event mask: %r' % events)
              # All the vfd_functions are no-ops on POSIX
              cdef int vfd = libev.vfd_open(fd)
              libev.ev_io_init(&self._watcher, <void *>gevent_callback_io, vfd, events)
              self.__watcher = <libev.ev_watcher*>&self._watcher
              self.__ss = &io_ss
.
.
.
and many more
2

Seems to be a known issue (from a few years ago) caused by a particular gevent + Cython version:

You could try first updating Cython and gevent, then try to install rasa-nlu again.
Also, when running pip, it might be good to add --no-cache-dir since what’s in your cache may be too old.

Problem could also be that (the Pipy-distributed) rasa-nlu is just too old (last release was 4 years ago - which means in today’s ML/NLU landscape it seems antique), which is bound to cause problems with dependencies. So, you really need to make sure you are running all of this in a special virtual Python env. Plus - make sure you only use a Python version that rasa-nlu explicitly says is supported, just to prevent problems.

Alternatively - do not install with pip install from the Pypi, but install rasa-nlu directly from the current source code in the GitHub. This will give you an up-to-date version.

3

It still errors the same, but given the age of rasa, perhaps I’ll look at an alternative package for NLU intent based processing. Thanks

4

Or look at the ‘rasa’ (not rasa-nlu) package first. Not sure how to two relate, but ‘rasa’ is being actively maintained: GitHub - RasaHQ/rasa: 💬 Open source machine learning framework to automate text- and voice-based conversations: NLU, dialogue management, connect to Slack, Facebook, and more - Create chatbots and voice assistants
Having taken a quick look, it seems ‘rasa/nlu’ is just a submodule of ‘rasa’, so the 'rasa` package is probably what you want.
Don’t know how familiar you are with other packages in this area, but you may also like the HuggingFace APIs: https://huggingface.co

5

This sounds suspiciously similar to this previous discussion: Getting requirements to build wheel did not run successfully exit code: 1

6

Indeed. Cython 3.0.2 did fix several issues but Cython 3 may not be compatible with some older extension module code.
To prevent these kind of issues, imo it would be good to always release wheels with pre-compiled dynamic link modules (also an ongoing discussion).