Pyinstaller with Tensorflow takes incorrect path for _checkpoint_ops.so file

Add the following to your spec file (finds tensorflow binaries and adds them to your .app in the main binary/file directory):

import os

tensorflow_location = '/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/site-packages/tensorflow'

tensorflow_binaries = []

for dir_name, sub_dir_list, fileList in os.walk(tensorflow_location): 
  for file in fileList:
    if file.endswith(".so"):
      full_file = dir_name + '/' + file
      print(full_file)
      tensorflow_binaries.append((full_file, '.'))

Make sure you also add the binaries to Analysis

a = Analysis(...,
             binaries=tensorflow_binaries,
             ...)

And make sure binaries are included in the .exe/.app builds. Here's a simple example:

pyz = PYZ(a.pure)
exe = EXE(pyz,
          a.scripts,
          a.binaries,
          a.zipfiles,
          a.datas,
          name='TestApp',
          debug=True,
          strip=None,
          upx=True,
          console=True, icon='Test.icns')
app = BUNDLE(exe,
    a.scripts,
    a.binaries,
    a.zipfiles,
    a.datas,
    name='Test.app',
    icon='Test.icns',
    bundle_identifier='com.Test.TestApp',
    info_plist={
        'NSHighResolutionCapable': 'True'
            },
    )

The last thing you need to do is modify the tensorflow load_library.py file to point to the relative path of the binaries during .app runtime.

The file is usually located here (relative to python install): site-packages/tensorflow/python/framework/load_library.py

You need to add the resource_path function and modify the load_op_library function to point the file_name to the relative path (and to ignore finding binaries by sub-folder).

def resource_path(relative_path):
    """Due to pyinstaller changing root dir of project filenames need to be processed in order to open properly

    Parameters
    ----------
    relative_path : str
        String containing filename path

    Returns
    -------
    path : str
        path relative to this file on local computer
    """
    import sys
    import os

    try:
        # PyInstaller creates a temp folder and stores path in _MEIPASS
        base_path = sys._MEIPASS
    except Exception:
        base_path = os.path.abspath(".")

    path = os.path.join(base_path, relative_path)
    return path

def load_op_library(library_filename):
  """Loads a TensorFlow plugin, containing custom ops and kernels.

  Pass "library_filename" to a platform-specific mechanism for dynamically
  loading a library. The rules for determining the exact location of the
  library are platform-specific and are not documented here. When the
  library is loaded, ops and kernels registered in the library via the
  `REGISTER_*` macros are made available in the TensorFlow process. Note
  that ops with the same name as an existing op are rejected and not
  registered with the process.

  Args:
    library_filename: Path to the plugin.
      Relative or absolute filesystem path to a dynamic library file.

  Returns:
    A python module containing the Python wrappers for Ops defined in
    the plugin.

  Raises:
    RuntimeError: when unable to load the library or get the python wrappers.
  """ 
  # REMOVE AFTER PYINSTALLER USE
  library_filename = resource_path(library_filename.split('/')[-1])

This will make tensorflow fetch the binaries by the relative path of your pyinstaller application (See: https://pythonhosted.org/PyInstaller/runtime-information.html#run-time-information).

Tensorflow will then fetch the binaries as packaged by the .spec file.

Remember to uncomment the library_filename = resource_path(library_filename.split('/')[-1]) line in load_library.py after Pyinstaller packaging (or else Tensorflow won't be able to find the binaries when running in your local Python env!)