1.13.0rc1 requires numpy >=1.16.0 instead of >=1.13.3 stated in setup.py in r1.13 branch???

[henrysky]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Windows 10 x64
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device: N/A
• TensorFlow installed from (source or binary): Binary
• TensorFlow version: 1.13.0rc1
• Python version: 3.7.1
• Installed using virtualenv? pip? conda?: pip
• Bazel version (if compiling from source): N/A
• GCC/Compiler version (if compiling from source): N/A
• CUDA/cuDNN version: 10 / 7.4.2
• GPU model and memory: MX150 2GB

Describe the problem
Tensorflow 1.13.0rc1 said the numpy requires >=1.13.3 (I had np 1.15.4) in setup.py but it only works with >=1.16.0. Either the setup.py is wrong or there is something wrong while building the wheel??

Provide the exact sequence of commands / steps that you executed before running into the problem

import tensorflow

Any other info / logs

PyBfloat16_Type.tp_base != nullptr

[jvishnuvardhan]

Sorry for the delay in resolving the issue @henrysky. Could you check whether this is an issue with the latest TF? Thanks!

[jvishnuvardhan]

In some cases, I think there are compatibility issues between modules when you install TF using conda. Could you uninstall python and tensorflow and reinstall following the instructions here. Please let me know how it progresses. Thanks!

[henrysky]

I follow the instruction to use official python builds instead of conda in a VM.

For python 3.6 and 3.7 I have tested, the error still the same when using numpy 1.15, but gone when using numpy 1.16

[jvishnuvardhan]

@henrysky Please check required packages here for TF1.13.1. In future, double check version of required packages before you install TF through Anaconda. Thanks!

REQUIRED_PACKAGES = [
'absl-py >= 0.7.0',
'astor >= 0.6.0',
'gast >= 0.2.0',
'google_pasta >= 0.1.2',
'keras_applications >= 1.0.6',
'keras_preprocessing >= 1.0.5',
'numpy >= 1.14.5, < 2.0',
'six >= 1.10.0',
'protobuf >= 3.6.1',
'tensorboard >= 1.13.0, < 1.14.0',
'tensorflow_estimator >= 1.13.0rc0, < 1.14.0rc0',
'termcolor >= 1.1.0',
]

[henrysky]

This is a clean windows 10 in VM. Using python instead of anaconda python

numpy 1.15.4 which tensorflow failed

numpy 1.16.2 which tensorflow okay

[henrysky]

@jvishnuvardhan I check with the latest tensorflow 2.0.0a0 same issue is there too

[combacsa]

In [1]: import numpy

In [2]: numpy.version
Out[2]: <module 'numpy.version' from '/Users/khbyun/anaconda3/envs/w4/lib/python3.6/site-packages/numpy/version.py'>

In [3]: numpy.version.version
Out[3]: '1.15.1'

In [4]: import tensorflow
---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
ModuleNotFoundError: No module named 'numpy.core._multiarray_umath'
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
ImportError: numpy.core.multiarray failed to import

The above exception was the direct cause of the following exception:

SystemError                               Traceback (most recent call last)
~/anaconda3/envs/w4/lib/python3.6/importlib/_bootstrap.py in _find_and_load(name, import_)

SystemError: <class '_frozen_importlib._ModuleLockManager'> returned a result with an error set
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
ImportError: numpy.core._multiarray_umath failed to import
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
ImportError: numpy.core.umath failed to import
2019-03-19 16:46:12.380357: F tensorflow/python/lib/core/bfloat16.cc:675] Check failed: PyBfloat16_Type.tp_base != nullptr
Abort trap: 6

According to numpy issue 11871, tensorflow >= 1.13 should depends on numpy >= 1.16 ...

numpy/numpy#11871

[combacsa]

It happens on macOS, too. Not only on windows.

[facaiy]

@av8ramit Hi, Amit. Do you know the problem?

[ctmckee]

The "ModuleNotFoundError: No module name 'numpy.core._multiarray_umath' occurs when trying to use tensorflowjs_converter

[Yilong94]

I can confirm that the error is still occuring. numpy==1.15.4 is not compatible with tensorflow==1.13.1, despite tensorflow v1.13.1 requiring only numpy>=1.13.3

[mihaimaruseac]

Going to fix this in 1.13 (with upcoming patch release), 1.14 (with upcoming patch release) and master (for new releases)

[mihaimaruseac]

Also notes that there is a workaround where you install numpy before tnesorflow or with the constraint at the same time as tensorflow: pip install numpy && pip install tensorflow or pip install numpy >= 1.16 tensorflow == 1.13