System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): ubuntu 18.04
• TensorFlow version and how it was installed (source or binary): tensorflow-cpu 2.1.0
• TensorFlow-Addons version and how it was installed (source or binary): 0.7.0
• Python version: 3.7
• Is GPU used? (yes/no): no

Describe the bug

In an environement where tensorflow-cpu is already installed, installing tensorflow-addons should not download and install tensorflow (now being the gpu version by default).

Code to reproduce the issue

FROM python:3.7

RUN pip install tensorflow-cpu
RUN pip install tensorflow-addons
RUN pip freeze | grep tensorflow

result:

tensorflow (gpu) is still being installed (400MB download) and the installation of tensorflow is now duplicated:

tensorflow==2.1.0
tensorflow-addons==0.7.0
tensorflow-cpu==2.1.0
tensorflow-estimator==2.1.0

A solution would be to remove the tensorflow dependency in the setup.py and let people install the version of tensorflow they want. It's not a perfect solution but this is the best one we found, at least for keras-tuner and autokeras. (keras-team/keras-tuner#211 and keras-team/autokeras#860)

The problem is similar to libraries that depend on Pillow and users have already pillow-SIMD installed.