如何在新的 venv 中升级到 Tensorflow2？

Question

Trying to upgrade from Tensorflow1.13 and tensorflow-gpu1.13 to Tensorflow 2. In Anaconda, have cloned previous venv (venvTF1) as venvTF2... Activate venvTF2 in command window... Then have upgraded pip, and then

pip --upgrade tensorflow

This successfully(?) installs tensorflow 2. If I do a

pip list

, it shows tensorflow 2 the list, not tensorflow 1.13 for venvTF2

However, if I run python in cmd window:

python
import tensorflow as tf
tf.test.is_built_with_cuda
True
tf.test.gpu_device_name
my gpu info prints here

But...

tf.__version__
'1.13.1'

So how to get new venvTF2 to work with Tensorflow2? (running Python 3.6.5)

Must also add that now tensorflow1.13 in original venv is corrupted...cannot import, DLL load failed...

Answer 1

I think this is because you have cloned the env. Because of the cloning, all the dependencied are the same. So it is same as running the command in older venv. If you have cloned, all the dependencies have come along with it.

What you need to do is to

Case 1 (Never do it): Either remove one by one the old dependencies.

Or Case 2: Best case:

conda create -n your_venv_name tensorflow=2.2

Suggestion : You should try install those packages which you are most dependent on. For example, if you want to have the tf 2.0 , install it first and then when you install python, it'll automatically install the version of python most suitable acording to the version of tf installed. Same with other packages.

Try to install as many packages with conda just like googling package_name conda install . You'll find something like conda install conda-forge your_package or conda install anaconda your_package_name