Issue installing
Closed this issue · 7 comments
Hi
Just tried a fresh install on a new VM (Ubuntu Desktop 16.04.6 64-bit), followed instructions but when I tried running tests_training.py I get the following error:
ModuleNotFoundError: No module named 'tensorflow.keras.layers.experimental.preprocessing'
...
ImportError: Keras requires Tensorflow 2.2 or higher.
Obviously a dependency issue...just not sure which has the wrong version... TF or Keras??
Any suggestions?
Thanks
Tim
Hello!
How are you installing aXeleRate? From pip install aXeleRate
or pip install git+https://github.com/AIWintermuteAI/aXeleRate
?
Either way it is really strange... I cannot find any code containing tensorflow.keras.layers.experimental.preprocessing --- and I don't remember adding it, because I am yet to update aXeleRate to tf.keras.
Hi
I did the following:
pip install git+https://github.com/AIWintermuteAI/aXeleRate
Followed by:
git clone https://github.com/AIWintermuteAI/aXeleRate.git
Then
python tests_training.com
Yeah it had me confused too as it is a new, clean environment.
Cheers
Tim
Just checked the versions that the pip install installed:
- keras: 2.4.2
- tensorflow: 1.15.3
I noticed an error when I try to update keras (InvalidArchiveError("Error with archive /home/.... Message from libarchive was: Failed to crate dir 'lib/python3.7/site-packages/tensorflow/python/profiler" So I deleted that conda env and created a new one.
This time it used keras 2.4.3 and TF 1.15.3
Same error when trying to run tests_training.com.
Then I tried conda install "keras<=2.3.2"
- (which seemed to do nothing more than update tensorflow-base-2.2.0) but when I do conda list
I still see keras: 2.4.3 (very strange)
I tried running tests_training.py again.. and it got a bit further, got error:
python tests_training.py
Using TensorFlow backend.
Traceback (most recent call last):
File "tests_training.py", line 2, in <module>
from axelerate import setup_training, setup_inference
File "/home/tim/Dev/ML/aXeleRate/axelerate/__init__.py", line 1, in <module>
from .train import setup_training
File "/home/tim/Dev/ML/aXeleRate/axelerate/train.py", line 19, in <module>
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.5)
AttributeError: module 'tensorflow' has no attribute 'GPUOptions'
Any suggestions?
Any idea what is different between your environment and mine? I specifically created this new environment so I could use the same as you.
Thanks again
Tim.
The issue was in Keras package update. I was under the impression that 2.3.1 is going to be last standalone keras version. But turned out that I was wrong and 2.4.0 was released a few days ago, which is only compatible with tf 2.0.
I changed the requirements for keras from keras >= 2.3.1 to keras == 2.3.1 in this commit
947eef3
Please try creating a fresh environment and installing again - btw if you use GPU it is better to install tensorflow-gpu package from conda, it will automatically install cuda/cudnn as well if you haven't done that already.
Anyways, check the latest commit and see if it resolves the problem :)
Thanks for this.
Just recreated my conda environment. keras now with 2.3.1 and TF 1.15.3.
When running tests_training I get an error regarding the target folder (/home/ubuntu/...etc) so need to modify hard coded paths in tests_training.py then it works!
Many thanks for your work and assistance with this - much appreciated!
Tim
From the config dictionary inside tests_training.py
"saved_folder": "/home/ubuntu/space safety/classifier",
Line 41 in 947eef3
Line 78 in 947eef3
Line 110 in 947eef3
I'll correct that in the next release, thank you for noticing this nonsense :)
Just correct them to something like "results/name-for-project-folder" for now.
My pleasure!