pip installation fails on modern images (Colab, Kaggle)

[maciejskorski]

Hi, I see the pip installation failing under dependencies of some important images: Colab and Kaggle. In particular, the current tutorial ISR_Traininig_Tutorial.ipynb is not working.
While workarounds exist (I managed to install from git), I believe many users would benefit from a solid fix.

The following error trace is under the recent (April 2023) Kaggle docker

(.venv) root@3a72fcea2525:/home/jupyter/utils# pip install ISR
Collecting ISR
  Using cached ISR-2.2.0-py3-none-any.whl (33 kB)
Requirement already satisfied: imageio in /opt/conda/lib/python3.10/site-packages (from ISR) (2.27.0)
Requirement already satisfied: numpy in /home/jupyter/.venv/lib/python3.10/site-packages (from ISR) (1.22.4)
INFO: pip is looking at multiple versions of isr to determine which version is compatible with other requirements. This could take a while.
  Using cached ISR-2.1.1-py3-none-any.whl (32 kB)
  Using cached ISR-2.1-py3-none-any.whl (32 kB)
Collecting Keras==2.2.4 (from ISR)
  Using cached Keras-2.2.4-py2.py3-none-any.whl (312 kB)
Collecting numpy==1.16.2 (from ISR)
  Using cached numpy-1.16.2.zip (5.1 MB)
  Preparing metadata (setup.py) ... done
Collecting ISR
  Using cached ISR-2.0.5-py3-none-any.whl (35 kB)
  Using cached ISR-2.0.2-py3-none-any.whl (29 kB)
  Using cached ISR-2.0.1-py3-none-any.whl (33 kB)
  Using cached ISR-2.0.0-py3-none-any.whl (32 kB)
  Using cached ISR-1.9.3-py3-none-any.whl (31 kB)
INFO: pip is looking at multiple versions of isr to determine which version is compatible with other requirements. This could take a while.
  Using cached ISR-1.9.2-py3-none-any.whl (31 kB)
  Using cached ISR-1.9.1-py3-none-any.whl (28 kB)
  Using cached ISR-1.9.0-py3-none-any.whl (27 kB)
ERROR: Cannot install isr==1.9.0, isr==1.9.1, isr==1.9.2, isr==1.9.3, isr==2.0.0, isr==2.0.1, isr==2.0.2, isr==2.0.5, isr==2.1, isr==2.1.1 and isr==2.2.0 because these package versions have conflicting dependencies.

The conflict is caused by:
    isr 2.2.0 depends on tensorflow==2.0.0
    isr 2.1.1 depends on tensorflow==2.0.0
    isr 2.1 depends on tensorflow==1.13.1
    isr 2.0.5 depends on tensorflow==1.13.1
    isr 2.0.2 depends on tensorflow==1.13.1
    isr 2.0.1 depends on tensorflow==1.13.1
    isr 2.0.0 depends on tensorflow==1.13.1
    isr 1.9.3 depends on tensorflow==1.13.1
    isr 1.9.2 depends on tensorflow==1.13.1
    isr 1.9.1 depends on tensorflow==1.13.1
    isr 1.9.0 depends on tensorflow==1.13.1

To fix this you could try to:
1. loosen the range of package versions you've specified
2. remove package versions to allow pip attempt to solve the dependency conflict

ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/topics/dependency-resolution/#dealing-with-dependency-conflicts

[incubiq]

Hi - hitting the same problem here... I can make it work on local machine (windows), but it fails the install in docker (ubuntu 20.04)
1/ Could you share exactly how you "managed to install from git" and bypass the issue?
2/ which python version do you use?
3/ which tensorflow version do you use?
Thanks

[psigelo]

I have the same problem, and trying to avoid problems I tried on a fresh new environment with the exact same results so it does not seems to be by the tensorflow version.

My python version is 3.10.12

[constantinpape]

The ISR pip packages only support tensorflow up to 2.0, which is only supported for up to python 3.7.
So if you want to install it via pip you need to use a python 3.7 environment.
(It would be nice to get a more up-to-date pip package!)