Skip to content
This repository has been archived by the owner on Mar 21, 2024. It is now read-only.

Can't find conda_merge #664

Closed
Bayesianworld opened this issue Feb 17, 2022 · 13 comments
Closed

Can't find conda_merge #664

Bayesianworld opened this issue Feb 17, 2022 · 13 comments

Comments

@Bayesianworld
Copy link

Bayesianworld commented Feb 17, 2022 •
edited by fepegar
Loading

Just installing the IE-DeepLearning software. Set up Anaconda and IE software. When I try running the HelloWorld model, I get the following error:

~username/mydir:~$ conda activate InnerEye
(InnerEye) ~username/mydir~$ export PYTHONPATH=`pwd`
(InnerEye) ~username/mydir:~$ python InnerEye/ML/runner.py --model=HelloWorld
Starting InnerEye runner at InnerEye/ML/runner.py
Traceback (most recent call last):
  File "InnerEye/ML/runner.py", line 30, in <module>
    from health_azure import AzureRunInfo, submit_to_azure_if_needed
  File "/users/markp/Projects/Microsoft/InnerEye2022/InnerEye-DeepLearning/hi-ml/hi-ml-azure/src/health_azure/__init__.py", line 6, in <module>
    from health_azure.datasets import DatasetConfig
  File "/users/markp/Projects/Microsoft/InnerEye2022/InnerEye-DeepLearning/hi-ml/hi-ml-azure/src/health_azure/datasets.py", line 15, in <module>
    from health_azure.utils import PathOrString, get_workspace
  File "/users/markp/Projects/Microsoft/InnerEye2022/InnerEye-DeepLearning/hi-ml/hi-ml-azure/src/health_azure/utils.py", line 20, in <module>
    import conda_merge
ModuleNotFoundError: No module named 'conda_merge'

AB#5020
@ant0nsc
Copy link
Contributor

ant0nsc commented Feb 21, 2022

Hi @Bayesianworld , that's a little puzzling. The environment definition does contain conda-merge, link
To double-check your environment is set up correctly, can you run

conda activate InnerEye
pip freeze | grep conda

It should output a line for conda-merge.

@Bayesianworld
Copy link
Author

Tried your suggestion. Returned immediately with no output. Here is my session:

markp@lambda:$ ls
InnerEye-DeepLearning/
markp@lambda:$ cd InnerEye-DeepLearning/
markp@lambda:$ conda activate InnerEye
(InnerEye) markp@lambda:$ export PYTHONPATH=pwd
(InnerEye) markp@lambda:$ pip freeze | grep conda
(InnerEye) markp@lambda:$ python InnerEye/ML/runner.py --model=HelloWorld
Starting InnerEye runner at InnerEye/ML/runner.py
Traceback (most recent call last):
File "InnerEye/ML/runner.py", line 30, in
from health_azure import AzureRunInfo, submit_to_azure_if_needed
File "/users/markp/Projects/Microsoft/InnerEye2022/InnerEye-DeepLearning/hi-ml/hi-ml-azure/src/health_azure/init.py", line 6, in
from health_azure.datasets import DatasetConfig
File "/users/markp/Projects/Microsoft/InnerEye2022/InnerEye-DeepLearning/hi-ml/hi-ml-azure/src/health_azure/datasets.py", line 15, in
from health_azure.utils import PathOrString, get_workspace
File "/users/markp/Projects/Microsoft/InnerEye2022/InnerEye-DeepLearning/hi-ml/hi-ml-azure/src/health_azure/utils.py", line 20, in
import conda_merge
ModuleNotFoundError: No module named 'conda_merge'

@ant0nsc
Copy link
Contributor

ant0nsc commented Feb 23, 2022

Odd. Can you please post the output of running pip freeze in your conda environment?
Can you please also check that you are on the latest version of the InnerEye code, and then update your conda environment via conda env update --file environment.yml?

@Bayesianworld
Copy link
Author

Bayesianworld commented Feb 25, 2022 via email

@ant0nsc
Copy link
Contributor

ant0nsc commented Mar 1, 2022

Hi Mark - sorry, I may have been a bit too brief with my suggestion. I'm interested in the output of pip freeze inside the InnerEye environment, without any filters, so you would need to run:

conda activate InnerEye
pip freeze

It should not matter which directory you run that in.

@Bayesianworld
Copy link
Author

Bayesianworld commented Mar 1, 2022 via email

@ant0nsc
Copy link
Contributor

ant0nsc commented Mar 2, 2022

That's all rather odd. I don't get why you don't have conda_merge in your environment, even though you say you built it with the latest environment.yml file. In addition, there are two odd things in your environment: numpy and pillow have these odd "@ file" versions. I'd suggest manually fixing all three by doing

conda activate InnerEye
pip uninstall numpy Pillow
pip install conda-merge numpy pillow

@Bayesianworld
Copy link
Author

Bayesianworld commented Mar 3, 2022 via email

@Bayesianworld
Copy link
Author

Bayesianworld commented Mar 3, 2022 via email

@ant0nsc
Copy link
Contributor

ant0nsc commented Mar 3, 2022

Hi @Bayesianworld , Something is really really broken with your conda environment. I checked in the output of pip freeze that you posted the other day - it contains azureml-sdk version 1.12.0, but it should be 1.36.0. Hence the problem with DockerConfig, which did not exist in 1.12. It would probably be best if you completely remove that conda environment, and re-build it using the environment definition from the latest InnerEye code in main branch.

Also, just FYI, it seems you are replying to the github comments via email - unfortunately, attachments don't come through there, apart from .txt files

@Bayesianworld
Copy link
Author

Bayesianworld commented Mar 4, 2022 via email

@ant0nsc
Copy link
Contributor

ant0nsc commented Mar 7, 2022

Unfortunately, I’m unable to reproduce the environment creation issue that you are facing. I just created a fresh enlistment, created a brand-new environment, and the HelloWorld models runs just fine.

Are you in a locked-down environment that uses pypi package mirrors or something alike? I have no other hypothesis what could have caused the problems with the AML packages – your environment has version 1.12.0, but we need 1.32.0.

@ant0nsc
Copy link
Contributor

ant0nsc commented May 16, 2022

Closing for lack of more information.

@ant0nsc ant0nsc closed this as not planned Won't fix, can't repro, duplicate, stale May 16, 2022
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@ant0nsc @Bayesianworld