`just train` fails with No module named 'pandas'

[shanxS]

Steps to reproduce:

1. Checkout git commit ba8fc6b
2. Clean up metaflow env: rm -rf .metaflow
3. run just train

Logs: https://gist.github.com/shanxS/57d1727a03ed2351b4f52192f034b22a

I was able to work around by explicitly adding pandas in common.py but that begs the question how is numpy import is working.

[svpino]

This error is so weird because it should find pandas (the library is referenced in pyproject.toml for uv to install it.)

I couldn't reproduce it using your steps.

When you said that you added pandas in common.py, what exactly do you mean? Did you import it or added it to the list of PACKAGES?

[leonelmorenoa]

I am going through the same issue, I tried idx and the devcontainer, see the screenshot below @svpino

[svpino]

@leonelmorenoa It looks like you are trying to run the common.py file in your screenshot?

Can you try just train instead?

[leonelmorenoa]

@svpino see now

[svpino]

Thanks for the screenshot.

Here is what you can do:

1. Ensure you have metaflow version >= 2.15.14 installed.
2. Run the following command: uv run pipelines/training.py --with retry run

Let me know if this works.

[leonelmorenoa]

@svpino Thanks, it worked. When I reach step 3.6 (transforming fold data), I see two buttons: "Open Transformer Implementation". If I click it, the common.py file opens. But when I press the "play" button (top-right), run the file, this error appears -see screenshot below

When I click "Run Unit Test", it shows 9 passed and 52 deselected. What do you think I should do?

This also occurs with the 3.7 step, training fold model: when I click 'Open Model Implementation', it opens the common.py file. Then, if I click 'Play' for that file (top-right), a pandas- error appears—even though the unit tests for the 3.7 model seem to run fine

[svpino]

@leonelmorenoa The common.py file is not runnable. It's a fail that contains the implementation of the transformers, but you can't run it directly. That's why you get an error when you click on the play button.

The unit tests are runnable. When you click on the blue button "Run unit tests", they will run and exercise the functions that implement the transformers. Another way to run that code is through the just train command, which will run the entire pipeline.

In summary, everything is working for you as expected.

[leonelmorenoa]

Thank you very much for your kindness and response @svpino .

I was able to progress without issues until part 6 of the extension. There, I clicked the Run Monitoring Pipeline button successfully, I ran "just monitor" without issues, but when I pressed Run Monitoring Card Server, it triggered the error shown in the screenshot. A

Afterwards, when running the unit test from that same section (part 6, the first one), the test fails and generates a lengthy error. The next section, which is 6.1, just traffic, also fails, it says “AttributeError: ‘NoneType’ object has no attribute ‘items’”.

`

[leonelmorenoa]

Any suggestions, Santiago? @svpino I deeply appreciate your comments