inconsistent program execution

[codercahol]

I get inconsistent results when I execute the code in database_creation:

The following is all being run on the Carnegie cloud in a jupyter notebook

I execute the following to initialize the jupyter notebook:

import sys
chlamy_impi_module_path = "{path-to-local-repo}"
if chlamy_impi_module_path not in sys.path:
    sys.path.append(chlamy_impi_module_path)

the functions from the module are exported as follows:

from chlamy_impi.well_segmentation_preprocessing.main import (
    main as well_segmentation_preprocessing,
)
from chlamy_impi.database_creation.main import main as database_creation

# for dev
from chlamy_impi.database_creation import utils as db_utils
from chlamy_impi.database_creation import error_correction as db_error_correction
from chlamy_impi.database_creation import main as db_main
from chlamy_impi.lib import mask_functions
from chlamy_impi.lib import npq_functions
from chlamy_impi.lib import y2_functions
from chlamy_impi.lib import fv_fm_functions
import chlamy_impi.paths as paths

When I run the database_creation/main.py:main() by executing cp.database_creation() or cp.db_main.main() I get an error in database_creation/main.py:merge_plate_and_well_info_dfs() saying that the i and j columns of the dataframe are NaNs.

When I run all the sub-functions of main() in the notebook it runs to completion with no error (i.e. I define and run local_main() as below):

def local_main():
    plat_info = cp.db_main.construct_plate_info_df()
    well_info = cp.db_main.construct_well_info_df()
    exptl_df = cp.db_main.merge_plate_and_well_info_dfs(plat_info, well_info)

    mut_df = cp.db_main.construct_mutations_dataframe()
    ident_df = cp.db_main.construct_identity_dataframe(mut_df)
    total = cp.db_main.merge_identity_and_experimental_dfs(exptl_df, ident_df)
    return total

[codercahol]

Further info: the wells that cause the issue are consistent, with the following indices: [4608, 8065, 16130, 18819, 22276, 34181] in the erroring code, the measurement times are all null, num_frames is 2x that of the working code, and the threshold values do not match

[codercahol]

[codercahol]

[murraycutforth]

Hey, I've just had a quick read. If I've understood correctly this is a code issue, and not an issue with the underlying data?

It's not immediately apparent to me why there should be a difference in execution depending on which functions are imported and run in the scope of the notebook. If it works correctly when the sub-functions are manually imported then that would suggest to me that there may be a name collision in the scope of the notebook? But I would expect that to cause the code to catastrophically fail rather than this subtle difference in a few rows.

In general it could be dangerous to import these main scripts as modules, because everything not wrapped in a if __name__ == '__main__': clause will be executed at the time of import. Not sure if that could be the cause..?

Have you checked if all the other rows in exptl_df are identical in both methods of running it? Also, just running the main.py script directly from the terminal is how I used the code when I was working on it, is it possible to try that method on the cluster?

Sorry I don't have enough time this week to try running anything myself!