wrong boolean actions in arguments.py

[rubenalv]

Arguments.py seems taken from the dMaSIF code, which had the same issue. In Argsparse, setting type=bool and default=True does not give what you initially expect. Eg, you cannot turn off the search using --search False. This could have implications on how the model was trained too?

Example:

args.py

import argparse
parser = argparse.ArgumentParser(description="Network parameters")  
parser.add_argument(
    "--search_wrong",
    type=bool,
    default=True,
    help="Predict matching between two partners",
)  
parser.add_argument(
    "--search_wrongFalse",
    type=bool,
    default=False,
    help="Predict matching between two partners",
)
parser.add_argument('--search_right', action=argparse.BooleanOptionalAction)  

test.py

from args import parser

# Parse the arguments
args = parser.parse_args()
print('search wrong, set to True in args.py')
print(args.search_wrong)
print('search_right, set as boolean action in args.py')
print(args.search_right)

Run python test.py search wrong, set to True in args.py True search wrongFalse, set to False in args.py False search_right, set as boolean action in args.py None

Run python test.py --search_wrong False --search_wrongFalse True --search_right search wrong, set to True in args.py True search wrongFalse, set to False in args.py True search_right, set as boolean action in args.py True

Run python test.py --search_wrong False --search_wrongFalse True --no-search_right search wrong, set to True in args.py True search wrongFalse, set to False in args.py True search_right, set as boolean action in args.py False

The only flag that gives exactly what you think it would is --search_right vs --no-search_right. It also works if you set the flag as boolean with False as default, but that is not the case in the current Arguments.py.

[phdymz]

In fact, many contents in Argsparse will not affect the operation of the program, such as True/False of search or site, because we have already calculated and stored the point cloud offline for the sake of convenience, and there is no need to change True/False to decide which task to perform as in dmasif.

[phdymz]

To perform each downstream task, just go to the folder of each downstream task and follow the instructions in the readme. Use the default value for search or site.

[rubenalv]

Ah, thank you. Inference would be OK, but if the flags do not work as expected, would that affect re-training? Eg the original dMaSIF had an AUROC that I substantially improved when retraining with a larger protein pair database (https://github.com/BioinfoMachineLearning/DIPS-Plus).

[phdymz]

The parameters related to training may only be network parameters, and parameters such as site/search should not be related to training. It is worth noting that if you plan to use a larger dataset, you need to preprocess the dataset into a point cloud. It should be fine to process it in the same format as the point cloud I uploaded.