Questions about reproducing the process of the experiment

[Nora-jsu]

Hello, I am reproducing your experiment recently, but I can't reproduce your effect in the report, the highest F-score is only 9, the following is my parameter configuration, please ask what needs to be changed in the parameter configuration of the large model, I look forward to your reply. Here is my parameter interface:

[axeber01]

Can you check if the model weights are loaded correctly? Try going to this line and use 'strict=True' instead

ngcc-seld/train_seldnet.py

Line 734 in 64aa259

model.load_state_dict(state_dict, strict=False)

Then set the learning rate to 0 (so that the weights are not changed after loading them) and the number of epochs to 1 (so that the evaluation is done faster). Then you should get the expected results.