Can anyone explain briefly how to submit our model in the competition? I'm finding the instructions confusing. :/ :(
Posted by: Nikolina @ Nov. 9, 2022, 7:53 p.m.Which part of the provided instructions is confusing? Please, point it out so that we can eventually improve it.
In any case, we provide an example of how to build a submission file in the 'starting_kit', which can be downloaded from the 'Participate/Files' tab in the competition. You have simply to provide in a zip file both the implementation of the class 'model', with an '__init__' and a 'predict' function. In the '__init__' you can load the weights of your model, while in the predict you compute the forward step given the input tensor 'X' of shape [None, H, W, 3].
Posted by: an2dl.competitions @ Nov. 9, 2022, 9:04 p.m.We try to submit our model but we get this error:
"
Traceback (most recent call last):
File "/multiverse/storage/lattari/Prj/postdoc/Courses/AN2DL_2022/Competition1_running_dir/worker_gpu0_dir/tmp/codalab/tmpnphcfp/run/program/score.py", line 278, in <module>
raise Exception('{} F1 is NaN'.format(c))
Exception: Species3 F1 is NaN
"
Our model works giving no error when we execute it in google colab. Only when we try to submit it, we get this error not allowing us to move further.
Most probably it is due to the predict function. We continue not understanding what we have to add there to make it work.
Thank you!
The log message explains that the Species3 class would lead to NaN on the test set. This happens because your prediction on the test probably has a TP number for the specific class equal to 0. This implies that if FP or FN were equal to zero, you would respectively obtain precision or recall equal to 0/0, and ditto for the f1 score. We preferred to make such predictions null by error to save you as many submissions as possible. In other words, errors of this type do not make you waste daily submissions.
Posted by: an2dl.competitions @ Nov. 9, 2022, 10:22 p.m.