thank you very much for pushing your amazing PET tool to this repo.
I am currently trying to investigate, how the random seed works, during different iteration per pattern.
If I understand correctly, PET trains and evaluates for each pattern n=repetitions independent models (stored in the folders final/p0-iY). This is done, to calculate later std deviation for the results.
You are describing that in your paper, too:
"each model is trained three times using different seeds and average results are reported"
https://arxiv.org/pdf/2001.07676v3.pdf
This would happend in pet/modeling.py in line: 326 and 327
set_seed(seed)
for pattern_id in pattern_ids:
for iteration in range(repetitions):
Do I interpret this correctly?
I am currently struggeling to understand how the different seeds for the model initialization per iteration is handled. We are setting a single seed in line 324.
Are the models initialized with diferent weights per iteration? Are there other seeds set?
Hi @timoschick ,
thank you very much for pushing your amazing PET tool to this repo. I am currently trying to investigate, how the random seed works, during different iteration per pattern.
If I understand correctly, PET trains and evaluates for each pattern n=repetitions independent models (stored in the folders final/p0-iY). This is done, to calculate later std deviation for the results.
You are describing that in your paper, too: "each model is trained three times using different seeds and average results are reported" https://arxiv.org/pdf/2001.07676v3.pdf
This would happend in pet/modeling.py in line: 326 and 327
set_seed(seed) for pattern_id in pattern_ids: for iteration in range(repetitions):Do I interpret this correctly? I am currently struggeling to understand how the different seeds for the model initialization per iteration is handled. We are setting a single seed in line 324. Are the models initialized with diferent weights per iteration? Are there other seeds set?
Thank youfso much in advance for your help!