What is it about robot collected data that makes researchers so anal? Like, dude, it's not even personal data. It's literally robot's joint motor recordings. It's not nuclear data, so why the fuck do you protect it like your life and your country depend on it?

I hope you get fisted by that data every night and how it will end up in oblivion sooner because you didn't publish it. You asshole.

We actually published most reactor operational data to the NRC, who in turn published it as public datasets

https://www.nrc.gov/data/#datasets

@SortOfTested lol. Big oof.
So now openAI gym acts more protective than nuclear scientists... Perfect.

@NoMad
Yep. The only thing strictly controlled was access, facility data, NRC regulated materials controls, schedules, maintenance routines, etc. A savvy analyst could derive most of that with a fair degree of accuracy from the datasets though.

@SortOfTested at the end of the day, that dataset is everything. Like, I see people who work with learning publishing their models, which is good if it was applicable, but they don't publish their sample. So I can't run anything else on the same data, neither can I confirm that what they've found is actually accurate. Like, half of similar papers to my work can't be referenced because something, either code, the main data, or the model is missing. I can't do comparison, I can't do any benchmarking, I can't even test my approach on the same data. Total waste of their data gathering efforts tbh.

@NoMad
I read that as "reproducibility is a problem in that machine learning industry, for many reasons."

@SortOfTested good deduction. My point was that I'm jealous of the nuclear researchers 😜

@NoMad
Me too. I was just a lowly engineer making the machines go bwaaaaaah and produce electrons ;)

Im not closely in this field, but I dabble and ye, I think datasets that don't contain any personal information should be released.

We're at the point where more data gives more people the chance to try something cool with it