search

OLA2022-Group-12 / OLA2022-Project

Project for the Online Learning Applications class at Politecnico di Milano a.a. 2021-2022
2 stars 0 forks source link

Environment #12

Closed raul-singh closed 2 years ago

raul-singh commented 2 years ago

Made a small change in the environment. I didn't quite like the sigmoidal functions as alpha functions, because a sigmoid is tough to tune to have the shape we would like, and even with that, its shape it's not correct. It starts flat, then it gets steep and then it's flat again. We don't want flatness in the beginning. I looked a the course material and they simply implement this kind of function with a negative exponential, which in my mind makes more sense! I did this change (although I probably screwed up something with git, in this case I summon thee, @barskern ) so now the alpha functions have an initial area where they are 0 (which is something we want) and then they increase almost linearly and start to flatten as they get closer to the upper bound. The function interface is still the same, I just edited also the example_environment with values that make more sense with this kind of function.

raul-singh commented 2 years ago

I just modified a couple of lines, but I clearly I messed up something with the merge. I think I shouldn't have used this branch? If I made some fatal mistakes, I'm sorry

barskern commented 2 years ago

Made a small change in the environment. I didn't quite like the sigmoidal functions as alpha functions, because a sigmoid is tough to tune to have the shape we would like, and even with that, its shape it's not correct. It starts flat, then it gets steep and then it's flat again. We don't want flatness in the beginning. I looked a the course material and they simply implement this kind of function with a negative exponential, which in my mind makes more sense! I did this change (although I probably screwed up something with git, in this case I summon thee, @barskern ) so now the alpha functions have an initial area where they are 0 (which is something we want) and then they increase almost linearly and start to flatten as they get closer to the upper bound. The function interface is still the same, I just edited also the example_environment with values that make more sense with this kind of function.

I think this is a great change! The sigmoidal function, though what seemed to be a good choice, was behaving a bit strangly. And also I think the negative exponential does seem like a more suiting function. Awesome!

I am hearby summoned hahahah. I think the right way to go would be to create a new branch of master before making these changes, instead of continuing of an old branch. I will fix it! :)

raul-singh commented 2 years ago

Thank you for your help! Another thing I'm not understanding is why the linting failed. I did not change much, did I do something wrong?

barskern commented 2 years ago

Thank you for your help! Another thing I'm not understanding is why the linting failed. I did not change much, did I do something wrong?

It was me that made a mistake when rebasing, I forgot to change user_classes to classes_parameters 😓