Elias Joa

An issue now is that we get large coefficients for the variables that often are added that are similar, while all the unique variables (for example per player) always has a coefficient of 1. This l...

I see. Well thanks for tipsing me about it! Fixed all my memory problems, even allowing me to parallelize it and finish the rating calculations twenty times faster.

Is there any documentation on ._normalize()? Cannot find it. 

I ended up using ._normalize(). Will hopefully solve the issues. If not, the solution will be to keep track of all coefficients and creating the LinExpression only right before solving the model. 

Oh wow i was not aware, so I can then use quicksum(List of linexressions)._normalize(). I made this function instead, which aggregates coefficients: def quicksum_acc(lin_exprs):    var_coeff_map =...

Great, thanks a lot! Furthermore, maybe there is a way to make the function actually able to input a list of LinExpressions, returning one LinExpression with coefficients aggregated. Like quicksum,...

Thanks! Any idea on the time complexity of this function? Experiencing a massive increase in runtime, but indeed no memory problems anymore. 

It does indeed happen during construction of these expressions. To give some perspective there are 62000 games, and around 24 players per game, and all of these have a specific base rating decision...