Handle temperature==0

[alejandromarcu]

I noticed that when playing it would often pick a sub-optimal choice (e.g. probs were 0.4, 0.3 and 0.3 and it would pick 0.3). That was because it's not correclty handling the temperature. When playing, temperature is 0 which means choose the best option (randomly between them if there's more than 1 with the max value)

[adamantivm]

LGTM

[jonbinney]

Is there any easy way to set temperature=_some_number_ for training, but do the intermittent mini-tournaments with temperature=0?

[alejandromarcu]

Is there any easy way to set temperature=_some_number_ for training, but do the intermittent mini-tournaments with temperature=0?

Good catch! for cases like this we introduced training_only_params. I think if we put temperature in that set, it will be removed when playing the tournament in the training and the default of 0 will be used. I'll try it out.