[BUG] Chapter 12 notebook treats probability of survival as probability of death

[gwtaylor]

In chapter 11's default implementation of choose_dead we see that a random 10% of agents die in every round. So 0.1 is the probability of death:

def choose_dead(self, ps):
        """Choose which agents die in the next timestep.
        ps: probability of survival for each agent
        returns: indices of the chosen ones
        """
        n = len(self.agents)
        is_dead = np.random.random(n) < 0.1
        index_dead = np.nonzero(is_dead)[0]
        return index_dead

When this is overridden to use differential survival, we flip < to > in the is_dead line to interpret ps as probability of survival:

class SimWithDiffSurvival(Simulation):
    def choose_dead(self, ps):
        """Choose which agents die in the next timestep.
        ps: probability of survival for each agent
        returns: indices of the chosen ones
        """
        n = len(self.agents)
        is_dead = np.random.random(n) > ps
        index_dead = np.nonzero(is_dead)[0]
        return index_dead

However, this doesn't happen in chapter 12 and I believe it's a bug that affects the conclusions made in the chapter.

Note that chapter 12 uses the same default implementation of choose_dead as chapter 11. It interprets 0.1 as probability of death. But when introducing differential survival, choose_dead is overridden as this:

# class PDSimulation(Simulation):

    def choose_dead(self, fits):
        """Choose which agents die in the next timestep.
        fits: fitness of each agent
        returns: indices of the chosen ones
        """
        ps = prob_survive(fits)
        n = len(self.agents)
        is_dead = np.random.random(n) < ps
        index_dead = np.nonzero(is_dead)[0]
        return index_dead

Note that < isn't flipped this time. So imagine that all the agents had infinite fitness and therefore their probability of survival was 1.0 . The line is_dead = np.random.random(n) < ps would return all True and kill them all off. This doesn't make sense.

Since choose_dead takes a parameter ps (stands for probability of survival), I think that we should flip < to > and use 0.9 in the default implementation so that the semantics are consistent. Then we never need to do a sign flip and there won't be confusion down the road.

However, this still leaves open the experiments & conclusions made in chapter 12, which, I believe, come with this bug in place.

[Captain-Blackstone]

Stumbled upon the same bug. It's not fixed yet.