[generic_imitation] different results depending on threshold notation

[leander-j]

For a boolean variable, three different ways of setting (all) thresholds yield three different results:

1. Specifying only for different source/target

    imi_abs_threshold = {'bool': {
                             ((False,), (True,)): 0,
                             ((True,), (False,)): 0}},

2. Setting all others as well using the '*' notation

    imi_abs_threshold = {'bool': {
                             ((False,), (True,)): 0,
                             ((True,), (False,)): 0,
                             ('*', '*'): 0}},

same as

    imi_abs_threshold = {'bool': {
                                ((False,), (True,)): 0,
                                ((True,), (False,)): 0,
                                ((False,), (False,)): 0,
                                ((True,), (True,)): 0,
                                },

3. Setting independently of source-target

    imi_abs_threshold = {'bool': 0},

same as

    imi_abs_threshold = {'bool': {
                                ('*', '*'): 0
                                }},

This qualitative finding is the same when using different values, e.g. 5 instead of 0. The same behavior is found in imi_rel_threshold.

The difference between 1. and 2. suggests that in the former it might not be allowed for a trait to keep its value if not specified (threshold set to ∞), which would be (in my opinion) counterintuitive. The difference of 2. and 3. is unclear to me.

Defaulting to ∞ (or 1.0 in the imi_rel_threshold case) is yet another question.

To reproduce:

culture = M.Culture(
    imi_rate = {
        'bool': 100,   # -> 10000 updates
        #'ord': 1000,  # *10 in batch -> 10000 updates
        #'pair': 20, # *~50 in batch -> 10000 updates
        '*': 0,
        },
    imi_type = 'complex',
    imi_batch_n = {'bool': 5, 'ord': 10},
    imi_p_in_batch = {'pair': 0.5},
    imi_network = M.Culture.acquaintance_network,
    imi_p_neighbor_drawn = {'bool': 0.5, 'pair': 1.0},
    imi_n_neighbors_drawn = {'ord': 5},
    imi_abs_threshold = {'bool': {
                                ((False,), (True,)): 5,
                                ((True,), (False,)): 5,
                                ('*', '*'): 5
                                },
            'ord': 1, 
            'pair': {
            ((('a','*'),), '*'): 0,
           # ('*', '*'): 0
        }},
    imi_include_own_trait = {'pair': True, '*': False},
    imi_delta = {'pair': 10.0},  # evaluation will be done by Individual.imi_evaluate_pair    
    imi_p_imitate = 1.0
    )