def test_no_pruning_unless_gain_is_less_than_threshold():
examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'}] * 99
examples += [{'data': {'weather': 'rainy'}, 'conclusion': 'sad'}]
tree = build(examples)
tree = prune(examples, tree, threshold=0.01)
assert (tree.decide({'weather': 'sunny'}) == {'happy': 1.0})
assert (tree.decide({'weather': 'rainy'}) == {'sad': 1.0})
def test_prune_can_remove_outliers_leaving_missing_example_data():
examples = [{'data': {}, 'conclusion': 'happy'}] * 99
examples += [{'data': {'weather': 'rainy'}, 'conclusion': 'sad'}]
tree = build(examples)
tree = prune(examples, tree, threshold=0.1)
assert (tree.decide({}) == {'happy': 1.0})
assert (tree.decide({'weather': 'rainy'}) == {'happy': 1.0})
def test_treat_gap_between_same_values_as_decidable():
examples = [{'data': {'rainfall': 0.1}, 'conclusion': 'happy'},
{'data': {'rainfall': 0.3}, 'conclusion': 'happy'},
{'data': {'rainfall': 0.5}, 'conclusion': 'sad'}]
tree = build(examples)
assert(tree.decide({'rainfall': 0.2}) == {'happy': 1.0})
assert(approximately(tree.decide({'rainfall': 0.4})) == {'happy': 0.66, 'sad': 0.33})
def test_treat_gap_between_numerical_values_as_undecidable():
examples = [{'data': {'rainfall': 0.1}, 'conclusion': 'happy'},
{'data': {'rainfall': 0.3}, 'conclusion': 'sad'}]
tree = build(examples)
assert(tree.decide({'rainfall': 0.0}) == {'happy': 1.0})
assert(tree.decide({'rainfall': 0.2}) == {'happy': 0.5, 'sad': 0.5})
assert(tree.decide({'rainfall': 0.4}) == {'sad': 1.0})
def test_prune_removes_outliers():
examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'}] * 99
examples += [{'data': {'weather': 'rainy'}, 'conclusion': 'sad'}]
tree = build(examples)
tree = prune(examples, tree, threshold=0.1)
assert (tree.decide({'weather': 'sunny'}) == {'happy': 1.0})
assert (tree.decide({'weather': 'rainy'}) == {'happy': 1.0})
def test_probablity_is_apportioned_down_the_tree():
examples = [{'data': {'sky': 'sunny', 'rain': 'no'}, 'conclusion': 'happy'},
{'data': {'sky': 'cloudy', 'rain': 'no'}, 'conclusion': 'cheerful'},
{'data': {'sky': 'sunny', 'rain': 'yes'}, 'conclusion': 'confused'},
{'data': {'sky': 'cloudy', 'rain': 'yes'}, 'conclusion': 'sad'}]
tree = build(examples)
assert(tree.decide({}) == {'happy': 0.25, 'sad': 0.25,
'cheerful': 0.25, 'confused': 0.25})
def test_missing_data_in_examples_is_treated_as_an_option():
examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'},
{'data': {}, 'conclusion': 'cheerful'},
{'data': {'weather': 'rainy'}, 'conclusion': 'sad'}]
tree = build(examples)
assert(tree.decide({'weather': 'sunny'}) == {'happy': 1.0})
assert(tree.decide({}) == {'cheerful': 1.0})
assert(tree.decide({'weather': 'rainy'}) == {'sad': 1.0})
def test_can_choose_from_a_three_choice_option():
examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'},
{'data': {'weather': 'cloudy'}, 'conclusion': 'cheerful'},
{'data': {'weather': 'rainy'}, 'conclusion': 'sad'}]
tree = build(examples)
assert(tree.decide({'weather': 'sunny'}) == {'happy': 1.0})
assert(tree.decide({'weather': 'cloudy'}) == {'cheerful': 1.0})
assert(tree.decide({'weather': 'rainy'}) == {'sad': 1.0})
def test_can_approximately_add_up():
examples = [{'data': {'a': 2.0, 'b': 3.0}, 'conclusion': 5.0},
{'data': {'a': 0.0, 'b': 2.0}, 'conclusion': 2.0},
{'data': {'a': 2.0, 'b': 2.0}, 'conclusion': 4.0},
{'data': {'a': 1.0, 'b': 3.0}, 'conclusion': 4.0}]
tree = build(examples)
assert(tree.decide({'a': 0.0, 'b': 2.0}) == {2.0: 1.0})
assert(tree.decide({'a': 0.0, 'b': 0.0}) == {2.0: 1.0})
assert(tree.decide({'a': 4.0, 'b': 4.0}) == {5.0: 1.0})
def test_can_decide_with_mixed_finite_and_numerical_data():
examples = [{'data': {'rainfall': 0.1, 'sky': 'sunny'}, 'conclusion': 'happy'},
{'data': {'rainfall': 0.1, 'sky': 'cloudy'}, 'conclusion': 'cheerful'},
{'data': {'rainfall': 2.0, 'sky': 'sunny'}, 'conclusion': 'cheerful'},
{'data': {'rainfall': 2.0, 'sky': 'cloudy'}, 'conclusion': 'sad'}]
tree = build(examples)
assert (tree.decide({'rainfall': 0.0, 'sky': 'sunny'}) == {'happy': 1.0})
assert (tree.decide({'rainfall': 2.1, 'sky': 'sunny'}) == {'cheerful': 1.0})
assert (tree.decide({'rainfall': 0.0, 'sky': 'cloudy'}) == {'cheerful': 1.0})
assert (tree.decide({'rainfall': 2.1, 'sky': 'cloudy'}) == {'sad': 1.0})
def test_two_fields_with_first_one_redundant():
examples = [{'data': {'weather': 'sunny', 'horoscope': 'bad'}, 'conclusion': 'sad'},
{'data': {'weather': 'rainy', 'horoscope': 'bad'}, 'conclusion': 'sad'},
{'data': {'weather': 'sunny', 'horoscope': 'good'}, 'conclusion': 'happy'},
{'data': {'weather': 'rainy', 'horoscope': 'good'}, 'conclusion': 'happy'}]
tree = build(examples)
assert(tree.decide({'weather': 'sunny', 'horoscope': 'bad'}) == {'sad': 1.0})
assert(tree.decide({'weather': 'sunny', 'horoscope': 'good'}) == {'happy': 1.0})
assert(tree.decide({'weather': 'rainy', 'horoscope': 'bad'}) == {'sad': 1.0})
assert(tree.decide({'weather': 'rainy', 'horoscope': 'good'}) == {'happy': 1.0})
def test_decide_on_two_xor_fields():
examples = [{'data': {'weather': 'sunny', 'horoscope': 'bad'}, 'conclusion': 'happy'},
{'data': {'weather': 'rainy', 'horoscope': 'bad'}, 'conclusion': 'sad'},
{'data': {'weather': 'sunny', 'horoscope': 'good'}, 'conclusion': 'sad'},
{'data': {'weather': 'rainy', 'horoscope': 'good'}, 'conclusion': 'happy'}]
tree = build(examples)
assert(tree.decide({'weather': 'sunny', 'horoscope': 'bad'}) == {'happy': 1.0})
assert(tree.decide({'weather': 'sunny', 'horoscope': 'good'}) == {'sad': 1.0})
assert(tree.decide({'weather': 'rainy', 'horoscope': 'bad'}) == {'sad': 1.0})
assert(tree.decide({'weather': 'rainy', 'horoscope': 'good'}) == {'happy': 1.0})
def test_datum_labels_do_not_affect_the_decision():
examples = [{
'data': {
'weather': 'clear'
},
'conclusion': 'happy'
}, {
'data': {
'weather': 'precipitation'
},
'conclusion': 'sad'
}]
tree = build(examples)
assert (tree.decide({'weather': 'clear'}) == {'happy': 1.0})
assert (tree.decide({'weather': 'precipitation'}) == {'sad': 1.0})
def test_conclusion_labels_do_not_affect_the_decision():
examples = [{
'data': {
'weather': 'sunny'
},
'conclusion': 'overjoyed'
}, {
'data': {
'weather': 'rain'
},
'conclusion': 'depressed'
}]
tree = build(examples)
assert (tree.decide({'weather': 'sunny'}) == {'overjoyed': 1.0})
assert (tree.decide({'weather': 'rain'}) == {'depressed': 1.0})
def test_decide_between_two_examples():
examples = [{
'data': {
'weather': 'sunny'
},
'conclusion': 'happy'
}, {
'data': {
'weather': 'rain'
},
'conclusion': 'sad'
}]
tree = build(examples)
assert (tree.decide({'weather': 'sunny'}) == {'happy': 1.0})
assert (tree.decide({'weather': 'rain'}) == {'sad': 1.0})
def test_can_handle_a_fuzzy_numerical_boundary():
examples = [{'data': {'rainfall': 0.1}, 'conclusion': 'happy'},
{'data': {'rainfall': 0.2}, 'conclusion': 'sad'},
{'data': {'rainfall': 0.3}, 'conclusion': 'happy'},
{'data': {'rainfall': 0.4}, 'conclusion': 'sad'}]
tree = build(examples)
assert(tree.decide({'rainfall': 0.05}) == {'happy': 1.0})
assert(tree.decide({'rainfall': 0.1}) == {'happy': 1.0})
assert(tree.decide({'rainfall': 0.15}) == {'happy': 0.5, 'sad': 0.5})
assert(tree.decide({'rainfall': 0.2}) == {'sad': 1.0})
assert(approximately(tree.decide({'rainfall': 0.25})) == {'happy': 0.33, 'sad': 0.66})
assert(tree.decide({'rainfall': 0.3}) == {'happy': 1.0})
assert(tree.decide({'rainfall': 0.35}) == {'happy': 0.75, 'sad': 0.25})
assert(tree.decide({'rainfall': 0.4}) == {'sad': 1.0})
assert(tree.decide({'rainfall': 0.45}) == {'sad': 1.0})
def test_field_names_do_not_affect_the_decision():
examples = [{
'data': {
'climate': 'sunny'
},
'conclusion': 'happy'
}, {
'data': {
'climate': 'rain'
},
'conclusion': 'sad'
}]
tree = build(examples)
assert (tree.decide({'climate': 'sunny'}) == {'happy': 1.0})
assert (tree.decide({'climate': 'rain'}) == {'sad': 1.0})
def test_prune_does_nothing_on_decisive_tree():
examples = [{
'data': {
'weather': 'sunny'
},
'conclusion': 'happy'
}, {
'data': {
'weather': 'rainy'
},
'conclusion': 'sad'
}]
tree = build(examples)
tree = prune(examples, tree)
assert (tree.decide({'weather': 'sunny'}) == {'happy': 1.0})
assert (tree.decide({'weather': 'rainy'}) == {'sad': 1.0})
def test_can_prune_tree_with_missing_example_data():
examples = [{
'data': {
'weather': 'sunny'
},
'conclusion': 'happy'
}, {
'data': {},
'conclusion': 'cheerful'
}, {
'data': {
'weather': 'rainy'
},
'conclusion': 'sad'
}]
tree = build(examples)
tree = prune(examples, tree, threshold=0.1)
assert (tree.decide({'weather': 'sunny'}) == {'happy': 1.0})
# assert(tree.decide({}) == {'cheerful': 1.0})
assert (tree.decide({'weather': 'rainy'}) == {'sad': 1.0})
def test_missing_example_determines_decision_emphatically_when_test_data_is_missing():
examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'},
{'data': {}, 'conclusion': 'cheerful'}]
tree = build(examples)
assert(tree.decide({}) == {'cheerful': 1.0})
def test_can_decide_on_single_numerical_value():
examples = [{'data': {'rainfall': 0.1}, 'conclusion': 'happy'}]
tree = build(examples)
assert(tree.decide({'rainfall': 0.0}) == {'happy': 1.0})
assert(tree.decide({'rainfall': 0.1}) == {'happy': 1.0})
assert(tree.decide({'rainfall': 0.2}) == {'happy': 1.0})
def test_with_no_data_falls_back_training_weights():
examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'},
{'data': {'weather': 'sunny'}, 'conclusion': 'happy'},
{'data': {'weather': 'rainy'}, 'conclusion': 'sad'}]
tree = build(examples)
assert(approximately(tree.decide({})) == {'happy': 0.66, 'sad': 0.33})
def test_with_no_data_result_is_divided():
examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'},
{'data': {'weather': 'rainy'}, 'conclusion': 'sad'}]
tree = build(examples)
assert(tree.decide({}) == {'happy': 0.5, 'sad': 0.5})
def test_produce_a_numerical_result_from_a_single_example():
examples = [{'data': {'rainfall': 'high'}, 'conclusion': 2.0}]
tree = build(examples)
assert(tree.decide({'rainfall': 'high'}) == {2.0: 1.0})
def test_choose_a_value_from_numerical_outcomes():
examples = [{'data': {'rainfall': 'high'}, 'conclusion': 2.0},
{'data': {'rainfall': 'low'}, 'conclusion': 0.0}]
tree = build(examples)
assert(tree.decide({'rainfall': 'high'}) == {2.0: 1.0})
assert(tree.decide({'rainfall': 'low'}) == {0.0: 1.0})
def test_only_one_conclusion():
examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'},
{'data': {'weather': 'rainy'}, 'conclusion': 'happy'}]
tree = build(examples)
assert(tree.decide({'weather': 'sunny'}) == {'happy': 1.0})
def test_decide_with_just_one_example():
examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'}]
tree = build(examples)
assert (tree.decide({'weather': 'sunny'}) == {'happy': 1.0})
def test_unseen_option_treated_as_missing_data():
examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'},
{'data': {'weather': 'rainy'}, 'conclusion': 'sad'}]
tree = build(examples)
assert(tree.decide({'weather': 'cloudy'}) == {'happy': 0.5, 'sad': 0.5})
def test_decide_based_on_two_examples_separated_numerically():
examples = [{'data': {'rainfall': 0.1}, 'conclusion': 'happy'},
{'data': {'rainfall': 2.7}, 'conclusion': 'sad'}]
tree = build(examples)
assert(tree.decide({'rainfall': 0.0}) == {'happy': 1.0})
assert(tree.decide({'rainfall': 2.9}) == {'sad': 1.0})
