def __init__(self, preprocessors, base_rules, params):
super().__init__(preprocessors, base_rules)
self.progress_advance_callback = None
assert params is not None
self.params = params
# top-level control procedure (rule ordering)
self.rule_ordering = params["Rule ordering"]
# top-level control procedure (covering algorithm)
self.covering_algorithm = params["Covering algorithm"]
if self.covering_algorithm == "exclusive":
self.cover_and_remove = self.exclusive_cover_and_remove
elif self.covering_algorithm == "weighted":
self.gamma = params["Gamma"]
self.cover_and_remove = self.weighted_cover_and_remove
# bottom-level search procedure (search algorithm)
self.rule_finder.search_algorithm.beam_width = params["Beam width"]
# bottom-level search procedure (search strategy)
self.rule_finder.search_strategy.constrain_continuous = True
# bottom-level search procedure (search heuristics)
evaluation_measure = params["Evaluation measure"]
if evaluation_measure == "entropy":
evaluator = EntropyEvaluator()
elif evaluation_measure == "laplace":
evaluator = LaplaceAccuracyEvaluator()
elif evaluation_measure == "wracc":
evaluator = WeightedRelativeAccuracyEvaluator()
self.rule_finder.quality_evaluator = evaluator
# bottom-level search procedure (over-fitting avoidance heuristics)
min_rule_cov = params["Minimum rule coverage"]
max_rule_length = params["Maximum rule length"]
self.rule_finder.general_validator.min_covered_examples = min_rule_cov
self.rule_finder.general_validator.max_rule_length = max_rule_length
# bottom-level search procedure (over-fitting avoidance heuristics)
default_alpha = params["Default alpha"]
parent_alpha = params["Parent alpha"]
self.rule_finder.significance_validator.default_alpha = default_alpha
self.rule_finder.significance_validator.parent_alpha = parent_alpha
# ADD COMMAND TO DEFINE LEARNER HERE
learners = {}
for beam_width in range(3, 11):
for min_covered_examples in range(7, 21):
for max_rule_length in range(2, 6):
if category == 'unordered':
learner = CN2UnorderedLearner()
else:
learner = CN2Learner()
if evaluator == 'laplace':
learner.rule_finder.quality_evaluator = LaplaceAccuracyEvaluator(
)
else:
learner.rule_finder.quality_evaluator = EntropyEvaluator()
learner.rule_finder.search_algorithm.beam_width = beam_width
learner.rule_finder.general_validator.min_covered_examples = min_covered_examples
learner.rule_finder.general_validator.max_rule_length = max_rule_length
learners[category + '_' + evaluator + '_' + str(beam_width) + '_' +
str(min_covered_examples) + '_' +
str(max_rule_length)] = learner
# We want to test our model now. The CrossValidation() function will do all the
# work in this case, which includes splitting the whole dataset into train and test subsets,
# then train the model, and produce results.
# So, simply initialize the CrossValidation() object from the 'testing' library
# and call it with input arguments 1) the dataset and 2) the learner.
# Note that the 'learner' argument should be in array form, i.e. '[learner]'.
cv = CrossValidation(k=10, random_state=44)
results = {}