def test_rule_num_covered():
rule = Rule([COND1, COND4])
rule_num_covered = rule.num_covered(FIRST_10_EXAMPLES)
len_assertion_df = len(
FIRST_10_EXAMPLES[(FIRST_10_EXAMPLES[FEATURE1] == VALUE1)
& (FIRST_10_EXAMPLES[FEATURE4] == VALUE4)])
assert (rule_num_covered == len_assertion_df)
def test_asruleset():
assert asruleset(Ruleset([Rule([Cond("hello", "world")])
])) == Ruleset([Rule([Cond("hello", "world")])])
assert asruleset("[[hello=world]]") == Ruleset(
[Rule([Cond("hello", "world")])])
assert asruleset(["[hello=world]"
]) == Ruleset([Rule([Cond("hello", "world")])])
def test_rule_covers():
rule = Rule([COND1, COND4])
rule_covers = rule.covers(FIRST_10_EXAMPLES)
assertion_df = FIRST_10_EXAMPLES[(FIRST_10_EXAMPLES[FEATURE1] == VALUE1)
& (FIRST_10_EXAMPLES[FEATURE4] == VALUE4)]
assert (set(rule_covers.index.tolist()) == set(
assertion_df.index.tolist()))
def grow_rule(pos_df,
neg_df,
possible_conds,
initial_rule=Rule(),
max_rule_conds=None,
verbosity=0):
""" Fit a new rule to add to a ruleset """
rule0 = copy.deepcopy(initial_rule)
if verbosity >= 4:
print(f'growing rule from initial rule: {rule0}')
rule1 = copy.deepcopy(rule0)
while (len(rule0.covers(neg_df)) > 0 and rule1 is not None) and (
max_rule_conds is None or len(rule1.conds) < max_rule_conds
): # Stop refining rule if no negative examples remain
rule1 = best_successor(rule0,
possible_conds,
pos_df,
neg_df,
verbosity=verbosity)
if rule1 is not None:
rule0 = rule1
if verbosity >= 4:
print(f'negs remaining {len(rule0.covers(neg_df))}')
if not rule0.isempty():
if verbosity >= 2: print(f'grew rule: {rule0}')
return rule0
else:
#warning_str = f"grew an empty rule: {rule0} over {len(pos_idx)} pos and {len(neg_idx)} neg"
#_warn(warning_str, RuntimeWarning, filename='base_functions', funcname='grow_rule')
return rule0
def grow_rule_cn(cn,
pos_idx,
neg_idx,
initial_rule=Rule(),
max_rule_conds=None,
verbosity=0):
""" Fit a new rule to add to a ruleset """
rule0 = copy.deepcopy(initial_rule)
rule1 = copy.deepcopy(rule0)
if verbosity >= 4:
print(f'growing rule from initial rule: {rule0}')
num_neg_covered = len(cn.rule_covers(rule0, subset=neg_idx))
user_halt = (max_rule_conds is not None
and len(rule1.conds) >= max_rule_conds)
while num_neg_covered > 0: # Stop refining rule if no negative examples remain
rule1 = best_rule_successor_cn(cn,
rule0,
pos_idx,
neg_idx,
verbosity=verbosity)
if rule1 is None: break
rule0 = rule1
num_neg_covered = len(cn.rule_covers(rule0, neg_idx))
if verbosity >= 4:
print(f'negs remaining: {num_neg_covered}')
if not rule0.isempty():
if verbosity >= 2: print(f'grew rule: {rule0}')
return rule0
else:
#warning_str = f"grew an empty rule: {rule0} over {len(pos_idx)} pos and {len(neg_idx)} neg"
#_warn(warning_str, RuntimeWarning, filename='base_functions', funcname='grow_rule_cn')
return rule0
def feat_to_num_rs(ruleset):
new_ruleset = Ruleset()
for rule in ruleset.rules:
new_rule = Rule()
for cond in rule.conds:
feat = cond.feature
val = cond.val
new_cond = Cond(FEAT2IDX[feat], val)
new_rule.conds.append(new_cond)
new_ruleset.rules.append(new_rule)
return new_ruleset
def best_rule_successor_cn(cn, rule, pos_idx, neg_idx, verbosity=0):
best_cond = None
best_gain = float('-inf')
rule_covers_pos_idx = cn.rule_covers(rule, pos_idx)
rule_covers_neg_idx = cn.rule_covers(rule, neg_idx)
for cond_action_step in cn.conds:
g = gain_cn(cn, cond_action_step, rule_covers_pos_idx,
rule_covers_neg_idx)
if g > best_gain:
best_gain = g
best_cond = cond_action_step
if verbosity >= 5: print(f'gain {rnd(best_gain)} {best_cond}')
return Rule(rule.conds + [best_cond]) if best_gain > 0 else None
def best_rule_successor_cn(cn, rule, pos_idx, neg_idx, verbosity=0):
"""Return for a Rule its best successor Rule according to FOIL information gain metric."""
best_cond = None
best_gain = float("-inf")
rule_covers_pos_idx = cn.rule_covers(rule, pos_idx)
rule_covers_neg_idx = cn.rule_covers(rule, neg_idx)
for cond_action_step in cn.conds:
g = gain_cn(cn, cond_action_step, rule_covers_pos_idx,
rule_covers_neg_idx)
if g > best_gain:
best_gain = g
best_cond = cond_action_step
if verbosity >= 5:
print(f"gain {rnd(best_gain)} {best_cond}")
return Rule(rule.conds + [best_cond]) if best_gain > 0 else None
def test_empty_rules_equal():
assert (Rule() == Rule())
def test_rules_len_2_unequal():
rule1 = Rule([COND1, COND2])
rule2 = Rule([COND2, COND3])
assert (rule1 != rule2)
def test_disordered_rules_len_2_equal():
rule1 = Rule([COND1, COND2])
rule2 = Rule([COND2, COND1])
assert (rule1 == rule2)
def test_rules_len_2_equal():
rule1 = Rule([COND1, COND2])
rule2 = Rule([COND1, COND2])
assert (rule1 == rule2)
def test_asrule():
assert asrule(Rule([Cond("hello",
"world")])) == Rule([Cond("hello", "world")])
assert asrule(["hello=world"]) == Rule([Cond("hello", "world")])
assert asrule("[hello=world]") == Rule([Cond("hello", "world")])
DF = pd.read_csv("house-votes-84.csv")
CLASS_FEAT = "Party"
DEFAULT_CLASS_FEAT = "Class"
POS_CLASS = "democrat"
SPLIT_SIZE = 0.6
X_DF = DF.drop(CLASS_FEAT, axis=1)
Y_DF = DF[CLASS_FEAT]
XY_NP = DF.values
X_NP = X_DF.values
Y_NP = Y_DF.values
NP_CLASS_FEAT = 0
IREP_RULESET_42 = Ruleset([
Rule([Cond("physician-fee-freeze", "n")]),
Rule([
Cond("synfuels-corporation-cutback", "y"),
Cond("education-spending", "n")
]),
])
RIP_RULESET_42 = Ruleset([
Rule([Cond("physician-fee-freeze", "n")]),
Rule([
Cond("synfuels-corporation-cutback", "y"),
Cond("education-spending", "n")
]),
Rule([
Cond("synfuels-corporation-cutback", "y"),
Cond("adoption-of-the-budget-resolution", "y"),
def test_rules_len_1_equal():
assert (Rule([COND1]) == Rule([COND1]))
def test_rules_len_1_unequal():
assert (Rule([COND1]) != Rule([COND2]))
def _ruleset_from_df(self, model_df):
rules = []
for _, row in model_df.iterrows():
rules.append(Rule([cond_fromstr(c) for c in row if isinstance(c, str)]))
return Ruleset(rules)