def test_fit_numeric_dataset():
irep = IREP(random_state=42)
rip = RIPPER(random_state=42)
irep.fit(
CREDIT_DF, class_feat=CREDIT_CLASS_FEAT, pos_class=CREDIT_POS_CLASS,
)
assert irep.ruleset_ == CREDIT_IREP_RULESET_42
def test_verbosity():
irep_v5 = IREP(random_state=42, verbosity=5)
rip_v5 = RIPPER(random_state=42, verbosity=5)
irep_v5.fit(DF, class_feat=CLASS_FEAT, pos_class=POS_CLASS)
assert irep_v5.ruleset_ == IREP_RULESET_42
rip_v5.fit(DF, class_feat=CLASS_FEAT, pos_class=POS_CLASS)
assert rip_v5.ruleset_ == RIP_RULESET_42
def test_fit_X_y_np():
irep = IREP(random_state=42)
rip = RIPPER(random_state=42)
irep.fit(X_DF, y=Y_DF, pos_class=POS_CLASS)
assert irep.ruleset_ == IREP_RULESET_42
rip.fit(X_DF, y=Y_DF, pos_class=POS_CLASS)
assert rip.ruleset_ == RIP_RULESET_42
def test_fit_Xy_df():
irep = IREP(random_state=42)
rip = RIPPER(random_state=42)
irep.fit(DF, class_feat=CLASS_FEAT, pos_class=POS_CLASS)
assert irep.ruleset_ == IREP_RULESET_42
rip.fit(DF, class_feat=CLASS_FEAT, pos_class=POS_CLASS)
assert rip.ruleset_ == RIP_RULESET_42
def test_fit_Xy_np():
irep = IREP(random_state=42)
rip = RIPPER(random_state=42)
irep.fit(XY_NP, y=None, class_feat=NP_CLASS_FEAT, pos_class=POS_CLASS)
assert irep.ruleset_ == feat_to_num_rs(IREP_RULESET_42)
rip.fit(XY_NP, y=None, class_feat=NP_CLASS_FEAT, pos_class=POS_CLASS)
assert rip.ruleset_ == feat_to_num_rs(RIP_RULESET_42)
def test_df_isnt_modified():
old_df = pd.read_csv("credit.csv")
df = old_df.copy()
irep = IREP(random_state=42)
irep.fit(CREDIT_DF, class_feat=CREDIT_CLASS_FEAT, pos_class=CREDIT_POS_CLASS)
assert df.equals(old_df)
old_df = pd.read_csv("credit.csv")
df = old_df.copy()
rip = RIPPER(random_state=42)
rip.fit(CREDIT_DF, class_feat=CREDIT_CLASS_FEAT, pos_class=CREDIT_POS_CLASS)
assert df.equals(old_df)
def test_fit_discrete_dataset():
irep = IREP(random_state=0, n_discretize_bins=11)
rip = RIPPER(random_state=0, n_discretize_bins=11)
discrete_df = CREDIT_DF.select_dtypes(float).applymap(lambda x: int(x % 10))
discrete_df[CREDIT_CLASS_FEAT] = CREDIT_DF[CREDIT_CLASS_FEAT]
irep.fit(discrete_df, class_feat=CREDIT_CLASS_FEAT, pos_class=CREDIT_POS_CLASS)
assert not (irep.ruleset_.isuniversal()) and not (irep.ruleset_.isnull())
rip.fit(discrete_df, class_feat=CREDIT_CLASS_FEAT, pos_class=CREDIT_POS_CLASS)
assert not (rip.ruleset_.isuniversal()) and not (rip.ruleset_.isnull())
def test_random_state():
# Party dataset
irep_rulesets = []
rip_rulesets = []
for _ in range(3):
irep = IREP(random_state=72)
rip = RIPPER(random_state=72)
irep.fit(DF, class_feat=CLASS_FEAT, pos_class=POS_CLASS)
rip.fit(DF, class_feat=CLASS_FEAT, pos_class=POS_CLASS)
irep_rulesets.append(irep.ruleset_)
rip_rulesets.append(rip.ruleset_)
assert all(rs == irep_rulesets[0] for rs in irep_rulesets)
assert all(rs == rip_rulesets[0] for rs in rip_rulesets)
# Credit dataset
irep_rulesets = []
rip_rulesets = []
for _ in range(3):
irep = IREP(random_state=72)
rip = RIPPER(random_state=72)
irep.fit(CREDIT_DF,
class_feat=CREDIT_CLASS_FEAT,
pos_class=CREDIT_POS_CLASS)
rip.fit(CREDIT_DF,
class_feat=CREDIT_CLASS_FEAT,
pos_class=CREDIT_POS_CLASS)
irep_rulesets.append(irep.ruleset_)
rip_rulesets.append(rip.ruleset_)
assert all(rs == irep_rulesets[0] for rs in irep_rulesets)
assert all(rs == rip_rulesets[0] for rs in rip_rulesets)
def test_fit_boolean_dataset():
irep = IREP(random_state=42)
rip = RIPPER(random_state=42)
def tobool(x):
if x == "y":
return 0
elif x == "n":
return 1
else:
return 2
bool_df = DF.copy()
for col in bool_df.drop("Party", axis=1).columns:
bool_df[col] = bool_df[col].map(tobool)
irep.fit(bool_df, class_feat="Party", pos_class="democrat")
assert not (irep.ruleset_.isuniversal()) and not (irep.ruleset_.isnull())
def test_df_isnt_modified():
# df shouldn't be affected by side-effects during model fitting
old_df = pd.read_csv("credit.csv")
df = old_df.copy()
irep = IREP(random_state=42)
irep.fit(CREDIT_DF,
class_feat=CREDIT_CLASS_FEAT,
pos_class=CREDIT_POS_CLASS)
assert df.equals(old_df)
old_df = pd.read_csv("credit.csv")
df = old_df.copy()
rip = RIPPER(random_state=42)
rip.fit(CREDIT_DF,
class_feat=CREDIT_CLASS_FEAT,
pos_class=CREDIT_POS_CLASS)
assert df.equals(old_df)
def test_infer_pos_class():
irep = IREP(random_state=42)
rip = RIPPER(random_state=42)
infer_df = DF.copy()
infer_df[CLASS_FEAT] = infer_df[CLASS_FEAT].map(
lambda x: 1 if x == "democrat" else 0
)
irep.fit(
infer_df, class_feat=CLASS_FEAT,
)
assert irep.ruleset_ == IREP_RULESET_42
rip.fit(
infer_df, class_feat=CLASS_FEAT,
)
assert rip.ruleset_ == RIP_RULESET_42
def test_fit_XY_rename_columns():
irep = IREP(random_state=42)
rip = RIPPER(random_state=42)
# With xy
irep.fit(
XY_NP,
y=None,
class_feat=CLASS_FEAT,
pos_class=POS_CLASS,
feature_names=DF.columns,
)
assert irep.ruleset_ == IREP_RULESET_42
rip.fit(
XY_NP,
y=None,
class_feat=CLASS_FEAT,
pos_class=POS_CLASS,
feature_names=DF.columns,
)
assert rip.ruleset_ == RIP_RULESET_42
irep = IREP(random_state=42)
rip = RIPPER(random_state=42)
# With x_y
irep.fit(
X_NP,
y=Y_NP,
class_feat=CLASS_FEAT,
pos_class=POS_CLASS,
feature_names=DF.drop(CLASS_FEAT, axis=1).columns,
)
assert irep.ruleset_ == IREP_RULESET_42
rip.fit(
X_NP,
y=Y_NP,
class_feat=CLASS_FEAT,
pos_class=POS_CLASS,
feature_names=DF.drop(CLASS_FEAT, axis=1).columns,
)
assert rip.ruleset_ == RIP_RULESET_42
def test_deprecated_bin_transformer():
deprecated_bin_transformer = {
"A11": [(0, 1), (1, 2), (2, 4), (4, 8), (8, 17), (17, 67)],
"A15": [
(0, 1),
(1, 9),
(10, 105),
(108, 351),
(351, 1004),
(1058, 4607),
(4700, 100000),
],
"A3": [
(0.0, 0.415),
(0.415, 0.79),
(0.79, 1.375),
(1.375, 2.04),
(2.04, 3.04),
(3.04, 4.71),
(4.75, 7.04),
(7.08, 10.665),
(10.75, 14.585),
(14.79, 28.0),
],
"A8": [
(0.0, 0.04),
(0.04, 0.165),
(0.165, 0.335),
(0.335, 0.71),
(0.75, 1.25),
(1.25, 1.835),
(1.835, 2.79),
(3.0, 5.04),
(5.085, 13.0),
(13.5, 28.5),
],
}
df = pd.read_csv("credit.csv")
irep = IREP()
irep.fit(df, class_feat="Class", pos_class="+")
irep.bin_transformer_ = deprecated_bin_transformer
preds = irep.predict(df)
rip = RIPPER()
rip.fit(df, class_feat="Class", pos_class="+")
rip.bin_transformer_ = deprecated_bin_transformer
preds = rip.predict(df)
def test_same_inputs_give_same_results():
for random_state in range(3):
irep_res = []
rip_res = []
irep = IREP(random_state=random_state)
rip = RIPPER(random_state=random_state)
irep.fit(DF, class_feat=CLASS_FEAT, pos_class=POS_CLASS)
irep_res.append(irep.ruleset_)
rip.fit(DF, class_feat=CLASS_FEAT, pos_class=POS_CLASS)
rip_res.append(rip.ruleset_)
irep = IREP(random_state=random_state)
rip = RIPPER(random_state=random_state)
irep.fit(X_DF, y=Y_DF, class_feat=CLASS_FEAT, pos_class=POS_CLASS)
irep_res.append(irep.ruleset_)
rip.fit(X_DF, y=Y_DF, class_feat=CLASS_FEAT, pos_class=POS_CLASS)
rip_res.append(rip.ruleset_)
irep = IREP(random_state=random_state)
rip = RIPPER(random_state=random_state)
irep.fit(X_DF, y=Y_DF, pos_class=POS_CLASS)
irep_res.append(irep.ruleset_)
rip.fit(X_DF, y=Y_DF, pos_class=POS_CLASS)
rip_res.append(rip.ruleset_)
irep = IREP(random_state=random_state)
rip = RIPPER(random_state=random_state)
irep.fit(
XY_NP,
y=None,
class_feat=CLASS_FEAT,
pos_class=POS_CLASS,
feature_names=DF.columns,
)
irep_res.append(irep.ruleset_)
rip.fit(
XY_NP,
y=None,
class_feat=CLASS_FEAT,
pos_class=POS_CLASS,
feature_names=DF.columns,
)
rip_res.append(rip.ruleset_)
irep = IREP(random_state=random_state)
rip = RIPPER(random_state=random_state)
irep.fit(
X_NP,
y=Y_NP,
class_feat=CLASS_FEAT,
pos_class=POS_CLASS,
feature_names=DF.drop(CLASS_FEAT, axis=1).columns,
)
irep_res.append(irep.ruleset_)
rip.fit(
X_NP,
y=Y_NP,
class_feat=CLASS_FEAT,
pos_class=POS_CLASS,
feature_names=DF.drop(CLASS_FEAT, axis=1).columns,
)
rip_res.append(rip.ruleset_)
assert all([res == irep_res[0] for res in irep_res])
assert all([res == rip_res[0] for res in rip_res])
def test_use_initial_model():
initial_model = "[[A9=t ^ A10=t]]"
expected_irep = ruleset_fromstr(
"""[[A9=t ^ A10=t] V
[A9=t ^ A7=h] V
[A9=t ^ A4=u ^ A7=v]]
"""
)
expected_rip = ruleset_fromstr(
"""[[A9=t ^ A10=t] V
[A9=t ^ A7=h] V
[A9=t ^ A4=u ^ A14=0 ^ A15=0-0] V
[A9=t ^ A6=w]]
"""
)
# From str
irep = IREP(random_state=1)
irep.fit(credit_df, class_feat='Class', pos_class='+',
initial_model=initial_model
)
assert irep.ruleset_ == expected_irep
rip = RIPPER(random_state=1)
rip.fit(credit_df, class_feat='Class', pos_class='+',
initial_model=initial_model
)
assert rip.ruleset_ == expected_rip
# From IREP
initial_irep_model = IREP()
initial_irep_model.init_ruleset(initial_model)
irep = IREP(random_state=1)
irep.fit(credit_df, class_feat='Class', pos_class='+',
initial_model=initial_irep_model
)
assert irep.ruleset_ == expected_irep
rip = RIPPER(random_state=1)
rip.fit(credit_df, class_feat='Class', pos_class='+',
initial_model=initial_irep_model
)
assert rip.ruleset_ == expected_rip
# From RIP
initial_rip_model = RIPPER()
initial_rip_model.init_ruleset(initial_model)
irep = IREP(random_state=1)
irep.fit(credit_df, class_feat='Class', pos_class='+',
initial_model=initial_rip_model
)
assert irep.ruleset_ == expected_irep
rip = RIPPER(random_state=1)
rip.fit(credit_df, class_feat='Class', pos_class='+',
initial_model=initial_rip_model
)
assert rip.ruleset_ == expected_rip
# No side-effects
assert initial_irep_model.ruleset_ == ruleset_fromstr(initial_model)
assert initial_rip_model.ruleset_ == ruleset_fromstr(initial_model)
from copy import deepcopy
import os
import pytest
import pandas as pd
from wittgenstein.irep import IREP
from wittgenstein.ripper import RIPPER
from wittgenstein.base import Ruleset, ruleset_fromstr, rule_fromstr
DF = pd.read_csv("mushroom.csv")
original_ruleset_str = "[[Odor=f] V [Gill-size=n] V [Spore-print-color=r] V [Odor=m]]"
original_ruleset = ruleset_fromstr(original_ruleset_str)
original_rules = original_ruleset.rules
original_irep = IREP(random_state=42)
original_irep.fit(DF, class_feat="Poisonous/Edible", pos_class="p")
# Ensure setup works
assert original_ruleset == original_irep.ruleset_
credit_df = pd.read_csv("credit.csv")
credit_class_feat = "Class"
credit_pos_class = "+"
credit_rip = RIPPER(random_state=42, verbosity=0)
credit_rip.fit(credit_df, class_feat="Class", pos_class="+")
credit_original_ruleset = ruleset_fromstr(
"[[A9=t ^ A10=t ^ A4=u ^ A1=b ^ A11=7-16] V \
[A9=t ^ A10=t ^ A4=u ^ A11=3-7] V \
[A9=t ^ A10=t ^ A14=0] V \
[A9=t ^ A10=t] V \
[A9=t ^ A7=h ^ A6=q]]"
def test_initruleset():
irep = IREP(random_state=42)
irep.init_ruleset()
irep.ruleset_ == Ruleset()
irep = IREP(random_state=42)
irep.init_ruleset(original_ruleset)
irep.ruleset_ == original_ruleset
irep = IREP(random_state=42)
irep.init_ruleset(original_ruleset_str)
irep.ruleset_ == original_ruleset