import pandas as pd
from scipy.io import arff
from rulekit import RuleKit
from rulekit.survival import SurvivalRules
from rulekit.params import Measures

from rulexai.explainer import RuleExplainer

GBSG2

Read data

dataset_path = "./data/GBSG2.arff"
data = pd.DataFrame(arff.loadarff(dataset_path)[0])

# code to change encoding of the file
tmp_df = data.select_dtypes([object])
tmp_df = tmp_df.stack().str.decode("utf-8").unstack()
for col in tmp_df:
    data[col] = tmp_df[col].replace({"?": None})

x = data.drop(["survival_status"], axis=1)
y = data["survival_status"]

Train RuleKit model

# RuleKit
RuleKit.init()

srv = SurvivalRules(survival_time_attr="survival_time")
srv.fit(values=x, labels=y)

<rulekit.survival.SurvivalRules at 0x176db91a880>

Rules

for rule in srv.model.rules:
    print(rule, rule.stats)

IF pnodes = (-inf, 3.50) THEN survival_status = {NaN} (p = 304.0, n = 0.0, P = 564.0, N = 0.0, weight = 0.9999999999999998, pvalue = 2.220446049250313e-16)
IF pnodes = (-inf, 17.50) AND progrec = (-inf, 9.50) AND age = <41.50, 52.50) AND estrec = <0.50, 29) THEN survival_status = {NaN} (p = 21.0, n = 0.0, P = 564.0, N = 0.0, weight = 0.9999999999909083, pvalue = 9.09172737095787e-12)
IF pnodes = <4.50, 19) AND progrec = (-inf, 11.50) AND age = <41.50, 64.50) AND estrec = <0.50, 41) THEN survival_status = {NaN} (p = 33.0, n = 0.0, P = 564.0, N = 0.0, weight = 1.0, pvalue = 0.0)
IF pnodes = <4.50, inf) AND progrec = (-inf, 25.50) THEN survival_status = {NaN} (p = 113.0, n = 0.0, P = 564.0, N = 0.0, weight = 1.0, pvalue = 0.0)
IF pnodes = <4.50, inf) AND progrec = (-inf, 99) THEN survival_status = {NaN} (p = 156.0, n = 0.0, P = 564.0, N = 0.0, weight = 1.0, pvalue = 0.0)
IF pnodes = <5.50, inf) AND progrec = (-inf, 135) THEN survival_status = {NaN} (p = 144.0, n = 0.0, P = 564.0, N = 0.0, weight = 1.0, pvalue = 0.0)
IF pnodes = <4.50, inf) AND progrec = (-inf, 233) THEN survival_status = {NaN} (p = 185.0, n = 0.0, P = 564.0, N = 0.0, weight = 1.0, pvalue = 0.0)
IF pnodes = (-inf, 4.50) AND progrec = <9, inf) AND age = <39.50, inf) THEN survival_status = {NaN} (p = 245.0, n = 0.0, P = 564.0, N = 0.0, weight = 1.0, pvalue = 0.0)
IF progrec = <107, inf) THEN survival_status = {NaN} (p = 168.0, n = 0.0, P = 564.0, N = 0.0, weight = 0.9999999989621143, pvalue = 1.0378856662995872e-09)
IF pnodes = <3.50, inf) AND progrec = (-inf, 105.50) THEN survival_status = {NaN} (p = 195.0, n = 0.0, P = 564.0, N = 0.0, weight = 1.0, pvalue = 0.0)

RuleXAI

explainer = RuleExplainer(model=srv, X=x, y=y, type="survival")
explainer.explain()

<rulexai.explainer.RuleExplainer at 0x176db937700>

Feature importance

explainer.feature_importances_

	attributes	importances
2	pnodes	460.222804
3	progrec	251.499862
0	age	20.523849
1	estrec	13.347720

Condition importance

explainer.condition_importances_

	conditions	importances
0	pnodes = <4.5, inf)	207.268572
1	pnodes = (-inf, 3.5)	67.394775
2	pnodes = <5.5, inf)	64.254026
3	pnodes = <3.5, inf)	64.104973
4	progrec = (-inf, 25.5)	48.923100
5	progrec = <107.0, inf)	37.252374
6	progrec = (-inf, 105.5)	33.962572
7	progrec = (-inf, 99.0)	33.423755
8	pnodes = (-inf, 4.5)	32.835122
9	progrec = (-inf, 135.0)	25.353218
10	progrec = (-inf, 11.5)	23.663185
11	progrec = (-inf, 9.5)	23.506762
12	pnodes = <4.5, 19.0)	18.150272
13	progrec = <9.0, inf)	13.146344
14	progrec = (-inf, 233.0)	12.268552
15	estrec = <0.5, 29.0)	10.450381
16	age = <41.5, 64.5)	9.275232
17	age = <41.5, 52.5)	8.077389
18	pnodes = (-inf, 17.5)	6.215064
19	age = <39.5, inf)	3.171229
20	estrec = <0.5, 41.0)	2.897339

Local explainability

explainer.local_explainability(x.iloc[0, :], pd.DataFrame(y).iloc[0, :], plot = True)

Example:
horTh                  no
age                  70.0
menostat             Post
tsize                21.0
tgrade                 II
pnodes                3.0
progrec              48.0
estrec               66.0
survival_time      1814.0
survival_status       1.0
Name: 0, dtype: object

Rules that covers this example:
IF pnodes = (-inf, 3.5) THEN survival_status = {NaN}
IF pnodes = (-inf, 4.5) AND progrec = <9.0, inf) AND age = <39.5, inf) THEN survival_status = {NaN}

Importances of the conditions from rules covering the example
             conditions  importances
0  pnodes = (-inf, 3.5)    67.394775
1  pnodes = (-inf, 4.5)    32.835122
2  progrec = <9.0, inf)    13.146344
3     age = <39.5, inf)     3.171229

	conditions	importances
0	pnodes = (-inf, 3.5)	67.394775
1	pnodes = (-inf, 4.5)	32.835122
2	progrec = <9.0, inf)	13.146344
3	age = <39.5, inf)	3.171229