XAI-DESReg

Installation

pip install xaidesreg

Example

Loading necessary libraries and dataset:

import pandas as pd 

# models 
from xgboost import XGBRegressor
from sklearn.ensemble import RandomForestRegressor
from sklearn.neighbors import KNeighborsRegressor
from sklearn.linear_model import LinearRegression 

# metrics 
from sklearn.model_selection import train_test_split  
from sklearn.preprocessing import LabelEncoder
from ucimlrepo import fetch_ucirepo

# dataset
from ucimlrepo import fetch_ucirepo   

# InfoDESReg 
from xaidesreg import DES

Loading data:

abalone = fetch_ucirepo(id=1) 
X = abalone.data.features 
y = abalone.data.targets  

le = LabelEncoder() 
X['Sex'] = le.fit_transform(X['Sex'])  

Split the dataset into training, validation for DES (DSEL) and testing.

TEST_SIZE = 0.2 

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=TEST_SIZE, random_state=42) 
X_train, X_dsel, y_train, y_dsel = train_test_split(X_train, y_train, test_size=0.2, random_state=42)   

1. Models pool generation

pool_models = [
    XGBRegressor(random_state=42), 
    RandomForestRegressor(random_state=42), 
    LinearRegression(), 
    KNeighborsRegressor(), 
]

2. Train the models (pool):

for model in pool_models: 
    model.fit(X_train, y_train)  

3. Usage of our library:

des = DES(pool_regressors=pool_models, 
          k=6, 
          knn_metric='minkowski', 
          metrics='mape', 
          threshold=0.1)

des.fit(X_dsel, y_dsel) 

4. Testing

des.score(X_test, y_test) # based on MSE 

5. Explainability

index = 47
X_test.iloc[index]

prediction, contribution_df, neighbors_df = des.predict_xai(X_test.iloc[[index]])

print(prediction) 

The content of contribution data frame:

Model	Predictions	Competence	Weights
XGBRegressor	9.133	0.064	15.609
RandomForestRegressor	8.890	0.078	12.853
LinearRegression	8.913	0.090	11.141
KNeighborsRegressor	9.000	0.072	13.846

The samples in the region of competence (neighbors_df):

Sex	Length	Diameter	Height	Whole_weight	Shucked_weight	Viscera_weight	Shell_weight	Target
2	0.565	0.44	0.125	0.802	0.3595	0.1825	0.215	9
2	0.55	0.425	0.15	0.8315	0.411	0.1765	0.2165	10
2	0.56	0.415	0.13	0.7615	0.3695	0.17	0.1955	8
2	0.545	0.41	0.12	0.793	0.434	0.1405	0.19	9
2	0.56	0.415	0.145	0.852	0.43	0.1885	0.205	8
2	0.54	0.42	0.135	0.8075	0.3485	0.1795	0.235	11