How to modify loop builing ML models and generated DataFrame with column presented variable removed-CodePudding

I have Pandas DataFrame like below:

Input data:

Y - binnary target
X1...X5 - predictors

Source code of DataFrame:

import pandas as pd
import numpy as np

from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_curve, roc_auc_score
from sklearn.metrics import accuracy_score
from sklearn.metrics import recall_score
from sklearn.metrics import precision_score
from sklearn.metrics import roc_auc_score
from sklearn import metrics
from xgboost import XGBClassifier

df = pd.DataFrame()
df["Y"] = [1,0,1,0]
df["X1"] = [111,12,150,270]
df["X2"] = [22,33,44,55]
df["X3"] = [1,1,0,0]
df["X4"] = [0,0,0,1]
df["X5"] = [150, 222,230,500]

Y   | X1  | X2  | X3    | X4    | X5
----|-----|-----|-------|-------|-----
1   | 111 | 22  | 1     | 0     | 150
0   | 12  | 33  | 1     | 0     | 222
1   | 150 | 44  | 0     | 0     | 230
0   | 270 | 55  | 0     | 1     | 500

My code: -> I Run XGBClassifier() model, where in each successive iteration of the loop one variable is removed So, each successive model is built with 1 less variable than the previous one, the last model in the iteration is built with only 1 predictor

X_train, X_test, y_train, y_test = train_test_split(df.drop("Y", axis=1)
                                                    , df.Y
                                                    , train_size = 0.70
                                                    , test_size=0.30
                                                    , random_state=1
                                                    , stratify = df.Y)

results = []
list_of_models = []
Num_var_in = []
predictors = X_train.columns.tolist()
Var_out = []

for i in X_train.columns:
    
    #model building
    model = XGBClassifier()
    model.fit(X_train, y_train)
    list_of_models.append(model)
    
    #evaluation
    results.append({"AUC_train": round(metrics.roc_auc_score(y_train, model.predict_proba(X_train)[:,1]), 5),
                    "AUC_test": round(metrics.roc_auc_score(y_test, model.predict_proba(X_test)[:,1]), 5),})
    
    #Num_var_in - number of predictors which was used to create model during that iteration
    Num_var_in.append(len(X_train.columns.tolist()))
    
    #Var_out - name of variable which was removed during that iteration
    if sorted(predictors) == sorted(X_train.columns.tolist()):
        Var_out.append(np.nan)
    else:
        Var_out.append(set(predictors) - set(X_train.columns.tolist()))
   
    #drop 1 predictor after each loop iteration
    X_train = X_train.drop(i, axis=1)
    X_test = X_test.drop(i, axis=1)

#save results to DataFrame
results = pd.DataFrame(results)
results["Num_var_in"] = Num_var_in
results["Var_out"] = Var_out
results.reset_index(inplace = True)
results.rename(columns = {"index":"Model"}, inplace = True)
results

Current output:

Requirements:

In output in column "Var_out" I need to have one variable that has been discarded in a given iteration, not all that have been discarded so far

Desire output:

Model | AUC_train  | AUC_test   | Num_var_in  | Var_out
------|------------|------------|-------------|---------
0     | 0.5        | 0.5        | 5           | NaN
1     | 0.5        | 0.5        | 4           | X1
2     | 0.5        | 0.5        | 3           | X2
3     | 0.5        | 0.5        | 2           | X3
4     | 0.5        | 0.5        | 1           | X4

How can I modify my code in Python so as to have output in Var_out like in "Desire output" ?

CodePudding user response：

You can use: (check # HERE comments)

results = []
list_of_models = []
Num_var_in = []
predictors = X_train.columns.tolist()
Var_out = [np.nan]  # HERE (init with nan)

for i in X_train.columns:
    
    #model building
    model = XGBClassifier()
    model.fit(X_train, y_train)
    list_of_models.append(model)
    
    #evaluation
    results.append({"AUC_train": round(metrics.roc_auc_score(y_train, model.predict_proba(X_train)[:,1]), 5),
                    "AUC_test": round(metrics.roc_auc_score(y_test, model.predict_proba(X_test)[:,1]), 5),})
    
    #Num_var_in - number of predictors which was used to create model during that iteration
    Num_var_in.append(len(X_train.columns.tolist()))
    
    #Var_out - name of variable which was removed during that iteration
    Var_out.append(i)  # HERE (just append the current column)
   
    #drop 1 predictor after each loop iteration
    X_train = X_train.drop(i, axis=1)
    X_test = X_test.drop(i, axis=1)

#save results to DataFrame
results = pd.DataFrame(results)
results["Num_var_in"] = Num_var_in
results["Var_out"] = Var_out[:-1]  # HERE (remove the last value)
results.reset_index(inplace = True)
results.rename(columns = {"index":"Model"}, inplace = True)

Output:

>>> results
   Model  AUC_train  AUC_test  Num_var_in Var_out
0      0    1.00000   0.98270           5     NaN
1      1    1.00000   0.98590           4      X1
2      2    1.00000   0.97790           3      X2
3      3    0.99981   0.97075           2      X3
4      4    0.92516   0.59971           1      X4

Minimal reproducible example:

import pandas as pd
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split

X, y = make_classification(n_samples=1000, n_features=5, n_classes=2, random_state=2023)

df = pd.DataFrame(X, columns=[f'X{i}' for i in range(1, X.shape[1] 1)])
df = pd.concat([pd.Series(y, name='Y'), df], axis=1)

X_train, X_test, y_train, y_test = \
    train_test_split(df.iloc[:, 1:], df['Y'], test_size=0.2, random_state=2023)