如何生成分类器的集合？

Question

我想为 GaussianNB、线性 SVC、MLPClassifier 和 DecisionTreeClassifier 构建一个包含 100 个分类器的集合，存储为一个列表。 然后我想测量每个分类器的单个弱学习器的性能。

首先，我定义了分类器的 10 倍评估。 然后，我构建了一个弱学习器来分析每个分类器。

我的代码引发AttributeError: 'tuple' object has no attribute 'predict'错误。

import numpy as np
import pandas as pd
from sklearn.model_selection import StratifiedKFold
from sklearn.metrics import accuracy_score
from sklearn.naive_bayes import GaussianNB
from sklearn.svm import SVC
from sklearn.neural_network import MLPClassifier
from sklearn.tree import DecisionTreeClassifier

df = pd.read_csv("mushroom_dataset.csv")
df.head()

班级	帽形	帽面	帽子颜色	瘀伤	气味
磷	X	秒	n	吨	磷
电子	X	秒	是	吨	一种

X = df.loc[:, df.columns != 'class'].values
y = df.loc[:, df.columns == 'class'].values.ravel()

# 10-fold CV evaluation of the classifiers
def eval_classifier(_clf, _X, _y):
    acc = []
    kf = StratifiedKFold(n_splits=10, shuffle=False, random_state=None)
    for train_index, test_index in kf.split(_X, _y):
        _clf.fit(_X[train_index], _y[train_index])
        y_pred = _clf.predict(_X[test_index])
        acc += [accuracy_score(_y[test_index], y_pred)]
    return np.array(acc)

# Main Priors
counts = np.unique(y, return_counts=True)
NBpriors = [counts[1][0]/len(y), counts[1][1]/len(y)]

# Naive Bayes with the priors parameter
gnb_priors = eval_classifier(GaussianNB(priors=NBpriors), X.todense(), y)
print(f'Naive Bayes CV accuracy={np.mean(gnb_priors):.2f} {chr(177)}{np.std(gnb_priors):.3f}')

# Build a weak learner
def weak_fit(_list_cols, _X, _y):
    estimators = {}
    Xs = _X[:, _list_cols]
    gnb = GaussianNB(priors=NBpriors).fit(Xs, _y)
    estimators["Gaussian"] = GaussianNB(priors=NBpriors).fit(Xs, _y)
    lsvc = SVC(kernel='linear', probability=True).fit(Xs, _y)
    estimators["Linear SVC"] = SVC(kernel='linear', probability=True).fit(Xs, _y)
    mlp = MLPClassifier().fit(Xs, _y)
    estimators["MLP"] = MLPClassifier().fit(Xs, _y)
    dt = DecisionTreeClassifier().fit(Xs, _y)
    estimators["Decision Tree"] = DecisionTreeClassifier().fit(Xs, _y)
    return estimators, gnb, lsvc, mlp, dt

def weak_predict(_clf, _list_cols, _X):
    Xs = _X[:, _list_cols]
    return _clf.predict(Xs), _clf.predict_proba(Xs)

# Use _m features randomly selected, a total of M weak learners
def features_randomsubset(_M, _m, nensemble=1):
    from numpy.random import choice
    # returns a list of list of column choices - subset features
    return [choice(_M, _m, replace=False) for _ in range(nensemble)]


def eval_weak(_X, _y, _nensemble, _nfeatures):
    acc = []
    for j in range(_nensemble):
        # Keep subset features, columns same for a 10-fold
        cols = features_randomsubset(_X.shape[1], _nfeatures, nensemble=1)
        # 10-fold CV
        kf = StratifiedKFold(n_splits=10, shuffle=False, random_state=None)
        results, names = list(), list()
        for train_index, test_index in kf.split(_X, _y):
            clf = weak_fit(cols[0], _X[train_index], _y[train_index])
            y_pred, y_prob = weak_predict(clf, cols[0], _X[test_index])
            acc += [accuracy_score(_y[test_index], y_pred)]
    return np.array(acc)

# evaluate the models and store results
results, names = list(), list()
scores = eval_weak(X.todense(), y, 100, 5)

追溯

> --------------------------------------------------------------------------- AttributeError                            Traceback (most recent call
> last) /tmp/ipykernel_4467/1067286612.py in <module>
>       1 # evaluate the models and store results
>       2 results, names = list(), list()
> ----> 3 scores = eval_weak(X.todense(), y, 100, 5)
>       4 for name, model in scores.clf.items():
>       5     results.append(scores)
> 
> <timed exec> in eval_weak(_X, _y, _nensemble, _nfeatures)
> 
> /tmp/ipykernel_4467/3891534381.py in weak_predict(_clf, _list_cols,
> _X)
>      15 def weak_predict(_clf, _list_cols, _X):
>      16     Xs = _X[:, _list_cols]
> ---> 17     return _clf.predict(Xs), _clf.predict_proba(Xs)
>      18 
>      19 # Use _m features randomly selected, a total of M weak learners
> 
> AttributeError: 'tuple' object has no attribute 'predict'