TPOT：多类数据分类失败

Question

我无法让 TPot（v. 0.9.2，Python 2.7）处理多类数据（尽管我在 TPot 的文档中找不到任何内容说它只进行二进制分类）。

下面提供了一个示例。它运行到 9%，然后因错误而死机：

RuntimeError: There was an error in the TPOT optimization process. 
This could be because the data was not formatted properly, or because
data for a regression problem was provided to the TPOTClassifier 
object. Please make sure you passed the data to TPOT correctly.

但是把 n_classes 改成 2 就可以正常运行了。

from sklearn.metrics import f1_score, make_scorer
from sklearn.datasets import make_classification
from tpot import TPOTClassifier

scorer = make_scorer(f1_score)
X, y = make_classification(n_samples=200, n_features=100,
                           n_informative=20, n_redundant=10,
                           n_classes=3, random_state=42)
tpot = TPOTClassifier(generations=10, population_size=20, verbosity=20, scoring=scorer)
tpot.fit(X, y)

Answer 1

事实上，TPOT 也应该适用于多类数据 - example in the docs 与 MNIST 数据集（10 个类）一起使用。

错误与f1_score有关；将您的代码保存在n_classes=3，并要求

tpot = TPOTClassifier(generations=10, population_size=20, verbosity=2)

（即使用默认的scoring='accuracy'）可以正常工作：

Warning: xgboost.XGBClassifier is not available and will not be used by TPOT.

Generation 1 - Current best internal CV score: 0.7447422496202984                                                                                
Generation 2 - Current best internal CV score: 0.7447422496202984                                                                                  
Generation 3 - Current best internal CV score: 0.7454927186634503                                                                                   
Generation 4 - Current best internal CV score: 0.7454927186634503             
Generation 5 - Current best internal CV score: 0.7706334316090413
Generation 6 - Current best internal CV score: 0.7706334316090413
Generation 7 - Current best internal CV score: 0.7706334316090413
Generation 8 - Current best internal CV score: 0.7706334316090413
Generation 9 - Current best internal CV score: 0.7757616367372464
Generation 10 - Current best internal CV score: 0.7808898418654516

Best pipeline: 

LogisticRegression(KNeighborsClassifier(DecisionTreeClassifier(input_matrix, criterion=entropy, max_depth=3, min_samples_leaf=15, min_samples_split=12), n_neighbors=6, p=2, weights=uniform), C=0.01, dual=False, penalty=l2)

TPOTClassifier(config_dict={'sklearn.linear_model.LogisticRegression': {'penalty': ['l1', 'l2'], 'C': [0.0001, 0.001, 0.01, 0.1, 0.5, 1.0, 5.0, 10.0, 15.0, 20.0, 25.0], 'dual': [True, False]}, 'sklearn.decomposition.PCA': {'iterated_power': range(1, 11), 'svd_solver': ['randomized']}, 'sklearn.feature_selection.Se...ocessing.PolynomialFeatures': {'degree': [2], 'interaction_only': [False], 'include_bias': [False]}},
        crossover_rate=0.1, cv=5, disable_update_check=False,
        early_stop=None, generations=10, max_eval_time_mins=5,
        max_time_mins=None, memory=None, mutation_rate=0.9, n_jobs=1,
        offspring_size=20, periodic_checkpoint_folder=None,
        population_size=20, random_state=None, scoring=None, subsample=1.0,
        verbosity=2, warm_start=False)

使用suggested in the docs求F1分数，即：

tpot = TPOTClassifier(generations=10, population_size=20, verbosity=2, scoring='f1')

再次产生你报告的错误，可能是因为f1_score中的default argument是average='binary'，这对于多类问题确实没有意义，而简单的f1仅适用于二进制问题（@ 987654324@).

在scoring 中明确使用 F1 分数的其他一些变体，例如f1_macro、f1_micro 或 f1_weighted 工作正常（未显示）。