ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.

[CBrauer]

When I run the following code:

from sklearn.ensemble import GradientBoostingRegressor
from sklearn.kernel_approximation import Nystroem
from sklearn.linear_model import ElasticNetCV, RidgeCV
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsRegressor
from sklearn.pipeline import make_pipeline, make_union
from sklearn.svm import LinearSVR
from tpot.builtins import StackingEstimator
from xgboost import XGBRegressor

# Average CV score on the training set was:-0.21141374399237495
exported_pipeline = make_pipeline(
    StackingEstimator(estimator=GradientBoostingRegressor(alpha=0.8, 
                                                          learning_rate=1.0,
                                                          loss="quantile",
                                                          max_depth=8,
                                                          max_features=0.7000000000000001,
                                                          min_samples_leaf=16,
                                                          min_samples_split=5,
                                                          n_estimators=100,
                                                          subsample=0.6500000000000001)),
    StackingEstimator(estimator=XGBRegressor(learning_rate=0.01,
                                             max_depth=5,
                                             min_child_weight=10,
                                             n_estimators=500,
                                             nthread=1, 
                                             subsample=0.55)),
    StackingEstimator(estimator=LinearSVR(C=1.0,
                                          dual=True,
                                          epsilon=0.1,
                                          loss="epsilon_insensitive",
                                          tol=0.01)),
    StackingEstimator(estimator=RidgeCV()),
    StackingEstimator(estimator=LinearSVR(C=20.0,
                                          dual=True,
                                          epsilon=0.0001,
                                          loss="epsilon_insensitive",
                                          tol=0.001)),
    Nystroem(gamma=0.15000000000000002,
             kernel="laplacian",
             n_components=6),
    StackingEstimator(estimator=ElasticNetCV(l1_ratio=0.05,
                                             tol=0.01)),
    KNeighborsRegressor(n_neighbors=91,
                        p=2,
                        weights="distance")
)
exported_pipeline.fit(X_train, y_train)
score = exported_pipeline.score(X_test, y_test)
print('\nScore: ', score)

I get:

ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.

Any help in making this warning message go away will be greatly appreciated.

Charles

[weixuanfu]

import warnings
with warnings.catch_warnings():
    warnings.simplefilter('ignore')
    exported_pipeline.fit(X_train, y_train)

Using warnings module may help.

[CBrauer]

I'm nervous about suppressing warnings. Should I trust the results anyway? One advice I got from the Web was to normalize the data. Can I simply put “RobustScaler()” as the first line to the pipe?

[weixuanfu]

I think StandardScaler() or RobustScaler() may avoid the warning but the pipeline may not work very well on the normalized data since the pipeline was evaluated on the raw data.

[bluexie]

标准化未实现，遇到相同的问题，已经解决，标准化一下训练集和测试集就好了。 Standardization is not implemented, encountered the same problem has been solved, standardized training and test sets look just fine.

[aghaPathan]

I encountered the same problem. I used GridSearchCV and gave a random array of possible increased number of iterations and according adjusted my array. Finally at some point it converged at a value.