Best models for mlp

best_models.py

import re

import nltk
from nltk.tokenize.regexp import RegexpTokenizer
from nltk.corpus import stopwords

from sklearn.svm import SVC
from sklearn.naive_bayes import BernoulliNB
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis as LDA
from sklearn.feature_selection import SelectPercentile
from sklearn.feature_selection import chi2
from sklearn.multiclass import OneVsRestClassifier
from sklearn.decomposition import TruncatedSVD
from sklearn.pipeline import make_pipeline, Pipeline
from sklearn.feature_extraction.text import TfidfVectorizer

from xgboost.sklearn import XGBModel

# some monkeypatching to get probability estimates from xgboost
def xgb_pred_proba(self, X):
    pred = self.predict(X).reshape(-1, 1)
    return np.hstack([np.zeros(len(pred)).reshape(-1, 1), pred])

XGBModel.predict_proba = xgb_pred_proba

__tokenizer = RegexpTokenizer(r"\w+")
__stop = set(stopwords.words("german"))
__word_regex = re.compile("\D+")

def _preprocess_doc(doc):
    return [w for w in __tokenizer.tokenize(doc.lower()) if w not in __stop and re.fullmatch(__word_regex, w)]

features = {}
models = {}

features['mlp_00'] = \
    make_pipeline(TfidfVectorizer(
                        analyzer='char_wb',
                        strip_accents='unicode',
                        tokenizer=_preprocess_doc),
                  SelectPercentile(
                       chi2, 
                       percentile=90)
                  )
models['mlp_00'] = {
    'BNB': BernoulliNB(alpha=0.118),
    'DTREE': OneVsRestClassifier(XGBModel(nthread=1,
                                     objective="binary:logistic",
                                     colsample_bytree=0.5,
                                     learning_rate=0.1,
                                     max_depth=20,
                                     min_child_weight=1),
                 n_jobs=-1
            ),
    'LDA':Pipeline([
                ("reduction", TruncatedSVD(n_components=2000)),
                ("lda", LDA(n_components=13)),
                ("clf", SVC(kernel="rbf", C=5))
            ])
}