Fine-tuning parameters in Logistic Regression

Question

I am running a logistic regression with a tf-idf being ran on a text column. This is the only column I use in my logistic regression. How can I ensure the parameters for this are tuned as well as possible?

I would like to be able to run through a set of steps which would ultimately allow me say that my Logistic Regression classifier is running as well as it possibly can.

from sklearn import metrics,preprocessing,cross_validation
from sklearn.feature_extraction.text import TfidfVectorizer
import sklearn.linear_model as lm
import pandas as p
loadData = lambda f: np.genfromtxt(open(f, 'r'), delimiter=' ')

print "loading data.."
traindata = list(np.array(p.read_table('train.tsv'))[:, 2])
testdata = list(np.array(p.read_table('test.tsv'))[:, 2])
y = np.array(p.read_table('train.tsv'))[:, -1]

tfv = TfidfVectorizer(min_df=3, max_features=None, strip_accents='unicode',
                      analyzer='word', token_pattern=r'\w{1,}', 
                      ngram_range=(1, 2), use_idf=1, smooth_idf=1, 
                      sublinear_tf=1)

rd = lm.LogisticRegression(penalty='l2', dual=True, tol=0.0001, 
                           C=1, fit_intercept=True, intercept_scaling=1.0, 
                           class_weight=None, random_state=None)

X_all = traindata + testdata
lentrain = len(traindata)

print "fitting pipeline"
tfv.fit(X_all)
print "transforming data"
X_all = tfv.transform(X_all)

X = X_all[:lentrain]
X_test = X_all[lentrain:]

print "20 Fold CV Score: ", np.mean(cross_validation.cross_val_score(rd, X, y, cv=20, scoring='roc_auc'))

print "training on full data"
rd.fit(X, y)
pred = rd.predict_proba(X_test)[:, 1]
testfile = p.read_csv('test.tsv', sep="\t", na_values=['?'], index_col=1)
pred_df = p.DataFrame(pred, index=testfile.index, columns=['label'])
pred_df.to_csv('benchmark.csv')
print "submission file created.."

Answer 1

You can use grid search to find out the best C value for you. Basically smaller C specify stronger regularization.

>>> param_grid = {'C': [0.001, 0.01, 0.1, 1, 10, 100, 1000] }
>>> clf = GridSearchCV(LogisticRegression(penalty='l2'), param_grid)
GridSearchCV(cv=None,
             estimator=LogisticRegression(C=1.0, intercept_scaling=1,   
               dual=False, fit_intercept=True, penalty='l2', tol=0.0001),
             param_grid={'C': [0.001, 0.01, 0.1, 1, 10, 100, 1000]})

See the GridSearchCv document for more details on your application.

Answer 2

You may use below code for more general details:

LR = LogisticRegression()
LRparam_grid = {
    'C': [0.001, 0.01, 0.1, 1, 10, 100, 1000],
    'penalty': ['l1', 'l2'],
    # 'max_iter': list(range(100,800,100)),
    'solver': ['newton-cg', 'lbfgs', 'liblinear', 'sag', 'saga']
}
LR_search = GridSearchCV(LR, param_grid=LRparam_grid, refit = True, verbose = 3, cv=5)

# fitting the model for grid search 
LR_search.fit(X_train , y_train)
LR_search.best_params_
# summarize
print('Mean Accuracy: %.3f' % LR_search.best_score_)
print('Config: %s' % LR_search.best_params_)

Answer 3

Grid search is a brutal way of finding the optimal parameters because it train and test every possible combination. best way is using bayesian optimization which learns for past evaluation score and takes less computation time.