def pick_best_params(model, X, y):
opt_scores = []
for C in np.arange(5.26, 6, 0.1):
for p in (False, ):
#model = linear_model.LogisticRegression(C=C, penalty=p) # the classifier we'll use
model.C = C
model.dual = p
# Train and Crossvalidate
scores, _ = my_cross_val_score(model, X, y, auc_roc, 3)
#print str(scores)
print "Mean AUC for p=%s, C=%f: %f" % (p, C, sp.mean(scores))
opt_scores.append((sp.mean(scores), p, C))
opt_scores = sorted(opt_scores, reverse=True)
print opt_scores
return opt_scores[0][-1]
def pick_best_params(model, X, y):
"""
Pick best params
"""
opt_scores = []
for C in np.arange(0.5, 4, 0.3):
for p in (True, ):
model.C = C
model.dual = p
# Train and Crossvalidate
scores, _ = my_cross_val_score(model, X, y, auc_roc, 3)
#print str(scores)
logging.debug("Mean AUC for p=%s, C=%f: %f" % (p, C, sp.mean(scores)))
opt_scores.append((sp.mean(scores), p, C))
opt_scores = sorted(opt_scores, reverse=True)
logging.debug(opt_scores)
return opt_scores[0][-1]
stop_val=feature_sel_stop,
kfold=4,
random_state=SEED*78,
features_sel=preselected,
remove_worst=feature_remove_worst)
X = X[:, features]
X_test = X_test[:, features]
logging.debug(X.shape)
logging.info("Encoding factors/transform data")
X, X_test = encode_factors(X, X_test)
logging.info("Find best params for model (C)")
model.C = pick_best_params(model, X, y)
# Get cross_val scores
logging.info("Cross-validate")
scores, _ = my_cross_val_score(model, X, y, auc_roc, 10, random_state=SEED*31)
#print str(scores)
logging.info("Mean AUC: %f" % (sp.mean(scores)))
# Train with whole dataset
logging.info("Train with the whole dataset")
model.fit(X, y)
preds = model.predict_proba(X_test)[:, 1]
# Save results
logging.info("Save results")
save_results(preds, "new_sal.csv")
