def xgb_cv(X,
y,
params,
cv,
states=1,
pipe=None,
early_stopping_rounds=20,
verbose=False):
importance = []
n = cv.get_n_splits() * states
scores = [[] for _ in range(cv.get_n_splits())]
iterations = [[] for _ in range(cv.get_n_splits())]
clf = XGBClassifier(**params, tree_method="gpu_hist")
with tqdm(total=n, ncols=50) as pbar:
for i, (train_index, test_index) in enumerate(cv.split(X)):
X_train, y_train = X.iloc[train_index], y.iloc[train_index]
X_test, y_test = X.iloc[test_index], y.iloc[test_index]
if pipe is not None:
X_train = pipe.fit_transform(X_train, y_train, X_test=X_test)
X_test = pipe.transform(X_test)
for state in range(states):
clf.random_state = state
clf.fit(X_train,
y_train,
eval_set=[(X_test, y_test)],
eval_metric="auc",
early_stopping_rounds=early_stopping_rounds,
verbose=verbose)
iterations[i].append(clf.best_iteration)
scores[i].append(clf.best_score)
importance.append([i, state] +
clf.feature_importances_.tolist())
pbar.update(1)
importance = pd.DataFrame(importance,
columns=["fold", "state"] +
X_train.columns.tolist())
# sort columns from most to least important
importance = importance[importance.mean().sort_values(
ascending=False).index]
importance = importance.set_index(["fold", "state"])
return {
"importance": importance,
"scores": scores,
"iterations": iterations
}
