wrong time estimation for model training

[pplonski]

when running training for long time (4h) the AutoML is spending only half of the time, because it don't want to train more models because of time limits

[pplonski]

sample code to reproduce:

train = pd.read_csv("../otto/train.csv")
X_cols = [f for f in train.columns if "feat" in f]

automl = AutoML(algorithms=["Baseline", "Decision Tree", "Random Forest", "LightGBM", "Xgboost"],
                total_time_limit=4*60*60,
                tuning_mode="Sport")

automl.fit(train[X_cols], train["target"])

Data is from Kaggle Otto competition. AutoML instead use of 14400 seconds is using~7200 secons.