🔥알림🔥
① 테디노트 유튜브 - 구경하러 가기!
② LangChain 한국어 튜토리얼 바로가기 👀
③ 랭체인 노트 무료 전자책(wikidocs) 바로가기 🙌
④ RAG 비법노트 LangChain 강의오픈 바로가기 🙌
⑤ 서울대 PyTorch 딥러닝 강의 바로가기 🙌

8 분 소요

경진대회에서 모델의 Hyperparameter 튜닝에 드는 노력과 시간을 절약하기 위하여 xgboost, lightgbm, catboost 3개의 라이브러리에 대하여 optuna 튜닝을 적용하여 예측 값을 산출해 내는 로직을 라이브러리 형태로 패키징 했습니다.

경진대회 BASELINE을 잡기 위한 optuna + [xgboost, lightgbm, catboost]

지원하는 예측 종류는

  • 회귀(regression)
  • 이진분류(binary classification)
  • 다중분류(multi-class classification)

입니다.

앞으로 라이브러리 개선작업을 통해 더 빠르게 최적화할 수 있도록 개선해 나갈 계획입니다.

Google Colab 예제

설치

!pip install -U teddynote
# 모듈 import 
from teddynote import models

샘플 데이터셋 로드

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import warnings

from sklearn.datasets import load_iris, load_boston, load_breast_cancer
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error

import lightgbm as lgb
import xgboost as xgb
import catboost as cb

from lightgbm import LGBMRegressor, LGBMClassifier
from xgboost import XGBRegressor, XGBClassifier
from catboost import CatBoostRegressor, CatBoostClassifier

warnings.filterwarnings('ignore')

SEED = 2021
# Binary Class Datasets
cancer = load_breast_cancer()
cancer_df = pd.DataFrame(cancer['data'], columns=cancer['feature_names'])
cancer_df['target'] = cancer['target']
cancer_df.head()
mean radius mean texture mean perimeter mean area mean smoothness mean compactness mean concavity mean concave points mean symmetry mean fractal dimension ... worst texture worst perimeter worst area worst smoothness worst compactness worst concavity worst concave points worst symmetry worst fractal dimension target
0 17.99 10.38 122.80 1001.0 0.11840 0.27760 0.3001 0.14710 0.2419 0.07871 ... 17.33 184.60 2019.0 0.1622 0.6656 0.7119 0.2654 0.4601 0.11890 0
1 20.57 17.77 132.90 1326.0 0.08474 0.07864 0.0869 0.07017 0.1812 0.05667 ... 23.41 158.80 1956.0 0.1238 0.1866 0.2416 0.1860 0.2750 0.08902 0
2 19.69 21.25 130.00 1203.0 0.10960 0.15990 0.1974 0.12790 0.2069 0.05999 ... 25.53 152.50 1709.0 0.1444 0.4245 0.4504 0.2430 0.3613 0.08758 0
3 11.42 20.38 77.58 386.1 0.14250 0.28390 0.2414 0.10520 0.2597 0.09744 ... 26.50 98.87 567.7 0.2098 0.8663 0.6869 0.2575 0.6638 0.17300 0
4 20.29 14.34 135.10 1297.0 0.10030 0.13280 0.1980 0.10430 0.1809 0.05883 ... 16.67 152.20 1575.0 0.1374 0.2050 0.4000 0.1625 0.2364 0.07678 0

5 rows × 31 columns

# Multi Class Datasets
iris = load_iris()
iris_df = pd.DataFrame(iris['data'], columns=iris['feature_names'])
iris_df['target'] = iris['target']
iris_df.head()
sepal length (cm) sepal width (cm) petal length (cm) petal width (cm) target
0 5.1 3.5 1.4 0.2 0
1 4.9 3.0 1.4 0.2 0
2 4.7 3.2 1.3 0.2 0
3 4.6 3.1 1.5 0.2 0
4 5.0 3.6 1.4 0.2 0
# Regression Datasets
boston = load_boston()
boston_df = pd.DataFrame(boston['data'], columns=boston['feature_names'])
boston_df['target'] = boston['target']
boston_df.head()
CRIM ZN INDUS CHAS NOX RM AGE DIS RAD TAX PTRATIO B LSTAT target
0 0.00632 18.0 2.31 0.0 0.538 6.575 65.2 4.0900 1.0 296.0 15.3 396.90 4.98 24.0
1 0.02731 0.0 7.07 0.0 0.469 6.421 78.9 4.9671 2.0 242.0 17.8 396.90 9.14 21.6
2 0.02729 0.0 7.07 0.0 0.469 7.185 61.1 4.9671 2.0 242.0 17.8 392.83 4.03 34.7
3 0.03237 0.0 2.18 0.0 0.458 6.998 45.8 6.0622 3.0 222.0 18.7 394.63 2.94 33.4
4 0.06905 0.0 2.18 0.0 0.458 7.147 54.2 6.0622 3.0 222.0 18.7 396.90 5.33 36.2

간단 사용법

optimize()


optimize(
    x,
    y,
    test_data=None,
    cat_features=None,
    eval_metric='f1',
    cv=5,
    seed=None,
    n_rounds=3000,
    n_trials=100,
)

입력 매개변수

  • x: Feature 데이터

  • y: Target 데이터

  • test_data: 예측 데이터 (test 데이터의 feature 데이터)

  • cat_features: 카테고리형 컬럼

  • eval_metric: 최적화할 메트릭 (‘f1’, ‘accuracy’, ‘recall’, ‘precision’, ‘mse’, ‘rmse’, ‘rmsle’)

  • cv: cross validation fold 개수

  • seed: 시드

  • n_rounds: 학습시 최대 iteration 횟수

  • n_trials: optuna 하이퍼파라미터 튜닝 시도 횟수

return

  • params: best 하이퍼파라미터

  • preds: test_data 매개변수에 데이터를 지정한 경우 이에 대한 예측 값

결과값 자동저장 기능

optimizer() 로 튜닝 + 예측한 결과는 numpy array 형식으로 자동 저장합니다.

  • 저장 경로: models 폴더

CatBoost + Optuna

이진분류(binary classification)

catboostoptuna = models.CatBoostClassifierOptuna(use_gpu=False)

params, preds = catboostoptuna.optimize(iris_df.drop('target', 1), 
                                        iris_df['target'], 
                                        test_data=iris_df.drop('target', 1),
                                        seed=321,
                                        eval_metric='recall', n_trials=3)

(np.squeeze(preds) == iris_df['target']).mean()

다중분류(multi-class classification)

catboostoptuna = models.CatBoostClassifierOptuna()

params, preds = catboostoptuna.optimize(cancer_df.drop('target', 1), 
                                        cancer_df['target'], 
                                        test_data=cancer_df.drop('target', 1),
                                        seed=321,
                                        eval_metric='recall', n_trials=3)

(np.squeeze(preds) == cancer_df['target']).mean()

회귀(regression)

for col in ['CHAS', 'RAD', 'ZN']:
    boston_df[col] = boston_df[col].astype('int')
    
catboostoptuna_reg = models.CatBoostRegressorOptuna(use_gpu=False)
        
params, preds = catboostoptuna_reg.optimize(boston_df.drop('target', 1), 
                                            boston_df['target'], 
                                            test_data=boston_df.drop('target', 1),
                                            # int, str 타입 이어야 한다. float는 허용하지 않음
                                            cat_features=['CHAS', 'RAD', 'ZN'],
                                            eval_metric='rmse', n_trials=3)

mean_squared_error(boston_df['target'], preds)

저장한 파일로부터 예측 값 (prediction) 불러오기

# 넘파이 array로 저장된 예측 결과를 로드할 수 있습니다.
models.load_prediction_from_file('models/CatBoostRegressor-0.87226.npy')
array([28.76168717, 21.97469764, 33.91423778, 36.25317587, 34.89063327,
       30.36592103, 21.62216055, 21.67925239, 16.13744104, 18.55325167,
       19.19114681, 20.39200754, 20.4851822 , 20.18992755, 19.53332984,
       19.95333642, 21.90489984, 17.43089179, 19.28067897, 18.75662987,
       13.27196673, 17.43784243, 16.54706384, 14.6641719 , 15.98753177,
       14.55392224, 16.07654144, 15.15717605, 18.46626451, 19.14277466,
       13.42305337, 16.08031477, 13.88189742, 15.42776871, 13.93921857,
       21.76392661, 20.96989226, 22.05468906, 23.01954988, 30.38238318,
       35.86242058, 27.86047868, 24.46726054, 24.81965244, 21.93943212,
       20.97466042, 19.40221015, 17.69406529, 14.9838815 , 18.13447461,
       20.67717755, 22.8037252 , 27.00797342, 22.50247985, 18.38411847,
       34.59829013, 25.66841753, 34.24698692, 23.25394891, 20.56548366,
       18.08249006, 15.98063826, 23.92861079, 24.18700647, 31.40892682,
       25.84846134, 19.79075001, 20.42532382, 18.62746158, 20.80022091,
       23.33964199, 21.12974915, 22.61460761, 23.59166397, 23.77889816,
       22.56759439, 20.36601319, 21.90976033, 20.22201645, 20.79508199,
       26.11453205, 23.92772329, 24.00113404, 22.73930675, 23.72290053,
       26.13084916, 21.3290855 , 24.25999642, 28.84149579, 30.07779444,
       25.15881721, 25.06305693, 23.06077441, 24.28432971, 20.45356778,
       27.53869606, 25.76019718, 39.81858701, 41.84946703, 34.55262548,
       24.02980442, 25.65445863, 20.71075989, 20.2305375 , 19.82863103,
       16.79407357, 17.75558312, 18.35677353, 19.65660082, 18.0508654 ,
       21.5689087 , 23.06746945, 18.09671889, 18.42912872, 21.51601955,
       17.52104662, 21.19149237, 21.53741825, 18.60982263, 19.8353869 ,
       21.78394929, 20.82479114, 19.52986522, 16.34316023, 19.12320234,
       20.51817182, 15.84022485, 14.82196509, 17.39604813, 15.71638676,
       19.23800906, 19.38498764, 20.40981153, 17.61717099, 15.42971604,
       17.70507511, 16.60916137, 18.13055393, 12.97506937, 16.57567281,
       15.20248486, 14.23643501, 13.36728264, 14.72898506, 13.55049322,
       14.9755267 , 16.03064387, 14.98655171, 15.15773699, 15.42038297,
       19.80594139, 18.72855912, 16.47855863, 16.13431554, 17.39118588,
       18.95159967, 16.80484215, 36.44802584, 27.31847993, 24.42104749,
       33.32831742, 48.17558575, 48.10500758, 50.10220898, 21.6445183 ,
       24.25216144, 46.87931622, 22.3243491 , 23.54260782, 23.23930775,
       20.11795752, 20.55089244, 21.96288645, 25.19124179, 23.64260695,
       28.62909465, 24.51323548, 24.80628451, 30.42589992, 37.3450949 ,
       38.4181347 , 30.23559046, 38.05903337, 31.71650322, 26.26537372,
       28.50830185, 45.52756479, 30.25326372, 30.14612204, 32.84831893,
       29.96189048, 29.83769357, 35.86649294, 31.06581569, 29.52652706,
       47.51160949, 36.3245743 , 29.81679975, 33.0799321 , 32.54773576,
       34.19546892, 25.76115073, 41.09042432, 48.59485047, 48.31403964,
       21.86601009, 22.41873571, 19.98276713, 23.41414041, 18.99296122,
       20.94222734, 19.85599395, 22.46012421, 23.23118424, 22.33826844,
       24.35427176, 22.84754784, 26.78443297, 22.02070077, 25.00887251,
       29.11967995, 21.35142102, 28.4845406 , 28.13266249, 43.55809324,
       43.92528527, 41.41070297, 33.03739803, 43.98911308, 34.88425066,
       23.95836477, 35.73025736, 43.16032433, 43.04690079, 27.64386148,
       24.70308422, 25.4884944 , 37.47134395, 24.75982658, 24.46682263,
       24.01787678, 20.55522458, 21.81256414, 25.1210334 , 18.84767363,
       18.65530061, 23.31803976, 20.90605542, 23.2340228 , 25.26291264,
       24.24455301, 29.6144702 , 31.89752872, 40.56324634, 23.53199626,
       19.96194628, 40.74456114, 48.1240842 , 36.47112068, 31.95761281,
       35.22692603, 41.08284384, 43.59715004, 32.69785587, 38.395269  ,
       29.9962736 , 31.91708957, 40.03140873, 46.20098888, 21.80400282,
       20.18597404, 25.8267357 , 25.12578985, 35.62999617, 33.24609222,
       32.1246326 , 32.1276531 , 33.20775978, 26.22963653, 35.73998533,
       44.73786966, 36.20293923, 46.15322327, 50.76547855, 31.24164863,
       24.72412287, 20.28457479, 22.92577672, 22.80344633, 23.12369478,
       31.17422305, 35.1362746 , 26.74748661, 23.54393337, 22.04051644,
       26.52810061, 25.09489421, 19.48346588, 26.88560752, 32.68850504,
       27.88899813, 23.20912832, 24.87111363, 32.28653345, 33.91915569,
       28.80376445, 35.01262449, 29.21634021, 25.79724389, 22.05887928,
       21.3824104 , 23.18025825, 20.00625817, 23.00639761, 22.75589549,
       17.94145995, 18.06389734, 18.02561674, 22.79710775, 20.21203719,
       23.45896878, 23.04214565, 22.94954438, 18.56859949, 24.6756368 ,
       25.56719479, 24.084167  , 20.2480269 , 20.91046936, 24.25174757,
       21.23424581, 16.49834945, 19.66746909, 21.4557298 , 21.65723782,
       20.06222367, 19.22627356, 18.87736706, 20.04211261, 19.71010834,
       19.77379091, 32.5851722 , 24.65104583, 24.97530297, 30.10281989,
       19.2971915 , 17.20332405, 23.85532316, 25.69790452, 26.13244964,
       23.82816976, 24.07280403, 22.60158478, 31.12142614, 21.94114302,
       23.35171169, 18.76796541, 18.91663659, 23.34380777, 20.45138255,
       22.45630745, 17.74251749, 21.83864988, 18.43330579, 23.54134213,
       26.68331628, 19.25077592, 22.754818  , 38.81018216, 49.26446287,
       50.79271301, 39.70994055, 47.02221074, 12.5120863 , 13.44238312,
       16.03211618, 14.14478663, 12.77288871, 11.2105293 , 12.32364567,
       12.25072646, 10.34790415, 10.53523456, 10.7030971 , 11.42738056,
        8.88825104,  9.34983733,  9.12203832, 11.97584357, 11.17106292,
       14.56801675, 16.83874389, 10.02438036, 13.9523593 , 13.26336838,
       13.90404049, 12.99194972, 11.10259056,  7.3404075 ,  9.94517994,
        7.68810727, 10.44760646, 12.67179266, 10.04602704,  9.2217173 ,
        9.49584146, 18.29746443, 26.3683067 , 16.04251865, 20.71751946,
       27.65855805, 16.86487185, 14.54875179, 15.70205646,  8.96022422,
        9.18930385, 10.93764513,  9.52636606,  9.02340857,  9.99152837,
       15.03190469, 14.90398513, 17.58508063, 13.3390705 , 13.65630505,
        9.17099086, 13.15485696, 11.95532373, 11.0969807 , 10.75318325,
       15.2116026 , 15.75387832, 18.8499195 , 14.90319747, 13.90606762,
       11.24903563, 12.25064494, 10.36012359,  9.54165187, 11.81341828,
       10.84391803, 13.36466617, 15.57695542, 14.26152587, 10.23491255,
       10.70045996, 14.70878743, 14.79437752, 14.33805741, 13.99509979,
       13.91139729, 15.20039866, 16.1576856 , 18.10186486, 13.92352322,
       14.85029811, 14.2714253 , 15.14535685, 16.80094021, 18.69923671,
       16.33460786, 17.54055978, 19.44845554, 20.11537833, 21.70623678,
       19.35919475, 16.46347524, 15.96509858, 17.28791419, 19.37198404,
       20.14604739, 20.67226853, 21.52497603, 22.75428627, 16.2353488 ,
       15.04075596, 17.64296268, 13.079737  , 16.40755067, 20.24207015,
       22.28444162, 23.89979801, 25.88883777, 21.55801563, 20.88892453,
       22.86809854, 20.05936591, 21.45465288, 15.16904835,  9.95506956,
       10.62074076, 14.6694633 , 19.00900648, 21.3145516 , 22.52886541,
       21.62849666, 18.76080567, 21.24361604, 21.34419004, 18.83961699,
       20.54947447, 21.35132386, 17.93859277, 23.9005454 , 22.68316054,
       17.21506894])

하이퍼파라미터 튜닝 시각화

# 튜닝 결과 시각화
catboostoptuna_reg.visualize()

capture-20211231-161430

capture-20211231-161439

XGBoost

이진분류(binary classification)

xgboptuna = models.XGBClassifierOptuna(use_gpu=False)
        
params, preds = xgboptuna.optimize(iris_df.drop('target', 1), 
                                   iris_df['target'], 
                                   test_data=iris_df.drop('target', 1),
                                   seed=321,
                                   eval_metric='recall', n_trials=3)

(preds == iris_df['target']).mean()

다중분류(multi-class classification)

xgboptuna_binary = models.XGBClassifierOptuna(use_gpu=False)
        
params, preds = xgboptuna_binary.optimize(cancer_df.drop('target', 1), 
                                          cancer_df['target'], 
                                          test_data=cancer_df.drop('target', 1), 
                                          eval_metric='accuracy', n_trials=3)

(preds == cancer_df['target']).mean()

회귀(regression)

xgboptuna_reg = models.XGBRegressorOptuna()
        
params, preds = xgboptuna_reg.optimize(boston_df.drop('target', 1), 
                                       boston_df['target'], 
                                       test_data=boston_df.drop('target', 1), 
                                       eval_metric='mse', n_trials=3)

mean_squared_error(boston_df['target'], preds)

LGBM

이진분류(binary classification)

lgbmoptuna_binary = models.LGBMClassifierOptuna()
        
params, preds = lgbmoptuna_binary.optimize(cancer_df.drop('target', 1), 
                                           cancer_df['target'], 
                                           test_data=cancer_df.drop('target', 1),
                                           eval_metric='accuracy', n_trials=3)

(preds == cancer_df['target']).mean()

다중분류(multi-class classification)

lgbmoptuna = models.LGBMClassifierOptuna()
        
params, preds = lgbmoptuna.optimize(iris_df.drop('target', 1), 
                    iris_df['target'], 
                    seed=321,
                    eval_metric='recall', n_trials=3)


(preds == iris_df['target']).mean()

회귀(regression)

lgbmoptuna_reg = models.LGBMRegressorOptuna()
        
params, preds = lgbmoptuna_reg.optimize(boston_df.drop('target', 1), 
                                        boston_df['target'], 
                                        test_data=boston_df.drop('target', 1), 
                                        eval_metric='mse', n_trials=3)

mean_squared_error(boston_df['target'], preds)

하이퍼파라미터 범위 수정 (custom)

lgbmoptuna = models.LGBMRegressorOptuna()

# 기본 값으로 설정된 하이퍼파라미터 출력
lgbmoptuna.print_params()
name: verbose, fixed_value: -1, type: fixed
name: lambda_l1, low: 1e-08, high: 5, type: loguniform
name: lambda_l2, low: 1e-08, high: 5, type: loguniform
name: path_smooth, low: 1e-08, high: 0.001, type: loguniform
name: learning_rate, low: 1e-05, high: 0.1, type: loguniform
name: feature_fraction, low: 0.5, high: 0.9, type: uniform
name: bagging_fraction, low: 0.5, high: 0.9, type: uniform
name: num_leaves, low: 15, high: 90, type: int
name: min_data_in_leaf, low: 10, high: 100, type: int
name: max_bin, low: 100, high: 255, type: int
name: n_estimators, low: 100, high: 3000, type: int
name: bagging_freq, low: 0, high: 15, type: int
name: min_child_weight, low: 1, high: 20, type: int

param_type에 관하여

param_typeint, uniform, loguniform, categorical, fixed 가 있습니다.

  • int, uniform, loguniform은 optuna의 search range 정의하는 파라미터와 같습니다.

예시)

- int 범위(int)

lgbmoptuna.set_param(models.OptunaParam('num_leaves', low=10, high=25, param_type='int'))



- 카테고리(categorical)

cboptuna.set_param(models.OptunaParam('bootstrap_type', categorical_value=['Bayesian', 'Bernoulli', 'MVS'], param_type='categorical'))



- 고정된 값(fixed)

cboptuna.set_param(models.OptunaParam('one_hot_max_size', fixed_value=1024, param_type='fixed'))

# 하이퍼파라미터 범위 정의
lgbmoptuna.set_param(models.OptunaParam('num_leaves', low=10, high=25, param_type='int'))
lgbmoptuna.set_param(models.OptunaParam('n_estimators', low=0, high=500, param_type='int'))
# 출력
lgbmoptuna.print_params()
name: verbose, fixed_value: -1, type: fixed
name: lambda_l1, low: 1e-08, high: 5, type: loguniform
name: lambda_l2, low: 1e-08, high: 5, type: loguniform
name: path_smooth, low: 1e-08, high: 0.001, type: loguniform
name: learning_rate, low: 1e-05, high: 0.1, type: loguniform
name: feature_fraction, low: 0.5, high: 0.9, type: uniform
name: bagging_fraction, low: 0.5, high: 0.9, type: uniform
name: num_leaves, low: 10, high: 25, type: int
name: min_data_in_leaf, low: 10, high: 100, type: int
name: max_bin, low: 100, high: 255, type: int
name: n_estimators, low: 0, high: 500, type: int
name: bagging_freq, low: 0, high: 15, type: int
name: min_child_weight, low: 1, high: 20, type: int
# 달라진 결과값 확인
params, preds = lgbmoptuna.optimize(boston_df.drop('target', 1), 
                                    boston_df['target'], 
                                    test_data=boston_df.drop('target', 1), 
                                    eval_metric='mse', n_trials=3)

trial에 대한 결과를 출력합니다.

# trial에 대한 결과 출력
lgbmoptuna.study.trials_dataframe()
number value datetime_start datetime_complete duration params_bagging_fraction params_bagging_freq params_feature_fraction params_lambda_l1 params_lambda_l2 params_learning_rate params_max_bin params_min_child_weight params_min_data_in_leaf params_n_estimators params_num_leaves params_path_smooth state
0 0 83.529143 2021-12-31 07:26:21.337195 2021-12-31 07:26:24.492885 0 days 00:00:03.155690 0.508583 9 0.650294 6.415572e-07 4.982401e-02 0.000030 205 6 22 363 14 4.464380e-08 COMPLETE
1 1 82.696414 2021-12-31 07:26:24.494212 2021-12-31 07:26:26.673476 0 days 00:00:02.179264 0.712115 12 0.703372 4.705313e-08 6.194418e-08 0.000061 254 17 39 323 22 2.821343e-08 COMPLETE
2 2 20.680800 2021-12-31 07:26:26.674743 2021-12-31 07:26:28.133769 0 days 00:00:01.459026 0.857708 11 0.580214 1.124464e-02 5.298364e-06 0.091827 230 20 76 485 20 2.478903e-05 COMPLETE

하이퍼파라미터 튜닝 결과 시각화

lgbmoptuna.visualize()

capture-20211231-161620

capture-20211231-161630

Best 하이퍼파라미터 출력

lgbmoptuna.get_best_params()
{'lambda_l1': 0.011244644026182967,
 'lambda_l2': 5.298363992080463e-06,
 'path_smooth': 2.4789027860002685e-05,
 'learning_rate': 0.09182657994717408,
 'feature_fraction': 0.5802144206891808,
 'bagging_fraction': 0.8577082120277062,
 'num_leaves': 20,
 'min_data_in_leaf': 76,
 'max_bin': 230,
 'n_estimators': 485,
 'bagging_freq': 11,
 'min_child_weight': 20}

댓글남기기