def train(self):
"""
## Auto Train with Logs of Boost Round ##
Model Name:
'lr': Logistic Regression
'rf': Random Forest
'et': Extra Trees
'gb': GradientBoosting
'xgb': XGBoost
'xgb_sk': XGBoost using scikit-learn module
'lgb': LightGBM
'lgb_sk': LightGBM using scikit-learn module
'cb': CatBoost
"""
TM = TrainingMode()
"""
Training Arguments
"""
train_args = {
'use_global_valid': False,
'use_custom_obj': False,
'show_importance': False,
'save_final_pred': True,
'save_final_pred_train': False,
'save_cv_pred': True,
'save_cv_pred_train': False,
'save_csv_log': True,
'loss_fuc': self.rmse,
'append_info': 'Yuanan Bike'
}
"""
Cross Validation Arguments
"""
cv_args = {'n_cv': 10}
"""
Base Parameters
"""
base_parameters = self.get_base_params('dnn')
"""
Auto Train with Logs of Boost Round
"""
pg_list = [[['learning_rate', [0.05]]]]
train_seed_list = [68]
cv_seed_list = [95]
TM.auto_train_boost_round('dnn',
num_boost_round=10,
n_epoch=1,
full_grid_search=True,
train_seed_list=train_seed_list,
cv_seed_list=cv_seed_list,
base_parameters=base_parameters,
parameter_grid_list=pg_list,
save_final_pred=True,
train_args=train_args,
cv_args=cv_args)
"""Train Different Rounds"""
def train(self):
"""
Model Name:
'lr': Logistic Regression
'rf': Random Forest
'et': Extra Trees
'ab': AdaBoost
'gb': GradientBoosting
'xgb': XGBoost
'xgb_sk': XGBoost using scikit-learn module
'lgb': LightGBM
'lgb_sk': LightGBM using scikit-learn module
'cb': CatBoost
'dnn': Deep Neural Networks
'stack_lgb': LightGBM for stack layer
'christar': Christar1991
'prejudge_b': PrejudgeBinary
'prejudge_m': PrejudgeMultiClass
'stack_t': StackTree
"""
TM = TrainingMode()
"""
Global Seed
"""
train_seed = random.randint(0, 500)
cv_seed = random.randint(0, 500)
# train_seed = 666
# cv_seed = 216 # 425 48 461 157
"""
Training Arguments
"""
train_args = {'train_seed': train_seed,
'prescale': False,
'postscale': False,
'use_scale_pos_weight': False,
'use_global_valid': False,
'use_custom_obj': False,
'show_importance': False,
'show_accuracy': True,
'save_final_pred': True,
'save_final_prob_train': False,
'save_cv_pred': False,
'save_cv_prob_train': False,
'save_csv_log': True,
'append_info': None}
"""
Cross Validation Arguments
"""
cv_args = {'n_valid': 4,
'n_cv': 20,
'n_era': 20}
# cv_args = self.get_cv_args('lgb_fi')
"""
Reduced Features
"""
reduced_feature_list = None
"""
Base Parameters
"""
base_parameters = self.get_base_params('xgb')
# base_parameters = None
"""
Train Single Model
"""
# TM.train_single_model('xgb', train_seed, cv_seed, num_boost_round=88,
# reduced_feature_list=reduced_feature_list, base_parameters=base_parameters,
# train_args=train_args, use_multi_group=False)
"""
Auto Train with Logs of Boost Round
"""
pg_list = [
[['learning_rate', [0.003]]]
]
# train_seed_list = [666]
# cv_seed_list = [216]
train_seed_list = None
cv_seed_list = None
TM.auto_train_boost_round('xgb', num_boost_round=100, grid_search_n_cv=20, n_epoch=100, full_grid_search=False,
use_multi_group=True, train_seed_list=train_seed_list, cv_seed_list=cv_seed_list,
base_parameters=base_parameters, parameter_grid_list=pg_list, save_final_pred=True,
reduced_feature_list=reduced_feature_list, train_args=train_args, cv_args=cv_args)
"""
Auto Grid Search Parameters
"""
# pg_list = [
# [['max_depth', (8, 9, 10, 11, 12)],
# ['feature_fraction', (0.5, 0.6, 0.7, 0.8, 0.9)],
# ['bagging_fraction', (0.6, 0.7, 0.8, 0.9)],
# ['bagging_freq', (1, 3, 5, 7)]]
# ]
# train_seed_list = [999]
# cv_seed_list = [888]
# # train_seed_list = None
# # cv_seed_list = None
# TM.auto_grid_search('lgb', num_boost_round=65, grid_search_n_cv=5, n_epoch=1, use_multi_group=True,
# full_grid_search=True, train_seed_list=train_seed_list, cv_seed_list=cv_seed_list,
# parameter_grid_list=pg_list, base_parameters=base_parameters, save_final_pred=False,
# reduced_feature_list=reduced_feature_list, train_args=train_args, cv_args=cv_args)
"""
Auto Train
"""
# TM.auto_train('lgb', n_epoch=10000, base_parameters=base_parameters,
# reduced_feature_list=reduced_feature_list, train_args=train_args, use_multi_group=False)
"""
Others
"""
# TM.train_single_model('dnn', train_seed, cv_seed, reduced_feature_list=reduced_feature_list,
# base_parameters=base_parameters, train_args=train_args, use_multi_group=False)
# TM.train_single_model('prejudge_b', train_seed, cv_seed, load_pickle=False,
# base_parameters=base_parameters, reduced_feature_list=reduced_feature_list,
# train_args=train_args, use_multi_group=False)
# TM.train_single_model('stack_lgb', train_seed, cv_seed, auto_idx=1,
# base_parameters=base_parameters, reduced_feature_list=reduced_feature_list,
# train_args=train_args, use_multi_group=False)
# TM.train_single_model('stack_pt', train_seed, cv_seed, reduced_feature_list=reduced_feature_list,
# base_parameters=base_parameters, train_args=train_args, use_multi_group=False)
# pg_list = [
# [['learning_rate', [0.00005]]],
# [['keep_probability', [0.4, 0.5, 0.6, 0.7, 0.8, 0.9]]],
# # [['unit_number',
# # [
# # [32, 16, 8],
# # [48, 24, 12],
# # [64, 32], [64, 32, 16],
# # [128, 64], [128, 64, 32], [128, 64, 32, 16],
# # [256, 128], [256, 128, 64], [256, 128, 64, 32], [256, 128, 64, 32, 16],
# # [200, 100, 50],
# # [2048, 512],
# # [288, 144, 72], [288, 144, 72, 36],
# # [216, 108, 54], [216, 108, 54, 27],
# # [128, 256, 128, 64], [64, 128, 64, 32], [128, 256, 128], [64, 128, 64]
# # ]]]
# ]
# train_seed_list = [666]
# cv_seed_list = [216]
# TM.auto_train_boost_round('dnn', train_seed_list, cv_seed_list, n_epoch=1, base_parameters=base_parameters,
# epochs=2, parameter_grid_list=pg_list, save_final_pred=True,
# reduced_feature_list=reduced_feature_list, grid_search_n_cv=20,
# train_args=train_args, use_multi_group=False)
# TM.auto_train('stack_t', n_epoch=2, stack_final_epochs=10, base_parameters=base_parameters,
# reduced_feature_list=reduced_feature_list, train_args=train_args, use_multi_group=False)
print('======================================================')
print('Global Train Seed: {}'.format(train_seed))
print('Global Cross Validation Seed: {}'.format(cv_seed))
def train(self):
"""
## Auto Train with Logs of Boost Round ##
Model Name:
'lr': Logistic Regression
'rf': Random Forest
'et': Extra Trees
'gb': GradientBoosting
'xgb': XGBoost
'xgb_sk': XGBoost using scikit-learn module
'lgb': LightGBM
'lgb_sk': LightGBM using scikit-learn module
'cb': CatBoost
"""
TM = TrainingMode()
"""
Training Arguments
"""
train_args = {
'use_global_valid': False,
'use_custom_obj': False,
'show_importance': False,
'save_final_pred': True,
'save_final_pred_train': False,
'save_cv_pred': False,
'save_cv_pred_train': False,
'save_csv_log': True,
'loss_fuc': self.rmse,
'append_info': 'Yuanan Bike'
}
"""
Cross Validation Arguments
"""
cv_args = {'n_cv': 10}
# cv_args = self.get_cv_args('xgb')
"""
Base Parameters
"""
# base_parameters = self.get_base_params('xgb')
base_parameters = None
"""
Auto Train with Logs of Boost Round
"""
pg_list = [
# [
# ['n_cv', [5,7,9]]
# ['valid_rate', valid_rate_list]
# ],
[
['n_cv', [5, 7, 9]],
['max_depth', [12, 13, 14]], # best 14
['learning_rate', [0.15, 0.2, 0.25]], #best 0.15
#['min_child_weight', [9]],
['subsample', [0.75, 0.8]], #best0.8
['colsample_bytree', [0.7]], #best 0.7
['colsample_bylevel', [0.77]], #best0.77
]
]
train_seed_list = None
cv_seed_list = None
TM.auto_train_boost_round('xgb',
num_boost_round=600,
n_epoch=3,
full_grid_search=True,
train_seed_list=train_seed_list,
cv_seed_list=cv_seed_list,
base_parameters=base_parameters,
parameter_grid_list=pg_list,
save_final_pred=True,
train_args=train_args,
cv_args=cv_args)
"""Train Different Rounds"""
def train(self):
"""
## Auto Train with Logs of Boost Round ##
Model Name:
'lr': Logistic Regression
'rf': Random Forest
'et': Extra Trees
'ab': AdaBoost
'gb': GradientBoosting
'xgb': XGBoost
'xgb_sk': XGBoost using scikit-learn module
'lgb': LightGBM
'lgb_sk': LightGBM using scikit-learn module
'cb': CatBoost
'dnn': Deep Neural Networks
'stack_lgb': LightGBM for stack layer
'christar': Christar1991
'prejudge_b': PrejudgeBinary
'prejudge_m': PrejudgeMultiClass
'stack_t': StackTree
"""
TM = TrainingMode()
"""
Strategy Arguments
"""
from models.strategy import Strategy
strategy_args = {'use_strategy': True,
'f_strategy': Strategy.calc_profit,
'buy_count': 15,
'fee': 0.0034,
'ascending': False,
'save_result': True,
'save_image': True}
"""
Training Arguments
"""
train_args = {'prescale': False,
'postscale': True,
'use_scale_pos_weight': False,
'use_global_valid': False,
'use_custom_obj': False,
'show_importance': False,
'show_accuracy': True,
'save_final_pred': True,
'save_final_prob_train': False,
'save_cv_pred': True,
'save_cv_prob_train': False,
'save_csv_log': True,
'strategy_args': strategy_args,
'append_info': 'forward_window'}
"""
Cross Validation Arguments
"""
from models.cross_validation import CrossValidation
cv_args = {'ensemble': True,
'n_cv': 12,
'n_era': 90,
'cv_generator': CrossValidation.forward_window,
'window_size': 40}
# cv_args = self.get_cv_args('xgb_fw')
"""
Reduced Features
"""
reduced_feature_list = None
"""
Base Parameters
"""
base_parameters = self.get_base_params('xgb_fw')
# base_parameters = None
"""
Auto Train with Logs of Boost Round
"""
# cv_weights_range = [self.get_cv_weight('range', 1, i+1) for i in [5, 8, 10, 12, 15, 20]]
# cv_weights_log = [self.get_cv_weight('log', 1, i+1) for i in [5, 8, 10, 12, 15, 20]]
# n_cv_list = [5, 8, 10, 12, 15, 20] * 6
# import numpy as np
# valid_rate_list = np.array([[i]*6 for i in [0.075, 0.1, 0.125, 0.15, 0.175, 0.2]]).reshape(-1,).tolist()
# cv_weights_list = cv_weights_log*6
pg_list = [
# [['n_cv', (8, 9, 10), (11, 12, 13), [15], (18, 20)],
# ['valid_rate', (0.075, 0.1, 0.125, 0.15, 0.166, 0.175, 0.2)],
# ['window_size', (32, 34, 36, 40, 42, 44, 46, 48)]]
# [['n_cv', n_cv_list],
# ['valid_rate', valid_rate_list],
# ['cv_weights', cv_weights_list]]
# [
# ['max_depth', [11]],
# ['min_child_weight', [9]],
# ['subsample', (0.88, 0.90, 0.92)],
# ['colsample_bytree', (0.86, 0.88, 0.9)],
# ['colsample_bylevel', (0.7, 0.75, 0.8)]
# ]
[['learning_rate', [0.003]]]
]
train_seed_list = [999]
cv_seed_list = [95]
TM.auto_train_boost_round('xgb', num_boost_round=100, n_epoch=1, full_grid_search=True,
use_multi_group=False, train_seed_list=train_seed_list, cv_seed_list=cv_seed_list,
base_parameters=base_parameters, parameter_grid_list=pg_list, save_final_pred=False,
reduced_feature_list=reduced_feature_list, train_args=train_args, cv_args=cv_args)
def train(self):
"""
Model Name:
'lr': Logistic Regression
'rf': Random Forest
'ab': AdaBoost
'gb': GradientBoosting
'xgb': XGBoost
'xgb_sk': XGBoost using scikit-learn module
'lgb': LightGBM
'lgb_sk': LightGBM using scikit-learn module
'cb': CatBoost
"""
TM = TrainingMode()
"""
Global Seed
"""
train_seed = random.randint(0, 500)
cv_seed = random.randint(0, 500)
# train_seed = 666
# cv_seed = 216 # 425 48 461 157
"""
Training Arguments
"""
train_args = {
'use_global_valid': False,
'use_custom_obj': False,
'show_importance': False,
'save_final_pred': True,
'save_final_pred_train': False,
'save_cv_pred': False,
'save_cv_pred_train': False,
'save_csv_log': True,
'loss_fuc': None,
'append_info': 'forward_window_postscale_mdp-11_sub'
}
"""
Cross Validation Arguments
"""
# cv_args = {'n_splits': 10,
# 'n_cv': 10}
cv_args = self.get_cv_args('xgb')
"""
Base Parameters
"""
# base_parameters = self.get_base_params('xgb')
base_parameters = None
"""
Train Single Model
"""
# TM.train_single_model('xgb', train_seed, cv_seed, num_boost_round=88,
# reduced_feature_list=reduced_feature_list, base_parameters=base_parameters,
# train_args=train_args, use_multi_group=False)
"""
Auto Train with Logs of Boost Round
"""
pg_list = [
# [
# ['n_cv', n_cv_list],
# ['valid_rate', valid_rate_list]
# ]
[['max_depth', [11]], ['min_child_weight', [9]],
['subsample', (0.88, 0.90, 0.92)],
['colsample_bytree', (0.86, 0.88, 0.9)],
['colsample_bylevel', (0.7, 0.75, 0.8)]]
# [['learning_rate', [0.003]]]
]
train_seed_list = [999]
cv_seed_list = [95]
TM.auto_train_boost_round('xgb',
num_boost_round=100,
n_epoch=1,
full_grid_search=True,
train_seed_list=train_seed_list,
cv_seed_list=cv_seed_list,
base_parameters=base_parameters,
parameter_grid_list=pg_list,
save_final_pred=False,
train_args=train_args,
cv_args=cv_args)
"""Train Different Rounds"""
# num_boost_round_list = [83, 85, 87]
# self.train_diff_round('xgb', TM, num_boost_round_list=num_boost_round_list, n_epoch=1, full_grid_search=True,
# train_seed_list=train_seed_list, cv_seed_list=cv_seed_list,
# base_parameters=base_parameters, parameter_grid_list=pg_list, save_final_pred=True,
# train_args=train_args, cv_args=cv_args)
"""
Auto Grid Search Parameters
"""
# pg_list = [
# [
# ['max_depth', (8, 9, 10)],
# ['min_child_weight', (2, 4, 6, 8)],
# ['subsample', (0.81, 0.84, 0.87, 0.9)],
# ['colsample_bytree', (0.8, 0.85, 0.9)],
# ['colsample_bylevel', (0.7, 0.75, 0.8)],
# ]
# ]
# train_seed_list = [999]
# cv_seed_list = [95]
# # train_seed_list = None
# # cv_seed_list = None
# TM.auto_grid_search('xgb', num_boost_round=95, n_epoch=1, full_grid_search=True,
# train_seed_list=train_seed_list, cv_seed_list=cv_seed_list,
# parameter_grid_list=pg_list, base_parameters=base_parameters, save_final_pred=False,
# train_args=train_args, cv_args=cv_args)
print('======================================================')
print('Global Train Seed: {}'.format(train_seed))
print('Global Cross Validation Seed: {}'.format(cv_seed))