def run_dbn_classification(root_path, train_symbols, predict_symbols, need_training, need_plot_training_diagram, need_predict):
paras = SP_Paras('dbn', root_path, train_symbols, predict_symbols)
paras.save = True
paras.load = False
paras.plot = need_plot_training_diagram
# 0_index: no norm 1_index: standard norm 2_index: minmax norm 3_index: zscore norm
paras.features = {'0_0':['frac_change', 'frac_high', 'frac_low'],
'3_0':['volume']}
paras.window_len = 2
paras.pred_len = 1
paras.valid_len = 20
paras.start_date = '2016-01-03'
paras.end_date = datetime.datetime.now().strftime("%Y-%m-%d")
paras.verbose = 1
paras.batch_size = 64
paras.epoch = 100
paras.model['hidden_layers'] = [256, 128, 64]
paras.model['dropout'] = 0.2
paras.model['activation'] = 'relu'
paras.model['optimizer'] = 'adam'
paras.out_class_type = 'classification'
paras.n_out_class = 7
paras.model['out_layer'] = paras.n_out_class
paras.model['loss'] = 'categorical_crossentropy'
paras.model['out_activation'] = 'softmax'
dbn_cla = dbn_classification(paras)
dbn_cla.run(need_training, need_predict)
return paras
def run_recommand_system(root_path, train_symbols, predict_symbols, need_training, need_plot_training_diagram, need_predict):
paras = SP_Paras('recommandSystem', root_path, train_symbols, predict_symbols)
paras.save = True
paras.load = False
paras.run_hyperopt = True
paras.plot = need_plot_training_diagram
paras.features = {'1_0_0':['week_day'],
'1_0_1':['c_2_o', 'h_2_o', 'l_2_o', 'c_2_h', 'h_2_l', 'vol_p'],
'1_1_0':['buy_amount', 'sell_amount', 'even_amount'],
'1_1_1':['buy_volume', 'sell_volume', 'even_volume'],
'1_1_2':['buy_max', 'buy_min', 'buy_average', 'sell_max', 'sell_min', 'sell_average', 'even_max', 'even_min', 'even_average']}
paras.window_len = [3]
paras.pred_len = 1
paras.valid_len = 20
paras.start_date = '2016-11-01'
paras.end_date = datetime.datetime.now().strftime("%Y-%m-%d")
paras.verbose = 0
paras.batch_size = 64
paras.epoch = 200
paras.out_class_type = 'classification'
paras.n_out_class = 7 # ignore for regression
rs = recommand_system(paras)
rs.run(need_training, need_predict)
def run_xgboost_classification(root_path, need_training,
need_plot_training_diagram, need_predict):
df = getStocksList_CHN(root_path)
df.index = df.index.astype(str).str.zfill(6)
df = df.sort_index(ascending=True)
predict_symbols = df.index.values.tolist()
paras = SP_Paras('xgboost', root_path, predict_symbols, predict_symbols)
paras.save = True
paras.load = False
paras.run_hyperopt = True
paras.plot = need_plot_training_diagram
# A_B_C format:
# A: require window split or not -> 0 for not, 1 for yes
# B: normalization method -> 0: none 1: standard 2: minmax 3: zscore
# C: normalization index, same normalization requires different index
paras.features = {
'0_0_0': ['week_day'],
'1_0_1': ['c_2_o', 'h_2_o', 'l_2_o', 'c_2_h', 'h_2_l', 'vol_p'],
'1_1_0': ['buy_amount', 'sell_amount', 'even_amount'],
'1_1_1': ['buy_volume', 'sell_volume', 'even_volume'],
'1_1_2': [
'buy_max', 'buy_min', 'buy_average', 'sell_max', 'sell_min',
'sell_average', 'even_max', 'even_min', 'even_average'
]
}
paras.window_len = [3]
paras.pred_len = 1
paras.valid_len = 20
paras.start_date = '2016-11-01'
paras.end_date = datetime.datetime.now().strftime("%Y-%m-%d")
paras.verbose = 1
paras.batch_size = 64
paras.epoch = 5000
paras.out_class_type = 'classification'
paras.n_out_class = 7 # ignore for regression
from hyperopt import hp
paras.hyper_opt = {
"max_depth": hp.randint("max_depth", 10),
"n_estimators": hp.randint("n_estimators",
20), #[0,1,2,3,4,5] -> [50,]
"gamma": hp.randint("gamma", 4), #0-0.4
"learning_rate": hp.randint("learning_rate",
6), #[0,1,2,3,4,5] -> 0.05,0.06
"subsample": hp.randint("subsample",
4), #[0,1,2,3] -> [0.7,0.8,0.9,1.0]
"min_child_weight": hp.randint("min_child_weight", 5),
}
# run
xgboost_cla = xgboost_classification(paras)
xgboost_cla.run(need_training, need_predict)
return paras
def run_lstm_classification(root_path, train_symbols, predict_symbols,
need_training, need_plot_training_diagram,
need_predict):
paras = SP_Paras('lstm', root_path, train_symbols, predict_symbols)
paras.save = True
paras.load = False
paras.plot = need_plot_training_diagram
# 0_index: no norm 1_index: standard norm 2_index: minmax norm 3_index: zscore norm
paras.features = {
'0_0': ['frac_change', 'frac_high', 'frac_low'],
'3_0': ['c_2_o', 'h_2_o', 'l_2_o', 'c_2_h', 'h_2_l', 'vol'],
'3_0': ['volume']
}
#paras.features = [['top', 'middle', 'bottom'], ['volume'], ['vol_stat'], ['close_-5_r', 'close_-10_r', 'close_-20_r', 'close_-60_r']]
#paras.window_len = [2, 4, 9]
paras.window_len = [44]
paras.pred_len = 1
paras.valid_len = 20
paras.start_date = '2012-01-03'
paras.end_date = datetime.datetime.now().strftime("%Y-%m-%d")
paras.verbose = 1
paras.batch_size = 64
paras.epoch = 100
#paras.model['hidden_layers'] = [[256, 128, 64], [256, 128, 64], [256, 128, 64]]
paras.model['hidden_layers'] = [[256, 128, 64]]
#paras.model['dropout'] = [[0.7, 0.5, 0.3], [0.6, 0.5, 0.4], [0.6, 0.5, 0.4]]
paras.model['dropout'] = [[0.7, 0.5, 0.3]]
#paras.model['activation'] = [['relu', 'relu', 'relu'], ['relu', 'relu', 'relu'], ['relu', 'relu', 'relu']]
paras.model['activation'] = [['relu', 'relu', 'relu']]
#paras.model['optimizer'] = ['adam', 'adam', 'adam']
paras.model['optimizer'] = ['adam']
paras.out_class_type = 'classification'
paras.n_out_class = 7 # ignore for regression
paras.model['out_layer'] = paras.n_out_class
paras.model['loss'] = 'categorical_crossentropy'
paras.model['out_activation'] = 'softmax'
# run
lstm_cla = rnn_lstm_classification(paras)
lstm_cla.run(need_training, need_predict)
return paras
def run_recommand_system(root_path, train_symbols, predict_symbols,
need_training, need_plot_training_diagram,
need_predict):
paras = SP_Paras('recommandSystem', root_path, train_symbols,
predict_symbols)
paras.save = True
paras.load = False
paras.run_hyperopt = True
paras.plot = need_plot_training_diagram
# A_B_C format:
# A: require window split or not -> 0 for not, 1 for yes
# B: normalization method -> 0: none 1: standard 2: minmax 3: zscore
# C: normalization index, same normalization requires different index
paras.features = {
'1_0_0': ['week_day'],
'1_0_1': ['c_2_o', 'h_2_o', 'l_2_o', 'c_2_h', 'h_2_l', 'vol_p'],
'1_1_0': ['buy_amount', 'sell_amount', 'even_amount'],
'1_1_1': ['buy_volume', 'sell_volume', 'even_volume'],
'1_1_2': [
'buy_max', 'buy_min', 'buy_average', 'sell_max', 'sell_min',
'sell_average', 'even_max', 'even_min', 'even_average'
]
}
paras.window_len = [3]
paras.pred_len = 1
paras.valid_len = 20
paras.start_date = '2016-11-01'
paras.end_date = datetime.datetime.now().strftime("%Y-%m-%d")
paras.verbose = 0
paras.batch_size = 64
paras.epoch = 200
paras.out_class_type = 'classification'
paras.n_out_class = 7 # ignore for regression
# run
rs = recommand_system(paras)
rs.run(need_training, need_predict)
def run_lstm_classification(root_path, need_training, need_plot_training_diagram, need_predict):
df = getStocksList_CHN(root_path)
df.index = df.index.astype(str).str.zfill(6)
df = df.sort_index(ascending = True)
predict_symbols = df.index.values.tolist()
paras = SP_Paras('lstm', root_path, predict_symbols, predict_symbols)
paras.save = True
paras.load = False
paras.run_hyperopt = True
paras.plot = need_plot_training_diagram
# A_B_C format:
# A: require window split or not -> 0 for not, 1 for yes
# B: normalization method -> 0: none 1: standard 2: minmax 3: zscore
# C: normalization index, same normalization requires different index
paras.features = {'1_0_0':['week_day'], # 1
'1_0_1':['c_2_o', 'h_2_o', 'l_2_o', 'c_2_h', 'h_2_l', 'vol_p'], # 5
'1_1_0':['buy_amount', 'sell_amount', 'even_amount'], # 3
'1_1_1':['buy_volume', 'sell_volume', 'even_volume'], # 3
'1_1_2':['buy_max', 'buy_min', 'buy_average', 'sell_max', 'sell_min', 'sell_average', 'even_max', 'even_min', 'even_average']} # 9
paras.pred_len = 1
paras.valid_len = 20
paras.start_date = '2016-11-01'
paras.end_date = datetime.datetime.now().strftime("%Y-%m-%d")
paras.verbose = 1
paras.out_class_type = 'classification'
paras.n_out_class = 7 # ignore for regression
paras.epoch = 1000
paras.window_len = [5]
paras.batch_size = 64
paras.model['hidden_layers'] = [210, 140, 210, 140, 210, 140, 70]
paras.model['dropout'] = 0.5
paras.model['activation'] = 'relu'
paras.model['optimizer'] = 'adam'
paras.model['learning_rate'] = 0.01
paras.model['out_activation'] = 'softmax'
paras.model['out_layer'] = paras.n_out_class
paras.model['loss'] = 'categorical_crossentropy'
from hyperopt import hp
paras.hyper_opt = {"batch_size_opt" :[32, 64, 128],
"activation_opt" :['relu', 'tanh', 'sigmoid'],
"optimizer_opt" :['sgd', 'rmsprop', 'adagrad', 'adam'],
}
paras.hyper_opt.update({
"batch_size" :hp.choice ("batch_size" , paras.hyper_opt['batch_size_opt'] ),
"dropout" :hp.uniform("dropout" , 0.3, 0.7 ),
"learning_rate" :hp.uniform("learning_rate", 0.005, 0.02 ),
"activation" :hp.choice ("activation" , paras.hyper_opt['activation_opt']),
"optimizer" :hp.choice ("optimizer" , paras.hyper_opt['optimizer_opt']),
})
lstm_cla = rnn_lstm_classification(paras)
lstm_cla.run(need_training, need_predict)
return paras
