RL_quant_trader/data_manager.py at main · wonjun-TIL/RL_quant_trader · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
# import os
import pandas as pd
import numpy as np
# from sklearn.preprocessing import StandardScaler

# import settings


COLUMNS_CHART_DATA = ['date', 'open', 'high', 'low', 'close', 'volume']

# 차트 데이터에서 전처리로 얻을 수 있는 자질
COLUMNS_TRAINING_DATA_V1 = [
    'open_lastclose_ratio', 'high_close_ratio', 'low_close_ratio',
    'close_lastclose_ratio', 'volume_lastvolume_ratio',
    'close_ma5_ratio', 'volume_ma5_ratio',
    'close_ma10_ratio', 'volume_ma10_ratio',
    'close_ma20_ratio', 'volume_ma20_ratio',
    'close_ma60_ratio', 'volume_ma60_ratio',
    'close_ma120_ratio', 'volume_ma120_ratio',
]

# COLUMNS_TRAINING_DATA_V1_RICH = [
#     'open_lastclose_ratio', 'high_close_ratio', 'low_close_ratio',
#     'close_lastclose_ratio', 'volume_lastvolume_ratio',
#     'close_ma5_ratio', 'volume_ma5_ratio',
#     'close_ma10_ratio', 'volume_ma10_ratio',
#     'close_ma20_ratio', 'volume_ma20_ratio',
#     'close_ma60_ratio', 'volume_ma60_ratio',
#     'close_ma120_ratio', 'volume_ma120_ratio',
#     'inst_lastinst_ratio', 'frgn_lastfrgn_ratio',
#     'inst_ma5_ratio', 'frgn_ma5_ratio',
#     'inst_ma10_ratio', 'frgn_ma10_ratio',
#     'inst_ma20_ratio', 'frgn_ma20_ratio',
#     'inst_ma60_ratio', 'frgn_ma60_ratio',
#     'inst_ma120_ratio', 'frgn_ma120_ratio',
# ]

# 차트 데이터 외에 기본적 분석 지표 per, pbr, roe, 코스피지수, 국채 3년 데이터
COLUMNS_TRAINING_DATA_V2 = [
    'per', 'pbr', 'roe',
    'open_lastclose_ratio', 'high_close_ratio', 'low_close_ratio',
    'close_lastclose_ratio', 'volume_lastvolume_ratio',
    'close_ma5_ratio', 'volume_ma5_ratio',
    'close_ma10_ratio', 'volume_ma10_ratio',
    'close_ma20_ratio', 'volume_ma20_ratio',
    'close_ma60_ratio', 'volume_ma60_ratio',
    'close_ma120_ratio', 'volume_ma120_ratio',
    'market_kospi_ma5_ratio', 'market_kospi_ma20_ratio',
    'market_kospi_ma60_ratio', 'market_kospi_ma120_ratio',
    'bond_k3y_ma5_ratio', 'bond_k3y_ma20_ratio',
    'bond_k3y_ma60_ratio', 'bond_k3y_ma120_ratio'
]

# COLUMNS_TRAINING_DATA_V3 = [
#     'per', 'pbr', 'roe',
#     'open_lastclose_ratio', 'high_close_ratio', 'low_close_ratio',
#     'diffratio', 'volume_lastvolume_ratio',
#     'close_ma5_ratio', 'volume_ma5_ratio',
#     'close_ma10_ratio', 'volume_ma10_ratio',
#     'close_ma20_ratio', 'volume_ma20_ratio',
#     'close_ma60_ratio', 'volume_ma60_ratio',
#     'close_ma120_ratio', 'volume_ma120_ratio',
#     'market_kospi_ma5_ratio', 'market_kospi_ma20_ratio',
#     'market_kospi_ma60_ratio', 'market_kospi_ma120_ratio',
#     'bond_k3y_ma5_ratio', 'bond_k3y_ma20_ratio',
#     'bond_k3y_ma60_ratio', 'bond_k3y_ma120_ratio',
#     'ind', 'ind_diff', 'ind_ma5', 'ind_ma10', 'ind_ma20', 'ind_ma60', 'ind_ma120',
#     'inst', 'inst_diff', 'inst_ma5', 'inst_ma10', 'inst_ma20', 'inst_ma60', 'inst_ma120',
#     'foreign', 'foreign_diff', 'foreign_ma5', 'foreign_ma10', 'foreign_ma20', 'foreign_ma60', 'foreign_ma120',
# ]터

def preprocess(data):
    windows = [5, 10, 20, 60, 120]
    for window in windows:
        data['close_ma{}'.format(window)] = data['close'].rolling(window).mean()
        data['volume_ma{}'.format(window)] = data['volume'].rolling(window).mean()
        data['close_ma%d_ratio' % window] = (data['close'] - data['close_ma%d' % window]) / data['close_ma%d' % window]
        data['volume_ma%d_ratio' % window] = (data['volume'] - data['volume_ma%d' % window]) / data['volume_ma%d' % window]

        # if ver == 'v1.rich':
        #     data['inst_ma{}'.format(window)] = data['close'].rolling(window).mean()
        #     data['frgn_ma{}'.format(window)] = data['volume'].rolling(window).mean()
        #     data['inst_ma%d_ratio' % window] = (data['close'] - data['inst_ma%d' % window]) / data['inst_ma%d' % window]
        #     data['frgn_ma%d_ratio' % window] = (data['volume'] - data['frgn_ma%d' % window]) / data['frgn_ma%d' % window]

    data['open_lastclose_ratio'] = np.zeros(len(data))
    data.loc[1:, 'open_lastclose_ratio'] = (data['open'][1:].values - data['close'][:-1].values) / data['close'][:-1].values
    data['high_close_ratio'] = (data['high'].values - data['close'].values) / data['close'].values
    data['low_close_ratio'] = (data['low'].values - data['close'].values) / data['close'].values
    data['close_lastclose_ratio'] = np.zeros(len(data))
    data.loc[1:, 'close_lastclose_ratio'] = (data['close'][1:].values - data['close'][:-1].values) / data['close'][:-1].values
    data['volume_lastvolume_ratio'] = np.zeros(len(data))
    data.loc[1:, 'volume_lastvolume_ratio'] = (
        (data['volume'][1:].values - data['volume'][:-1].values)
        / data['volume'][:-1].replace(to_replace=0, method='ffill').replace(to_replace=0, method='bfill').values
    )

    # if ver == 'v1.rich':
    #     data['inst_lastinst_ratio'] = np.zeros(len(data))
    #     data.loc[1:, 'inst_lastinst_ratio'] = (
    #         (data['inst'][1:].values - data['inst'][:-1].values)
    #         / data['inst'][:-1].replace(to_replace=0, method='ffill').replace(to_replace=0, method='bfill').values
    #     )
    #     data['frgn_lastfrgn_ratio'] = np.zeros(len(data))
    #     data.loc[1:, 'frgn_lastfrgn_ratio'] = (
    #         (data['frgn'][1:].values - data['frgn'][:-1].values)
    #         / data['frgn'][:-1].replace(to_replace=0, method='ffill').replace(to_replace=0, method='bfill').values
    #     )

    return data


def load_data(fpath, date_from, date_to, ver='v2'):
    # if ver == 'v3':
    #     return load_data_v3(code, date_from, date_to)

    header = None if ver == 'v1' else 0
    data = pd.read_csv(fpath, thouusands=',', header=header, converters={'date': lambda x: str(x)})


    # if ver == 'v1':
    #     data.columns = ['date', 'open', 'high', 'low', 'close', 'volume']

    # # 날짜 오름차순 정렬
    # data = data.sort_values(by='date').reset_index()

    # 데이터 전처리
    data = preprocess(data)

    # 기간 필터링
    data['date'] = data['date'].str.replace('-', '')
    data = data[(data['date'] >= date_from) & (data['date'] <= date_to)]
    data = data.dropna()

    # 차트 데이터 분리
    chart_data = data[COLUMNS_CHART_DATA]

    # 학습 데이터 분리
    training_data = None
    if ver == 'v1':
        training_data = data[COLUMNS_TRAINING_DATA_V1]
    # elif ver == 'v1.rich':
    #     training_data = data[COLUMNS_TRAINING_DATA_V1_RICH]
    elif ver == 'v2':
        data.loc[:, ['per', 'pbr', 'roe']] = data[['per', 'pbr', 'roe']].apply(lambda x: x / 100)
        training_data = data[COLUMNS_TRAINING_DATA_V2]
        training_data = training_data.apply(np.tanh)
    else:
        raise Exception('Invalid version.')

    return chart_data, training_data


# def load_data_v3(code, date_from, date_to):
#     df = None
#     for filename in os.listdir('D:\\dev\\rltrader\\data\\v3'):
#         if filename.startswith(code):
#             df = pd.read_csv(os.path.join('D:\\dev\\rltrader\\data\\v3', filename), thousands=',', header=0, converters={'date': lambda x: str(x)})
#             break

#     # 날짜 오름차순 정렬
#     df = df.sort_values(by='date').reset_index()

#     # 표준화
#     scaler = StandardScaler()
#     scaler.fit(df[COLUMNS_TRAINING_DATA_V3].dropna().values)

#     # 기간 필터링
#     df['date'] = df['date'].str.replace('-', '')
#     df = df[(df['date'] >= date_from) & (df['date'] <= date_to)]
#     df = df.dropna()

#     # 차트 데이터 분리
#     chart_data = df[COLUMNS_CHART_DATA]

#     # 학습 데이터 분리
#     training_data = df[COLUMNS_TRAINING_DATA_V3]
#     training_data = pd.DataFrame(scaler.transform(training_data.values), columns=COLUMNS_TRAINING_DATA_V3)

#     return chart_data, training_data