時系列分析 - データ準備編(BTC価格予測)

何をしているか

• 時系列予測の題材として、BTCの価格予測をする
• それに際して、ひとまずどこかからデータを引っ張ってくる必要がある
• ひとまず、binanceのデータを活用して、データの調達をした

作業詳細

データ取得

仮想通貨取引所の大手であるbinanceを公開してくれている。 www.binance.com

ここから、データを引っ張ってくる。

# --
import time
import datetime
from datetime import datetime as dt
# --
import hmac
import hashlib
import requests
import json

from binance.client import Client

client = Client()
klines_BTC = client.get_historical_klines("BTCUSDT", Client.KLINE_INTERVAL_15MINUTE, '15 June, 2022', '10 Jan 2023')

taisho_cols=['Time','Open','High','Low','Close','Volume','Close_Time','Quote_Volume','Number_of_Trades','Taker_buy_base_asset_volume','Taker_buy_quote_asset_volume','Ignore']
df_BTC = pd.DataFrame(df_klines, columns = [taisho_cols])

f_maxmin = lambda x: dt.fromtimestamp(float(x)/1000).strftime('%Y%m%d%H%M%S')
df_BTC["timestamp"] = df_BTC["Close_Time"].apply(f_maxmin)

ひとまず、いろいろ抽出できてるっぽい

	('Time',)	('Open',)	('High',)	('Low',)	('Close',)	('Volume',)	('Close_Time',)	('Quote_Volume',)	('Number_of_Trades',)	('Taker_buy_base_asset_volume',)	('Taker_buy_quote_asset_volume',)	('Ignore',)	('timestamp',)
0	1655251200000	22136.4	22153.2	21882	21934	1424.74	1655252099999	3.13436e+07	21181	744.015	1.63642e+07	0	20220615091459
1	1655252100000	21935.4	22100	21904.3	21981.7	755.669	1655252999999	1.66372e+07	13551	377.503	8.31082e+06	0	20220615092959
2	1655253000000	21981.7	22147	21942.3	22047.9	523.564	1655253899999	1.15532e+07	11850	261.831	5.77717e+06	0	20220615094459
3	1655253900000	22047.9	22091.5	21741.6	21808.3	1217.73	1655254799999	2.66141e+07	17946	574.265	1.25462e+07	0	20220615095959
4	1655254800000	21808.3	21968.9	21770.7	21951	770.302	1655255699999	1.68429e+07	13102	422.603	9.2407e+06	0	20220615101459

データ加工

ネットの記事を参考に、過去実績系の特徴量・指標系特徴量をつくる。

def make_lag_feature(x, col_name, sbn_flg=True):
    # ---
    add_lag_list = ["lag2","lag4","lag12","lag48","lag96","lag144","lag288"]
    add_std_list = ["std2","std4","std12","std48","std96","std144","std288"]
    # ---
    col_nm1 = [f"{i}_{col_name}" for i in add_lag_list]
    col_nm2 = [f"{i}_{col_name}" for i in add_std_list]
    # ---
    for i in col_nm1:
        # --
        suchi = int(re.sub(r"\D", "", i))
        x[i]  = x[col_name].shift(suchi)
        # --
        if sbn_flg:
            col_nm2 = f"abs_{i}"
            x[col_nm2]  = (x[col_name]-x[i]).abs()
        else:
            col_nm2 = f"abs_{i}"
            x[col_nm2]  = (x[i]).abs()
    # --
    for i in col_nm1:
        # --
        suchi = int(re.sub(r"\D", "", i))
        x[i]  = x[col_name].rolling(suchi).std()
    # --
    return x

def make_feature(x):
    # --
    base_col = ["Open", "High", "Low", "Close", "Volume"]
    cate_col = ["op_cl", "op_hi", "op_lo", "hi_cl", "hi_lo", "lo_cl"]
    # ---
    x["op_cl"] = x["Open"] - x["Close"]
    x["op_hi"] = x["Open"] - x["High"]
    x["op_lo"] = x["Open"] - x["Low"]
    x["hi_cl"] = x["High"] - x["Close"]
    x["hi_lo"] = x["High"] - x["Low"]
    x["lo_cl"] = x["Low"] - x["Close"]
    # --
    for col_name in base_col:
        x = make_lag_feature(x, col_name, sbn_flg=True)
    for col_name in cate_col:
        x = make_lag_feature(x, col_name, sbn_flg=False)
    return x


def make_ta_feature(x):
    # Add Simple Moving Average (SMA) indicators
    x["sma7"] = SMAIndicator(close=x["Close"], window=7, fillna=True).sma_indicator()
    x["sma25"] = SMAIndicator(close=x["Close"], window=25, fillna=True).sma_indicator()
    x["sma99"] = SMAIndicator(close=x["Close"], window=99, fillna=True).sma_indicator()
    # Add Bollinger Bands indicator
    indicator_bb = BollingerBands(close=x["Close"], window=20, window_dev=2)
    x['bb_bbm'] = indicator_bb.bollinger_mavg()
    x['bb_bbh'] = indicator_bb.bollinger_hband()
    x['bb_bbl'] = indicator_bb.bollinger_lband()
    # Add Parabolic Stop and Reverse (Parabolic SAR) indicator
    indicator_psar = PSARIndicator(high=x["High"], low=x["Low"], close=x["Close"], step=0.02, max_step=2, fillna=True)
    x['psar'] = indicator_psar.psar()
    # Add Moving Average Convergence Divergence (MACD) indicator
    x["MACD"] = macd(close=x["Close"], window_slow=26, window_fast=12, fillna=True) # mazas
    # Add Relative Strength Index (RSI) indicator
    x["RSI"] = rsi(close=x["Close"], window=14, fillna=True) # mazas
    return x

def create_features(x):
    # --
    x = make_feature(x)
    x = make_ta_feature(x)
    return x

次回はいろいろアルゴリズムに入れて、挙動みていく予定