Qlib：Microsoft発のAI量的投資フレームワーク（39kスター）

import qlib
from qlib.constant import REG_CN

# Qlibの初期化（中国市場向け設定）
qlib.init(provider_uri="~/.qlib/qlib_data/cn_data", region=REG_CN)

# データプロバイダーの確認
from qlib.data import D
market_data = D.features(["000001.SZ"], fields=["$open", "$close"], start_time="2020-01-01", end_time="2020-12-31")
print(market_data)

from qlib.data import D
from qlib.backtest import collect_data

# 日足データの取得と前処理
fields = [
    "$open",
    "$close",
    "$high",
    "$low",
    "$volume",
    "Ref($close, 1) / $close - 1",  # 前日比リターン
]

df = D.features(
    ["000001.SZ"],
    fields=fields,
    start_time="2020-01-01",
    end_time="2020-12-31",
    freq="day"
)
print(df.head())

from qlib.backtest import backtest
from qlib.strategy import BaseStrategy

class SimpleStrategy(BaseStrategy):
    def generate_signal(self):
        # 前日の終値から当日の終値への変化率
        returns = self.df["$close"].pct_change()
        # リターンが正の日に買い、負の日に売る
        signal = (returns > 0).astype(int)
        return signal

# バックテスト実行
strategy = SimpleStrategy()
results = backtest(
    strategy=strategy,
    start_time="2020-01-01",
    end_time="2020-12-31"
)
print(results.summary_stats())

from qlib.data import D
import numpy as np

# 複数の特徴を定義
features = [
    "$close",
    "Ref($close, 1) / $close - 1",  # リターン
    "($high - $low) / $close",      # ボラティリティ
    "$volume / Ref($volume, 5)",    # 出来高の変動
    "Mean($close, 20) / $close - 1", # 20日移動平均との乖離
]

df = D.features(
    instruments=["000001.SZ", "000002.SZ", "000004.SZ"],
    fields=features,
    start_time="2019-01-01",
    end_time="2021-12-31",
    freq="day"
)

# NaN値を処理
df = df.fillna(method='bfill').fillna(method='ffill')
print(f"データ形状: {df.shape}")
print(df.describe())

from qlib.data import D
from qlib.model.trainer import task_train
from qlib.utils import get_pre_trading_date
import lightgbm as lgb
import numpy as np

# 訓練用の特徴セットを準備
training_fields = [
    "Ref($close, 1) / $close - 1",
    "($high - $low) / $close",
    "Mean($close, 5) / $close - 1",
    "Mean($close, 20) / $close - 1",
    "Ref($volume, 1) / Mean($volume, 5) - 1",
]

# 訓練データの取得
train_df = D.features(
    instruments=["000001.SZ"],
    fields=training_fields + ["Ref($close, -1) / $close - 1"],  # 翌日リターン（目的変数）
    start_time="2019-01-01",
    end_time="2020-12-31",
    freq="day"
)

# NaN値処理とデータ分割
train_df = train_df.dropna()
X_train = train_df[training_fields].values
y_train = (train_df["Ref($close, -1) / $close - 1"] > 0).astype(int).values

# LightGBMモデルの訓練
model = lgb.LGBMClassifier(n_estimators=100, max_depth=7, learning_rate=0.05)
model.fit(X_train, y_train, eval_set=[(X_train[:1000], y_train[:1000])], verbose=10)

print(f"モデル精度: {model.score(X_train, y_train):.4f}")