kr_gld的作业

实现了 ：

xgboost 和 stockrank 的策略。以及 超参测试

相同因子的情况下， stockrank 要比 xgboost 更好些。

实现 ：
xgboost 的策略

因子设计

c_pct_rank(dividend_yield_ratio) as rank_div_ratio

c_pct_rank(total_market_cap) as rank_cap

c_pct_rank(close) as real_close

c_rank(close / m_lag(close, 20)) as rank_mount

c_pct_rank(m_avg(turn,20)) as rank_turn

m.tune

result = M.tune.run(
    "search",
    [
        # 叶子节点数量为30, 树的数量为20
        {"m5.number_of_leaves": 30, "m5.number_of_trees": 20, '__outputs__': ['m7']},

        # 叶子节点数量为40, 树的数量为30
        {"m5.number_of_leaves": 50, "m5.number_of_trees": 30, '__outputs__': ['m7']},

        # 叶子节点数量为80, 树的数量为40
        {"m5.number_of_leaves": 80, "m5.number_of_trees": 60, '__outputs__': ['m7']},

        # 叶子节点数量为100, 树的数量为50
        {"m5.number_of_leaves": 100, "m5.number_of_trees": 50, '__outputs__': ['m7']},
    ],
)

实现：

stockrank 的策略

因子设计：

c_pct_rank(dividend_yield_ratio) as rank_div_ratio

c_pct_rank(total_market_cap) as rank_cap

c_pct_rank(close) as real_close

c_rank(close / m_lag(close, 20)) as rank_mount

c_pct_rank(m_avg(turn,20)) as rank_turn

m.tune

[

{"m5.number_of_leaves": 10, "m5.number_of_trees": 20, 'outputs': ['m6']},

{"m5.number_of_leaves": 20, "m5.number_of_trees": 20, 'outputs': ['m6']},

{"m5.number_of_leaves": 30, "m5.number_of_trees": 30, 'outputs': ['m6']},

{"m5.number_of_leaves": 40, "m5.number_of_trees": 30, 'outputs': ['m6']},

],