Skip to main content

CodeStr

CI Python License Ruff

CodeStr 是一个专为量化因子挖掘设计的 DSL → Polars Expr 表达式计算引擎,提供高效的表达式转译、缓存与执行。

安装

git clone https://github.com/huangbogeng/codestr.git
cd codestr
uv sync --extra dev

快速开始

import polars as pl
from codestr import CodeStr

# 标准面板数据 (time, entity)
df = pl.DataFrame({
    "datetime": ["2024-01-01", "2024-01-01", "2024-01-02", "2024-01-02"],
    "asset":    ["A", "B", "A", "B"],
    "close":    [100.0, 200.0, 101.0, 198.0],
    "volume":   [1000.0, 2000.0, 1100.0, 1900.0],
})

cs = CodeStr(df, index=("datetime", "asset"))

# 交互式查询 — 结果自动缓存
result = cs.sql(
    "ts_mean(close, 5) as ma5",
    "cs_rank(close) as rank",
    "close / ts_delay(close, 1) - 1 as ret",
)
print(result)

API 模式

模式 API 行为
纯编译 cs.compile(expr) -> pl.Expr 无副作用,返回 Polars 表达式
交互式 cs.sql(expr, lazy=False) -> pl.DataFrame 有状态,自动缓存与复用
静态验证 cs.validate_expr(*exprs) -> list[dict] 无副作用,按当前 schema 干编译
# 纯编译 — 表达式可被任意 DataFrame 消费
expr = cs.compile("ts_mean(close, 5) as ma5")
other_df.with_columns(expr)

# 交互式 — 适合逐步构建因子
cs.sql("close + volume as total")
cs.sql("ts_mean(total, 5) as total_ma5")  # 复用上一步的 total

静态验证

validate_expr() 复用 sql() 的 planner,在当前 LazyFrame schema 上解析 UDF、列引用、类型兼容性和混合窗口,但只调用 collect_schema(),不会 collect() 数据或修改引擎缓存:

results = cs.validate_expr(
    "sin(1.0) as invalid",
    "ts_mean(cs_moderate(close), 5) as factor",
)

每个结果包含 exprvalidstageerror_typemessage。 失败阶段分为 structuralcompileschema。批量输入相互独立, 后一条表达式不能引用同批前一条表达式新建的别名。依赖实际数据值、 只在物化时出现的错误不属于该 API 的检查范围。

窗口配置

CodeStr 使用 partition_by(实体分组轴)和 order_by(时间排序轴)控制窗口算子:

# 默认配置
cs = CodeStr(df)
# index=("datetime", "asset")
# → TS: over(partition_by=["asset"], order_by=["datetime"])
# → CS: over(partition_by=["datetime"], order_by=["asset"])

# 自定义列名
cs = CodeStr(df, index=("trade_date", "stock_code"))

# 多列窗口 — 按行业+股票分组,按日期+逐笔序号排序
cs = CodeStr(df,
    index=("trade_date", "stock_code"),
    partition_by=["industry", "stock_code"],
    order_by=["trade_date", "tick"],
)
算子类别 窗口规则
TS (时序) over(partition_by=partition_by, order_by=order_by)
CS (截面) over(partition_by=order_by, order_by=partition_by)

混合窗口

CodeStr.sql() 会自动把 TS/CS 混合窗口拆成连续的 lazy projection:

cs.sql(
    "close * 2 as scaled",
    "ts_mean(cs_moderate(scaled), 60) as factor",
)

其语义等价于先生成 cs_moderate(scaled) 中间列,再沿资产时间轴 计算 ts_mean。中间列不会出现在返回结果中。

cs.compile() 只能返回一个 pl.Expr,因此会明确拒绝需要多阶段 执行的 TS/CS 混合窗口。TS→TS 和 CS→CS 同域嵌套不受影响。

自定义算子

from codestr.udf.registry import udf
import polars as pl

@udf(category="ts")
def ts_ewm(expr: pl.Expr, windows, partition_by=None, order_by=None):
    """指数加权移动平均"""
    return expr.ewm_mean(halflife=windows).over(
        partition_by=partition_by, order_by=order_by
    )

cs.sql("ts_ewm(close, 10) as ewm10")

内置算子

基础算子 (base_udf):abs, log, sqrt, square, cube, sin, cos, tan, exp, sigmoid, sign, clip, trunc, between, cast, max, min, sum, mean, arg_max, arg_min, if_, fib

截面算子 (cs_udf):cs_rank, cs_zscore, cs_demean, cs_mean, cs_std, cs_var, cs_skew, cs_ic, cs_corr, cs_slope, cs_resid, cs_qcut, cs_midby, cs_meanby

时序算子 (ts_udf):ts_mean, ts_ema, ts_sum, ts_std, ts_var, ts_skew, ts_kurt, ts_max, ts_min, ts_mid, ts_delay, ts_delta, ts_mad

滚动统计与 ts_ema 支持关键字参数 min_samples。滚动统计省略该参数时沿用 Polars 默认值 None(需要完整窗口),ts_ema 则沿用 Polars 默认值 1

cs.sql("ts_mean(close, 20, min_samples=5) as ma20")
cs.sql("ts_ema(close, 10, min_samples=3) as ema10")

项目结构

src/codestr/
├── engine.py            # CodeStr 引擎入口
├── compiler.py          # AST → Polars Expr 编译器
├── parser.py            # DSL 解析器 (Lark LALR grammar)
├── syntax.py            # AST 节点定义
├── tokens.py            # Token 定义
├── errors.py            # 异常类型
└── udf/
    ├── registry.py      # UDF 注册中心 (@udf 装饰器)
    ├── base_udf.py      # 基础算子
    ├── cs_udf.py         # 截面算子 (Cross-Section)
    └── ts_udf.py         # 时序算子 (Time-Series)

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

codestr-0.3.2.tar.gz (125.4 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

codestr-0.3.2-py3-none-any.whl (26.0 kB view details)

Uploaded Python 3

File details

Details for the file codestr-0.3.2.tar.gz.

File metadata

  • Download URL: codestr-0.3.2.tar.gz
  • Upload date:
  • Size: 125.4 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for codestr-0.3.2.tar.gz
Algorithm Hash digest
SHA256 29497646cffb7968809e00a94a4951d8c96caa7dd5fa3770a0bd5f15a6c7e01e
MD5 865694931a1314fc8014ead40857f853
BLAKE2b-256 dc44529e307863b9735578a648e0b150d8cf09dfef8f063ff5a060b930e9502a

See more details on using hashes here.

Provenance

The following attestation bundles were made for codestr-0.3.2.tar.gz:

Publisher: release.yml on huangbogeng/codestr

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file codestr-0.3.2-py3-none-any.whl.

File metadata

  • Download URL: codestr-0.3.2-py3-none-any.whl
  • Upload date:
  • Size: 26.0 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for codestr-0.3.2-py3-none-any.whl
Algorithm Hash digest
SHA256 93c71bae4f41c10b21d7e4b80470149d31feb8988d59997c0806d81e52b2cf00
MD5 9ba5fa95d50bd908d7b49fe9a17a38db
BLAKE2b-256 a2c2d666ff83c0fd764a66c6a352af96606b4ccfc79c8b518cbec7a28e7ec534

See more details on using hashes here.

Provenance

The following attestation bundles were made for codestr-0.3.2-py3-none-any.whl:

Publisher: release.yml on huangbogeng/codestr

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Release history Release notifications | RSS feed

This release

0.3.2 This release

2 files

0.3.1

2 files

0.3.0

2 files

0.2.1

2 files

0.2.0

2 files

0.1.0

2 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page