Conversation
- Switch license from MIT to Apache 2.0 - Remove legacy setup.py, keep only pyproject.toml - Fix package discovery to exclude assets directory - Change default sort order to [timestamp, symbol] for cross-sectional workflows - Update README license badge Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
清除 WQ-style, phandas 等引用 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
測試預期行為：跳過 null，rank 用 method=average Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- 所有截面運算子從 has_null→kill_all 改為 per-row null check - rank: method=min → method=average（ties 取平均排名） - zscore: ddof=1 → ddof=0（母體標準差） - quantile: 重寫為 rank→shift→inverse CDF，保留 [DEV] Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- ts_std_dev, ts_covariance, ts_zscore, ts_cv: ddof=0 - ts_corr, ts_autocorr: 保留 ddof=1（相關係數不受 ddof 影響） - 全面 min_periods→min_samples 消除 deprecation warning Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
原 ts_quantile 改名 ts_percentile（rolling 百分位數） 新 ts_quantile 匹配 spec：rank → inverse CDF transform Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
filter=True: add/subtract 把 null 視為 0，multiply 把 null 視為 1 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ts_skewness: 手算公式完全算錯（符號和數值都不對），改用 rolling_skew(bias=True) ts_decay_linear: 6 行 shift 邏輯改為 1 行 rolling_mean(weights=[1,2,...,n]) 兩者均經 verify_against_polars.py 驗證與原生 Polars 完美匹配 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
一行呼叫顯示：維度、覆蓋率、null/NaN/inf 計數、分布統計、rank turnover 統計量排除 null/NaN/inf 後計算，避免污染 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
統計量先 per-timestamp 截面計算，再顯示 mean +/- std over time cs_mean, cs_std, cs_skew 各自顯示時序穩定性 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New operators: - ts: ts_step, ts_decay_exp_window, days_from_last_change, kth_element, last_diff_value, inst_tvr, ts_delta_limit, ts_regression (10 rettypes), trade_when - cs: winsorize, truncate, left_tail, right_tail, quantile (exported) - base: densify, bucket - group: group_mean (weighted), group_median, group_backfill Fixes: - ts_quantile gaussian driver: zscore -> rank+ppf (Acklam approximation) - winsorize/truncate: null-safe clip via max/min_horizontal - bucket range_params: fix float accumulation drift - normalize: add use_std, limit params - scale: add longscale, shortscale params - log type annotation: float -> Optional[float] Removals: - ln() redundant with log() - Factor.describe() moved to analytics layer - __all__ deduplication: ops/__init__.py is single source of truth Tests: consolidated 11 files -> 5+conftest (1:1 module mapping), 131 tests Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- CLAUDE.md: complete development standards (git workflow, coding standards, commit convention, code review rules, versioning, known limitations) - CI: GitHub Actions workflow for pytest on Python 3.10-3.13 + ruff lint - Pre-commit: ruff lint/format + pytest hooks - PR template: change type checklist with numerical impact section - CHANGELOG.md: structured change log with [NUMERICAL] tags - pyproject.toml: dev dependencies (pytest, ruff, pre-commit), tool configs - _validation.py: input validation helpers for operator parameters - .gitignore: exclude .claude/ local settings Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…anup [NUMERICAL] Three changes that affect computation results: 1. ts_product: silently returned null for negative inputs due to log-sum-exp trick. Replaced with sign-magnitude decomposition (count negatives for sign, exp(sum(log(abs(x)))) for magnitude, zero detection). 2. ts_covariance: used ddof=0 (population) inconsistent with ts_corr (ddof=1). Aligned to ddof=1 (sample covariance). This changes output values. 3. ts_count_nans: used min_samples=1 unlike all other ts_* operators. Aligned to min_samples=window for consistency. 4. ts_regression rettype=7 (MSE): added explicit zero guard for window=2 (n-2=0). Previously relied on implicit Inf-to-null conversion. Also moved DEV operators (hump, ts_arg_max, ts_arg_min) to _dev.py, removed from public API. These use Python callbacks and are not production-grade. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
[NUMERICAL] divide() and inverse() previously produced Inf when dividing by zero, relying on Factor constructor's implicit Inf-to-null conversion. Now explicitly returns null where abs(denominator) < 1e-10, matching the zero-guard pattern used by all other division operations in the library. Added tests for divide() (basic, zero denominator, scalar, scalar zero) and reverse() (negate, null preservation) -- both previously had zero test coverage. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
_balance() previously hardcoded interval="1d", which would silently drop all intraday data when loading hourly or minutely datasets. The skeleton was built with daily timestamps only, and the left join discarded any rows not matching midnight timestamps. load() now accepts an interval parameter (default "1d") that propagates to _balance(). Supports any Polars duration string (e.g., "1h", "5m"). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Factor constructor now validates required columns [timestamp, symbol, factor] and raises ValueError with clear message if missing - Added __hash__ = None to Factor class (explicit, since __eq__ returns Factor not bool -- matches pandas semantics, documented in docstring) - Applied ruff auto-formatting across all modules (import sorting, Union -> X | Y syntax for Python 3.10+) - No behavioral changes beyond the new constructor validation Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- publish.yml: triggered by v* tags, builds and publishes to PyPI via Trusted Publisher (no API tokens stored in repo) - CLAUDE.md: added complete release process (7 steps), versioning table with examples, one-time GitHub setup instructions Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- publish.yml: now runs full test suite before publishing, creates GitHub Release with auto-generated notes and build artifacts - CLAUDE.md: comprehensive rewrite with 12 numbered sections covering all development practices from daily workflow to release process. This is the single source of truth for any developer or LLM. Sections: project overview, architecture, git workflow, coding standards, testing, commit convention, code review, versioning, release process, one-time setup, commands reference, known limitations. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Version 0.2.0 includes production audit fixes with multiple numerical corrections. See CHANGELOG.md for full details. Key numerical changes: - ts_product: now handles negative values correctly - ts_covariance: aligned to ddof=1 (sample) matching ts_corr - divide/inverse: explicit zero-division guards
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Production audit: 8 bug fixes (4 numerical), project scaffolding, CI/CD automation.
Change Type
Numerical Impact
Testing