mirror of
https://github.com/microsoft/qlib.git
synced 2026-11-11 21:41:06 +08:00
XianBWandCopilot
7c95268c0c
* security: harden artifact serialization and loading * ci: restore dependency and title lint compatibility * fix: address artifact test and lint failures * ci: handle current MLflow and Windows OSQP changes * ci: use lxml release with cross-platform wheels * fix: defer optional CVXPY solver imports * docs: mark notebook model artifacts as trusted * docs: mark position artifacts as trusted * style: document intentional lazy import * ci: skip broken released OSQP on Windows Python 3.8 * fix: mark online model artifacts as trusted * test: make MLflow client timing check robust * ci: retry nondeterministic benchmark smoke test * fix: cap filelock before fork safety regression * fix: use filelock release compatible with multiprocessing * fix: preserve artifact loading compatibility after hardening * fix(security): validate derived artifact paths and extend regressions * fix(security): preserve extended data artifact round trips * fix(security): retain legacy NumPy and pandas artifact classes Allow the exact MaskedArray and SparseDtype module paths emitted by the older dependency versions in CI. Preserve restricted loading and cover both old and new paths. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * fix(data): avoid unused trading calendar lookup in PIT normalization PIT normalization uses report dates and calendar-day offsets, so do not request an unused Baostock calendar. Cover annual and quarterly normalization without online calendar access. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * fix(security): require explicit trust in artifact workflows Expose default-off artifact trust at online, delayed-training and meta-workflow entry points. Keep data loading restricted, preserve legacy default recorder calls and restored workflow defaults, and document migration and trust boundaries. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * test(workflow): cover complete DDG-DA and HIST artifact flows Exercise real training, cache and recorder restoration, saved-task replay, signals and portfolio backtests. Fix the cache trust, grouping, relative-path, LightGBM early-stopping and HIST CPU blockers exposed by full execution. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * fix(security)!: unify artifact trust and publish upgrade guidance Use trusted consistently across public APIs and CLI options. Migrate pre-release saved consent only at serialization boundaries, reject invalid or conflicting settings, and preserve restricted data reads. Add a canonical upgrade guide with README, installation, changelog and runtime-error entry points. BREAKING CHANGE: executable artifact restoration requires explicit trusted=True. HIST mappings use JSON and high-frequency cache paths must remain within artifact_root. See docs/start/artifact_migration.rst for upgrading existing workflows. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * fix(security): remove trial-only trust compatibility Target upgrades from pre-PR Qlib rather than artifacts generated by intermediate PR revisions. Retain restricted defaults for saved objects without consent, unify current APIs, and remove state/keyword migration branches and documentation. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * ci: avoid failed dataset probes and space download retries Download the same existing v2 archives used by successful jobs without probing nonexistent package-version assets. Allow five download attempts 60 seconds apart; preserve the full matrix, datasets, noninteractive setup and failure propagation. Cover the exact archives and retry policy in offline CI tests. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * docs: add practical artifact migration examples Show explicit consent for restored online components while preserving the original manager, schema-aware Arrow conversions, timezone-preserving ZoneInfo migration, and restricted round trips into a separate recorder. Explain shared-component scope, staging artifacts and trusted-source prerequisites. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * fix(workflow): preserve fresh artifact workflows with scoped trust Keep immutable cache references intact at open rolling tails. Add explicit per-load record-template consent and per-artifact collector loading options without widening other reads. Restore the original template class after parent-load failures, and support audited business-frequency and NumPy record representations through exact pickle globals. Document the scoped APIs and cover fresh trained-model reads, cache generation, numerical round trips and refusal paths. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Introduction
Due to the non-stationary nature of the environment of the financial market, the data distribution may change in different periods, which makes the performance of models build on training data decays in the future test data. So adapting the forecasting models/strategies to market dynamics is very important to the model/strategies' performance.
The table below shows the performances of different solutions on different forecasting models.
Alpha158 Dataset
Here is the crowd sourced version of qlib data: https://github.com/chenditc/investment_data/releases
wget https://github.com/chenditc/investment_data/releases/latest/download/qlib_bin.tar.gz
mkdir -p ~/.qlib/qlib_data/cn_data
tar -zxvf qlib_bin.tar.gz -C ~/.qlib/qlib_data/cn_data --strip-components=2
rm -f qlib_bin.tar.gz
| Model Name | Dataset | IC | ICIR | Rank IC | Rank ICIR | Annualized Return | Information Ratio | Max Drawdown |
|---|---|---|---|---|---|---|---|---|
| RR[Linear] | Alpha158 | 0.0945 | 0.5989 | 0.1069 | 0.6495 | 0.0857 | 1.3682 | -0.0986 |
| DDG-DA[Linear] | Alpha158 | 0.0983 | 0.6157 | 0.1108 | 0.6646 | 0.0764 | 1.1904 | -0.0769 |
| RR[LightGBM] | Alpha158 | 0.0816 | 0.5887 | 0.0912 | 0.6263 | 0.0771 | 1.3196 | -0.0909 |
| DDG-DA[LightGBM] | Alpha158 | 0.0878 | 0.6185 | 0.0975 | 0.6524 | 0.1261 | 2.0096 | -0.0744 |
- The label horizon of the
Alpha158dataset is set to 20. - The rolling time intervals are set to 20 trading days.
- The test rolling periods are from January 2017 to August 2020.
- The results are based on the crowd-sourced version. The Yahoo version of qlib data does not contain
VWAP, so all related factors are missing and filled with 0, which leads to a rank-deficient matrix (a matrix does not have full rank) and makes lower-level optimization of DDG-DA can not be solved.