Qlib AI Quant

Microsoft qlib AI quant research platform: Alpha158/TFT feature engineering, prediction models, portfolio optimization, and multi-frequency backtest in one stack. Multi-market (A-share / US / HK).

Factor AnalysisQuantitativeAIBacktestMachine Learning
✓ 0 reported success·v0.1.0·

Overview

Qlib AI Quant is Microsoft's open-source AI-driven quantitative investment platform (github.com/microsoft/qlib). It covers the full research pipeline from raw data processing, feature engineering (Alpha158 / Alpha360 / TFT — Temporal Fusion Transformer), model training (GBDT / DNN / Transformer), to portfolio optimization and multi-frequency backtest. Supports A-share, US, and HK markets with a built-in factor library and Handler framework for rapid strategy iteration. Typical use cases: factor efficacy research, ML stock selection, multi-frequency data fusion (daily + minute), portfolio rebalance. This skill embeds constraints for common qlib pitfalls: frequency conversion NaN handling, Handler initialization order, TFT data format requirements. The host AI applies these automatically.

Blueprint Source

finance-bp-087

microsoft/qlibd5379c52 source files

Constraints

86total
61fatal
61 must-not-violate

Evidence Quality

Confidence89%

Medium confidence — review before critical use

61 non-negotiable constraints

FATALdomain_rulefinance-C-001

WHENWhen implementing expression load with index range parameters

ACTIONAllow start_index to exceed end_index in Expression.load

CONSEQUENCEInvalid index range causes data corruption where features return empty or misaligned series, breaking calendar index alignment for all downstream consumers

domain-constraint
FATALdomain_rulefinance-C-002

WHENWhen implementing a new Expression subclass

ACTIONImplement _load_internal abstract method to return pd.Series indexed by calendar

CONSEQUENCESubclass without _load_internal raises NotImplementedError, preventing feature calculation and breaking the entire data pipeline

domain-constraint
FATALdomain_rulefinance-C-008

WHENWhen using PIT (Point-in-Time) database expressions

ACTIONCreate PIT expressions with positive right extended window (end_ws > 0) referencing future periods

CONSEQUENCEFuture period references in PIT cause ValueError and break point-in-time data integrity, leading to look-ahead bias in backtests

domain-constraint

FAQ

Discussion (0)

No comments yet. Be the first to share!

Changelog

v0.1.02026-04-23·Contributors: tangweigang-jpg

v0.1.0: Initial release on Doramagic.ai. Microsoft qlib AI quant platform with bilingual metadata, 47 anti-pattern constraints (each with GitHub issue provenance), and 3 FAQs.