Qlib

Local setup (Mac)

Qlib is Python-based. Recommended: conda env.

1) Install Python tooling

If you don’t have Python/conda installed:

• Install Miniforge/Conda (recommended), then create a clean env.

2) Create env + install

conda create -n qlib python=3.11 -y
conda activate qlib

# Mac tip (especially Apple Silicon):
brew install libomp

pip install -U pip
pip install pyqlib

Or dev install:

git clone https://github.com/microsoft/qlib.git
cd qlib
pip install -e .

Data

Qlib needs a dataset. The upstream README notes the official dataset is sometimes disabled; community releases exist.

Example (from upstream README):

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=1
rm -f qlib_bin.tar.gz

Quick “does it run” test

Try the code-based workflow example:

• https://github.com/microsoft/qlib/blob/main/examples/workflow_by_code.py

How we’ll use this to improve alpha-engine

We will not switch alpha-engine to Qlib. Instead we’ll import Qlib discipline:

1. Factorize our entry score

   • turn current gates (momentum, buy pressure, liquidity, txns, quoteConfidence) into numeric factors
   • normalize (z-score) and combine into a single score

2. Offline evaluation loop

   • use replay logs as our “dataset”
   • build a simple backtest/eval harness that compares factor variants (A/B)

3. Export artifacts

   • JSON report of factor weights + performance deltas
   • keep a changelog that ties config fingerprint → results

Guardrails

• Read-only research by default.
• No live trading changes without explicit approval.