Optimization, Native in SQL

DecidB extends DuckDB with declarative optimization.
Select optimal subsets of data with constraints and objectives — no external tools required.

Quick Example

SELECT item, value, weight, x AS selected
FROM Items
DECIDE x IS BOOLEAN
SUCH THAT
    SUM(x * weight) <= 50
MAXIMIZE SUM(x * value);

Why DecidB?

• Native SQL — Express optimization as a SQL extension. No context switching between your database and an external solver.
• Zero Data Movement — Solve directly on database buffers. No export/import overhead.
• Declarative — Define what to optimize, not how. The system handles ILP formulation automatically.
• Built on DuckDB — Columnar storage, vectorized execution, and an embedded HiGHS solver. Optional Gurobi support for commercial workloads.

DECIDE Syntax

SELECT select_list
FROM table_expression
[WHERE ...]
DECIDE variable_name [IS type] [, ...]
SUCH THAT constraint [AND constraint ...]
[MAXIMIZE | MINIMIZE] SUM|MIN|MAX(linear_expression)

Feature	Details
Variable types	IS BOOLEAN (0/1), IS INTEGER (non-negative, default), IS REAL (continuous)
Constraint operators	=, <, <=, >, >=, <>, BETWEEN, IN
Aggregates	SUM(), COUNT(), AVG(), MIN(), MAX()
Conditional	expression WHEN condition (postfix)
Grouping	SUM(expr) op rhs PER column — one constraint per distinct group

For the full syntax specification, see [context/descriptions/00_project_overview/syntax_reference.md](context/descriptions/00_project_overview/syntax_reference.md).

Building from Source

DecidB requires CMake, Python3, and a C++11 compliant compiler.

make release

Build output:

• CLI: build/release/decidb
• Library: build/release/src/libduckdb.so

Running Tests

make decide-test       # Run DECIDE differential tests
make decide-setup      # Setup test environment only

Tests are located in [test/decide/](test/decide/).

Example Problem Domains

DecidB can express a wide range of optimization problems directly in SQL:

• Knapsack / Packing — Maximize value within weight or budget limits
• Diet / Nutrition — Meet nutritional targets while minimizing cost
• Portfolio Selection — Maximize return under risk constraints
• Resource Allocation — Assign limited resources to maximize output
• Production Planning — Optimize production quantities with capacity constraints
• Assignment — Assign items to resources optimally
• Scheduling — Complex constraint satisfaction over time slots
• Multi-period Optimization — Time-indexed decision problems with carry-over constraints

Documentation

Full documentation lives in [context/descriptions/](context/descriptions/). Start with the README there for navigation and reading order.

Key areas:

• [00_project_overview/](context/descriptions/00_project_overview/) — Syntax specification
• [01_pipeline/](context/descriptions/01_pipeline/) — Query processing architecture
• [03_expressivity/](context/descriptions/03_expressivity/) — Feature status (WHEN, PER, aggregates, etc.)
• [04_optimizer/](context/descriptions/04_optimizer/) — Optimizer rewrite strategies

Research

DecidB is developed by the HUDA Lab (NYU Abu Dhabi), and UMass Amherst.

Foundation papers:

• Scalable Package Queries in Relational Database Systems (VLDB 2016)
• Scalable Computation of High-Order Optimization Queries (ACM Communications, 2019)
• Scaling Package Queries to a Billion Tuples (VLDB 2024)

Contact: [email protected]

License

DecidB is released under the MIT License. See LICENSE for details.