Skip to content

jasonmorton/Cateno

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Warning: this is very early stage, not ready for use. It is public just to share with some interested folks.

Cateno

======= Cateno is a system for computational category theory and applications. It provides an interactive calculator for free morphism expressions and string diagram generation for monoidal categories. It also handles concrete categories, and can be used as a typed numerical linear algebra system. Algorithms are implemented using catgorical interfaces: tell Cateno how the objects, morphisms, and structure maps are implemented, and you have access to fast algorithms for normal forms, inference, tree decomposition, contraction, marginalization, and so on. This enables fast development of models of uncertainty.

Check out the static demos (Ipython/Jupyter notebooks): qubits, and screencasts:

Cateno is primarily written in Julia. It can be used in an IPython/Jupyter notebook, interactively at the Julia prompt (with diagram-drawing in a separate window), or as a set of libraries in your programs. In the future we plan to make available bindings or ports to other languages such as Python, Scala, and Rust.

Goals

A practical computer algebra system for computational (monoidal) category theory should:

  1. Manipulate abstract categorical quantities such as morphism terms in a REPL.
  2. Compile code expressed categorically to an efficient implementation in a particular category (e.g. numerical linear algebra, (probabilistic) databases, quantum simulation, belief networks).
  3. Scale to be useful for practical computational problems in modeling uncertainty in data analysis, statistics, physics, computer science.

There is more than one way to accomplish (1), and our design choices are driven by (2) and (3) as well. In particular we use wrapper classes (wrap an untyped object in a typed wrapper) and representations (functorial bindings) whenever possible. Note that Cateno is not intended as a proof assistant; rather it is a consumer of proofs that, e.g., something is a compact closed category. Cateno then lets you use that knowledge to compute things and build models.

For people interested primarily in modeling, rather than computational category theory in its own right, we aim to:

  1. Ingest qualitative domain knowlege expressed with flow charts/diagrams from non-programmers.
  2. Attach multiple competing quantitative explanations (e.g.\ probabilistic, differential equation, discrete dynamical) and test them.
  3. Port algorithms, using best solution in each component (e.g. tree decomposition).
  4. Scale to useful size, exploiting parallelsm and concurrency.
  5. Pay only a small abstraction cost in final assembly program.

From the modeling side, we want something like MATLAB, NumPy, R, but that:

  • allows a higher level of abstraction in describing algorithms,
  • can handle more types of ``data'' than matrices of floats or probability distributions, and
  • treats morphism expressions as first class to enable rewriting, syntax tricks.

We'd like to think that Computational category theory is the numerical linear algebra of the 21st century, and this is the first steps toward a system that makes that real.

Now: reducing an applied math problem to numerical linear algebra means you can solve it using BLAS, LAPACK primitives, matrix decompositions, etc.

Future: reduce your uncertainty/information-processing applied math problem to (computational) category theory, solve it using generic engines and libraries with matching abstractions.

Design Principles

A kind of category (called a doctrine, for example monoidal category, compact closed category, or well-supported compact closed category) is represented as a Typeclass (also called a trait or interface). These describe the common interface available to manipulate morphism expressions in any particular category.

Everything in Cateno is represented in terms of five modular components:

  • a doctrine (e.g. ``compact closed category''), which is a typeclass

  • an instance (implementation) of a doctrine (e.g. matrices or relations as CCC), implemented as a pair of types which together implement the doctrine. A value of the morphism type of an instance is either:

    • a morphism expression (e.g.\ $f \ot (g \circ \delta_A \circ h)$) in a free language over a tensor signature, or
    • a concrete value in an implementation (e.g.\ $\begin{pmatrix}1 & 3\4 & 5 \end{pmatrix}$)
  • a representation (a functor) between implementations (usually free $\rightarrow$ concrete). This is usually a binding $f = \begin{pmatrix}1 & 3\4 & 5 \end{pmatrix}$) for each symbol, that satisfies some axioms.

  • algorithms are expressed in terms of the defining methods of the doctrine (e.g. $\otimes$, $\delta$) or operadically (more on this later)

From the modeling perspective, there are five parts to a Cateno model.

  • Morphism Term: a human-readable qualitative model, captured by a labeled generalized graph; fixes the relationships, suggests qualitative rules and syntax of the model
  • Doctrine: formal categorical syntax constraining the quantitative models of uncertainty that can be attached, rewrite rules, available constructions
  • Value: a machine-processed quantitative model in which the graph is interpreted and the data summarized, e.g. probabilistically as a Bayesian network, in Hilbert space for a quantum circuit, or with rate constants in a chemical reaction network
  • Representation: the interpretation assigning quantitative meaning to the qualitative description (generalizing the mathematical idea of a representation of a quiver or algebra)
  • Algorithms, categorically expressed, for processing and analyzing data. Make quantitative predictions, choose the model which best explains a given system (often a variant of belief propagation).

About the name

The word cateno is a Latin verb that means "to chain together" or "to bind." Working with Cateno involves chaining together lego-like pieces to make a larger model, and binding morphism variables in a representation. The project is intended to help build bridges across disciplines by making their analogous models literally the same, so we like the fact that cateno is the root of the English word catenary, which suggests such a role. Finally it starts with "cat" just as category, although "category" is Greek in origin.

About

A system for computational category theory and applications

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published