Usage

Venture can be invoked in two styles: as a standalone language, and as a Python library.

Interactive console:

$ venture
venture[script] > assume x (normal 0 1)
venture[script] > observe (normal x 1) 2

Source file:

$ cat prog.vnt
[assume x (normal 0 1)]
[observe (normal x 1) 2]
$ venture puma -f prog.vnt

Python library:

from venture.shortcuts import *
v = make_church_prime_ripl()
v.assume("x", "(normal 0 1)")
v.observe("(normal x 1)", 2)

Python library (batch invocation):

import venture.shortcuts as s
v = s.Puma().make_church_prime_ripl()
v.execute_program("""
    [assume x (normal 0 1)]
    [observe (normal x 1) 2]
""")

The expressive power of the two methods is equivalent, because the standalone language can be extended with plugins and callbacks written in Python, which can then manipulate the Venture system programmatically.

Standalone Invocation

The Venture program is self-documenting as to its invocation pattern. We reproduce here the help messages for representative invocation modes.

$ venture -h
usage: venture [-h] [--version] [-f FILE] [-e EVAL] [-i] [--abstract-syntax] [-L LIBRARY]
               [--prelude]

A general-purpose probabilistic computing platform.

optional arguments:
  -h, --help            show this help message and exit
  --version             show program's version number and exit
  -f FILE, --file FILE  execute the given file
  -e EVAL, --eval EVAL  execute the given program
  -i, --interactive     enter interactive interpreter after completing script file
  --abstract-syntax     interpret abstract rather than concrete syntax
  -L LIBRARY, --library LIBRARY
                        load a Python file in the local directory.  If the file defines
                        a __venture_start__ function, call it with the ripl to
                        initialize the library.
  --prelude             load model prelude on startup (default: off)

optional modes:
  lite                  (default) VentureScript, tracing models with the Lite backend
  puma                  VentureScript, tracing models with the Puma backend
  server                Start a VentureScript command server
  remote                Connect to a running VentureScript command server

Run venture {mode} -h for mode-specific options.

$ venture lite -h
usage: venture lite [-h] [-f FILE] [-e EVAL] [-i] [--abstract-syntax] [-L LIBRARY] [--prelude]

VentureScript, tracing models with the Lite backend

optional arguments:
  -h, --help            show this help message and exit
  -f FILE, --file FILE  execute the given file
  -e EVAL, --eval EVAL  execute the given program
  -i, --interactive     enter interactive interpreter after completing script file
  --abstract-syntax     interpret abstract rather than concrete syntax
  -L LIBRARY, --library LIBRARY
                        load a Python file in the local directory. If the file defines a
                        __venture_start__ function, call it with the ripl to initialize the
                        library.
  --prelude             load model prelude on startup (default: off)

$ venture server -h
usage: venture server [-h] [-p PORT] backend

Start a VentureScript server

positional arguments:
  backend               the backend to trace models in

optional arguments:
  -h, --help            show this help message and exit
  -p PORT, --port PORT  the port to listen on (default 8082)

$ venture remote -h
usage: venture remote [-h] [--host HOST] [-p PORT]

Connect to a running VentureScript server

optional arguments:
  -h, --help            show this help message and exit
  --host HOST           the host to connect to (default 127.0.0.1)
  -p PORT, --port PORT  the port to connect on (default 8082)

Library Invocation

TODO

import venture.shortcuts as ven
ripl = ven.Lite().make_combined_ripl()
ripl.assume ...
ripl.observe ...
ripl.infer ...