• Web PPL
• Venture

Read paper for details

To run a shell: python run_program.py [-e ENGINE]

To run a program: python run_program.py [-e ENGINE] [-p PROGRAM]

Flags:

ENGINE:  r = randomdb, t = traces, rt = reduced traces
PROGRAM:  The directory of some Church program.

All servers conform to a common REST API.

To run the python server: python server.py [-e ENGINE] [-v] [-p PORT]

To run a C server: python server.py [-s BINARY] [-v] [-p PORT]

Flags:

ENGINE:  r = randomdb, t = traces, rt = reduced traces
BINARY:  The directory of some C binary
PORT: The number of some port
The -v flag toggles whether the server should print its activity

See ec2_setup.txt

You can also compile a faster version of the server by doing the following:

1. (OPTIONAL) Open socket_server.py, and change engine_type to what you'd like.
2. Install pypy.
3. Run (replace the path to the pypy folder as appropriate): python /Applications/pypy/rpython/bin/rpython socket_server.py For a tracing JIT version (NOT YET READY): python /Applications/pypy/rpython/bin/rpython --opt=jit socket_server.py WARNING: This can take a long time.
4. This should create a binary called socket_server-c. You may want to rename it

See values.py for class definition

See xrp.py for examples

Remember to set self.sample, depending on whether your XRP is a sampler or rescorer!

See directives.py for examples. (Search for "PRIMITIVE PROCEDURES".) The syntax for expressions can be found near the bottom of expressions.py (or you can simply follow the examples). Don't forget the 3rd argument must be set to 'True'.

• Propagation should use a priority queue (it is currently wrong, sometimes)
• Traces could be more intelligent at propagating up through arguments (and then, we can use the proper XRP if)
• Some Jenkins tests fail
• Outermost observe shouldn't be required
• Assumes resampling XRPs are all deterministic
• Needs to use unsample