Skip to content

wslee/euphony

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Euphony

Euphony: a probabilistic model-guided program synthesizer

Build (tested on Linux)

$ ./build
$ . bin/setenv

Learning a PHOG model from training data

$ bin/run_phog_learner [filename for PHOG] [training instances]

Here, "training instaces" are just SyGuS problems along with solutions. Any functions defined in a SyGuS formulation are regarded as solutions of the SyGuS problem and used for learning. For example, at the bottom of benchmarks/string/train/bikes.sl, you can see a function definition

(define-fun f_1 ((name String)) String (str.substr name 0 (- (str.len name) 3)))

which will be regarded as a solution of the problem of bikes.sl.

Example. Learning a PHOG for the STRING domain

$ bin/run_phog_learner phog_str benchmarks/string/train/*.sl

You will be able to see file phog_str generated. You can use the PHOG to guide the search as follow:

$ bin/run_with_new_phog phog_str benchmarks/string/test/phone-5.sl 

which will be much fater than solving using EUSolver (bin/run_string_eusolver) In a similar manner, you can learn PHOGs for the BITVEC and CIRCUIT domains as well using benchmarks/bitvec/train/*.sl and benchmarks/circuit/train/*.sl

Choosing hyper-parameters

There are parameters that should be properly determined for learning. On the top of bin/run_phog_learner, you can find several variables. Increasing lambda_val can avoid overfitting. By increasing values of max_iter* and pool_size, you can put more computing resources for learning better models.

Reproducing the experimental results in the paper

# Run the experiments
$ ./artifact [string | bitvec | circuit] [--timeout <sec> (default: 3600)] [--memory <GB> (default: 16)]
# Table 4,5,6
$ ./artifact [string | bitvec | circuit] --timeout 3600
# Table 4,5,6 without EUSOLVER
$ ./artifact [string | bitvec | circuit] --timeout 3600 --only_euphony
# Figure 8
$ ./artifact [string | bitvec | circuit] --timeout 3600 --only_euphony --strategy [pcfg | uniform | pcfg_uniform]

Reproducing Table 7 (comparison between Euphony and FlashFill)

Running Euphony

$ ./artifact string_flashfill --timeout 600 --only_euphony [--memory <GB> (default: 16)]

Running FlashFill

  1. Modify the first line in "bin/run.ps1" to set an output directory.
  2. Launch Windows PowerShell and run the PowerShell script "bin/run.ps1".

Run Euphony on a single SyGus file

$ ./bin/run_[string | bitvec | circuit] [a SyGuS input file]
# For example
$ ./bin/run_string benchmarks/string/test/exceljet1.sl

About

Euphony: a probabilistic model-guided program synthesizer

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published