Getting started

Option 1

First you will have to solve non-trivial task of deploying LAPS on your machine. For instructions refer to Deploying LapsTrans to Ubuntu 20

Option 2

Alternatively, if you have access to the Uni Heidelberg HPC, you can find deployed instance at /home/container/lapstrans_home/LapsTrans Clone it, or use it as is. To source the correct python and opam environments, necessary for running laps execute

source /home/container/lapstrans_home/env.sourcefile

How to - Translate

To execute the translation script, navigate to the ec/ directory, type in the terminal

python3.7 lapstrans_extensions/translate.py -i <Path to the file containing the function to translate> config lapstrans_extensions/config.ini

or

python3.7 lapstrans_extensions/translate.py --cli config lapstrans_extensions/config.ini

For interactive input of functions.

The results will be written to lapstrans_extensions/working_dir/translation.json, but you could modify that path and more through command line options and/or config.ini file. For more instructions go to Configuration

How to - Data generation

To execute the training data generation script, navigate to the ec/ directory, type in the terminal

python3.7 lapstrans_extensions/generate_training_data.py -i <Path to the file containing the training functions> config lapstrans_extensions/config.ini

This generates both language data for LAPS, as well as solved task examples for the dreamcoder portion of it.

The results will be written to ec/data/list/<Output name>, but you could modify that path and more through command line options and/or config.ini file. For more instructions go to Configuration

Notes

We have built a mechanism to infer the possible input-output types for the given function, differentiating between List[int], List[bool], bool, int. Since python is dynamically typed, this might cause unexpected consequences. An obvious illustration is that + operand can be used for both list concatenation, addition, and even

>>> True+True
2

So the function that you have written with a specific purpose in mind, might very well take and return other types flawlessly.

Our code does not differentiate, except it detects identity mappings. Moreover in research we discovered that if the function returns bool or List[bool], and was detected to be an identity mapping very often this is unexpected behaviour, and thus we treat it as such. Such bool interpretation will be ignored.

In other cases, however, all legal type combinations will be processed into training data/ translation examples, and passed onto dreamcoder.

Operations such as sum(l: List) will work both with conventional integer list, and boolean list (effectively counting True elements).