This section is meant to give an overview of design decisions for practical implementation of the theory provided above. It is build up in the following way: we start off with a naive implementation and then optimize this in later subsections.
The code accompanying this subsection can be found here.
Design decisions:
- Ordering or program enumeration (lexicographic)
- Programs that halted do not benefit from longer run times, only consider programs with runtime limit.
- Profile (short circuit of array equality)
Python requirements: pip install -r requirements.txt
For generating images and animations:
- imagemagick: https://imagemagick.org/script/download.php
- pdflatex: https://miktex.org/download
- ghostscript: https://www.ghostscript.com/download/gsdnld.html
- update
config.pywith the paths
usage: levin_search.py [-h] [--version] [--work_tape_size WORK_TAPE_SIZE]
[--program_tape_size PROGRAM_TAPE_SIZE]
[--primitives_set {DEFAULT,WEIGHT}]
[--search_log SEARCH_LOG]
[--solutions_file SOLUTIONS_FILE]
{POSITION,COUNT,EVEN,ODD,FIZZ,FIZZ_COMPLETE,BUZZ,BUZZ_COMPLETE,FIZZBUZZ,FIZZBUZZ_COMPLETE,NEGATIVE_ONE,NEGATIVE_ONE_TWO_THREE}
search_length solutions_dir
Levin search for Discovering Low Complexity Neural Network Weights
positional arguments:
{POSITION,COUNT,EVEN,ODD,FIZZ,FIZZ_COMPLETE,BUZZ,BUZZ_COMPLETE,FIZZBUZZ,FIZZBUZZ_COMPLETE,NEGATIVE_ONE,NEGATIVE_ONE_TWO_THREE}
Which task to run? Tasks can be defined in `task.py`
search_length Search for programs up to this length.
solutions_dir Directory to store solutions, created if not exists
optional arguments:
-h, --help show this help message and exit
--version show program's version number and exit
--work_tape_size WORK_TAPE_SIZE
The size of the work tape
--program_tape_size PROGRAM_TAPE_SIZE
The size of the program tape
--primitives_set {DEFAULT,WEIGHT}
Use the following set of primitives
--search_log SEARCH_LOG
Store the log of the deterministic levin search
process (.csv).
--solutions_file SOLUTIONS_FILE
Store the solutions found in this file (.json)
Example usage Count task:
python levin_search.py COUNT 4usage: run_program.py [-h] [--version] [--work_tape_size WORK_TAPE_SIZE]
[--program_tape_size PROGRAM_TAPE_SIZE]
[--n_weights N_WEIGHTS]
[--primitives_set {DEFAULT,WEIGHT}]
{file,string} ... log_file
Program Running for Discovering Low Complexity Neural Network Weights
positional arguments:
{file,string}
log_file Store the logs in this file (.jsonl)
optional arguments:
-h, --help show this help message and exit
--version show program's version number and exit
--work_tape_size WORK_TAPE_SIZE
The size of the work tape
--program_tape_size PROGRAM_TAPE_SIZE
The size of the program tape
--n_weights N_WEIGHTS
The number of weights
--primitives_set {DEFAULT,WEIGHT}
Use the following set of primitives
Example to run from file:
python run_program.py file program_file.txt program_log.jsonlExample to run from command line string:
python run_program.py string 1,0,2,0 program_log.jsonlusage: program_convert.py [-h] [--version] [--work_tape_size WORK_TAPE_SIZE]
[--program_tape_size PROGRAM_TAPE_SIZE]
program_file {table,animation} ...
Levin search for Discovering Low Complexity Neural Network Weights
positional arguments:
program_file Which log file to use?
{table,animation}
table Table with the program instructions
animation Animate the program run
optional arguments:
-h, --help show this help message and exit
--version show program's version number and exit
--work_tape_size WORK_TAPE_SIZE
The size of the work tape
--program_tape_size PROGRAM_TAPE_SIZE
The size of the program tape
Example usage for generating a table:
python program_convert.py program_file.jsonl table program_table.mdExample usage for generating an animation:
python program_convert.py program_file.jsonl animation animation.gif