Entropy-controlled contexts in Python

Overview

Build Status PyPI version Linux PyPI status

Twitter Follow Twitter URL

Python module ordered

ordered module is the opposite to random - it maintains order in the program.

import random 
x = 5
def increase():
    global x
    x += 7
def decrease():
    global x
    x -= 2

while x != 22:  
    random.choice([increase, decrease])()  
    # takes long time to exit ...

vs.

import random, ordered
x = 5
def increase():
    global x
    x += 7
def decrease():
    global x
    x -= 2

with ordered.orderedcontext(): # entropy-controlled context
    while x != 22: 
        random.choice([increase, decrease])()  
    # exits immediately with correct result
pass 

Ordered contexts are environments of controlled entropy. Contexts allow you to control which portions of the program will be guaranteed to exit with minimum state-changing steps. Raising any exceptions is also avoided by providing the correct "anti-random" choice() results.

Usage

ordered is a Python 3.8+ module. Use cases include automated decisionmaking, manufacturing control, robotics, automated design, automated programming and others.

You describe the world as Python objects and state-modifying methods. Defining an entropy-controlled context allows you to set up a goal for the system to satisfy all constraints and reach the desired state.

To define constraints you add assert statements in your code and inside ordered context. Then you add a function-calling loop while : random.choice()(random.choice()) . To exit the context the engine will have to call correct functions with correct arguments and end up with a staisfying state (see examples below).

Requirements

  • Linux (tested on Ubuntu 20.04+)
  • Python 3.8 in virtualenv
  • Recommended: PyPy compatible with Python 3.7+, installed globally.
# In Python3.8 virtualenv on Linux:
$ pip install ordered

Entropy Context Objects

# ... normal python code
with ordered.orderedcontext():  
    # ... entropy-controlled context, guaranteed to exit without exceptions
# ... normal python code
  • ordered.orderedcontext()

    Return a context manager and enter the context. SchedulingError will be raised if exit is not possible.

    Inside ordered context functions random.choice and ordered.choice are equivalent and no randomness is possible. If choice() is called without parameters then gc.get_objects() (all objects in Python heap) is considered by default.

    Optional returned context object allows to set parameters and limits such as timeout and max_states.

    Warning: not all Python features are currently supported and thus ordered might fail with internal exception. In this case a rewrite of user code is needed to remove the usage of unsupported features (such as I/O, lists and for loops.)

    Warning: ordered requires all entropy-controlled code to be type-hinted.

# ...
def decrease():
    global x
    assert x > 25  # when run inside context this excludes cases when x <= 25
                   # thus increasing amount of overall steps needed to complete
    x -= 2
# ...
with ordered.orderedcontext(): # entropy-controlled context
    while x < 21:  # exit if x >= 21
        random.choice([increase, decrease])()  
    assert x < 23  # only x == 21 or 22 matches overall

ordered.choice()

  • ordered.choice(objects=None)

    Choose and return the object that maintains maximum order in the program (minimum entropy). Any exception increases entropy to infinity so choices leading to exceptions will be avoided. Inside the entropy controlled context, random.choice is equivalent to ordered.choice (and also random.choices in the sense that it may return any amount of parameters when used as argument-generator in choice(*choice())).

    objects is a list of objects to choose from. If objects is None then gc.get_objects() is assumed by default.

    Warning: current implementation of while ... ordered loop is hard-coded to the form shown in examples. while loops with other statements than a single-line choice() are not supported. Add your code to other parts of context and/or functions and methods in your program

ordered.side_effect(lambda: )

  • ordered.side_effect(lamdba=[lambda function])

    Execute the supplied lambda function as a side-effect avoiding the compilation and subsequent effect analysis by ordered. This is useful when I/O is easier schdeuled right within the entropy-controlled part of the program or when you know that the code to be executed has no useful effect on the state of the problem of interest.

    side_effect may only be used when importred into global namespace using from ordered import side_effect

    from ordered import side_effect
    
    def move(t: Truck, l: Location):
        "Move truck to any adjacent location"
        assert l in t.location.adjacent
        t.locaiton = l
        t.distance += 1
        side_effect(lambda: print(f"This {__name__} code can have any Python construct and is not analysed. Current value is {t.distance}"))

Examples:

Object Oriented Code

Preferred way of implementing software models with ordered is object-oriented:

import ordered

class MyVars:
    x: int
    steps: int
    def __init__(self) -> None:
        self.x = 0
        self.steps = 0

    def plus_x(self):
        self.x += 3
        self.count_steps()

    def minus_x(self):
        self.x -= 2
        self.count_steps()
    
    def count_steps(self):
        self.steps += 1

m = MyVars()
m.x = 5
with ordered.orderedcontext():
    while m.x != 12:  
        ordered.choice()()  

print("Steps:", steps)

Pouring problem

A classic bottle pouring puzzle. You are in the possession of two bottles, one with a capacity of 3 litres and one with a capacity of 5 litres. Next to you is an infinitely large tub of water. You need to measure exactly 4 litres in one of the bottles. You are only allowed to entirely empty or fill the bottles. You can't fill them partially since there is no indication on the bottles saying how much liquid is in them. How do you measure exactly 4 litres?

from ordered import orderedcontext, choice
class Bottle:
    volume: int
    fill: int
    def __init__(self, volume: int):
        self.volume = volume
        self.fill = 0
    def fill_in(self):
        self.fill += self.volume
        assert self.fill == self.volume
    def pour_out(self, bottle: "Bottle"):
        assert self != bottle
        can_fit = bottle.volume - bottle.fill
        sf = self.fill
        bf = bottle.fill
        if self.fill <= can_fit:
            bottle.fill += self.fill
            self.fill = 0
            assert self.fill == 0
            assert bottle.fill == bf + sf
        else:
            bottle.fill += can_fit
            self.fill -= can_fit
            assert bottle.fill == bottle.volume
            assert self.fill == sf - can_fit
    def empty(self):
        self.fill = 0
b1 = Bottle(3)
b2 = Bottle(5)
with orderedcontext():
  while b2.fill != 4: 
      choice([Bottle])()
pass

NOTE: Be careful with importing from a module into global namespace and using choice()() without parameters in global scope. Current implementation load all global objects including the orderedcontext and choice and cause an error

Learning a function

ordered can be used

from ordered import choice, orderedcontext
from dataclasses import dataclass 

@dataclass
class Point:
   x: int
   y: int
   
data = [Point(1,1), Point(2,4), Point(3,9)]

# TODO: create_function creates a nonrandom function out of Node objects with `ordered.choice`
# TODO: run_function runs a node tree with a value and returns result
    
with orderedcontext():
    f = create_function()
    for point in data:
        assert run_function(f, point.x) == point.y
# context exit guarantees that create_function() constructs a correct function to describe input

# TODO: approximate function learning example

Work-in-progress functions

ordered.relaxedcontext()

Guaranteed to find an exit. Modifies the program if required.

Method ordered.def(heap_in_out: List)

Defines a function from a list of input and output heaps. The more examples of heaps are supplied, the better is the function.

Status

Although the system is in use by several industry organizations, ordered is under heavy development. Expect rapid changes in language support, performance and bugs.

Limitations

Python Language

Overall we have a relatively complete support of 'basic' use of object-oriented programming style. However, there are some hard limitaions and work-in-progress items that are yet to be documented.

Try to avoid multiline code as we have several places where line continuation may break during compilation.

Built-ins support is minimal. No I/O can be executed except for in explicit side_effect() calls.

None of the "ordered data structures" are supported: this includes list, dict and tuple. Use set or create your own data structures based on objects and classes.

Loops are not supported, including while and for besides the main while..choice() loop as described above - define your problem by creating functions that can be iteratively called by while.. choice() to overcome this.

Support of missing features is a current work in progress.

Integer Math

Math implementation is simple and works up to count 20-50 depedning on available resources. Future development includes switching to register-based math and monotonic-increase heuristics to support any numbers.

Symbolic Execution Performance

Current implementaion of Python code compilation is naive and doesn't scale well. The simpler your code, the faster it will compile. Future development includes implementing smarter symboic execution heuristics, pre-calculated database and statistical methods.

Model Universality

Current model can efficiently handle a limited set of problem classes and might require significantly more resources than would be needed with a more complete model. HyperC team provides more complete models for specific industry per request. Future development includes adding a universal pruning based on statistical methods as amount of data available to HyperC team grows.

Science behind ordered

ordered is based on translating a Python program to AI planning problem and uses a customized fast-downward as a backend. Additionally, we're implementing machine learning and pre-computed matrices on various levels to vastly improve performance on larger problems.

Contributing

For any questions and inquries please feel free contact Andrew Gree, [email protected].

Support

Module ordered is maintained by HyperC team, https://hyperc.com (CriticalHop Inc.) and is implemented in multiple production envorinments.

Investor Relations

HyperC is fundraising! Please contact at [email protected].

Owner
HyperC
Production-ready artificial Intelligence-based scheduling and planning ("AI planning")
HyperC
Regression Metrics Calculation Made easy

Regression Metrics Mean Absolute Error Mean Square Error Root Mean Square Error Root Mean Square Logarithmic Error Root Mean Square Logarithmic Error

Ashish Patel 12 Jan 02, 2023
A Randomizer Oracle

Tezos Randomizer Tezod Randomizer "Oracle". It's a smart contract that you can call to get a random number between X and Y (for now). It uses entropy

Asbjorn Enge 19 Sep 13, 2022
This is a package that allows you to create a key-value vault for storing variables in a global context

This is a package that allows you to create a key-value vault for storing variables in a global context. It allows you to set up a keyring with pre-defined constants which act as keys for the vault.

Data Ductus 2 Dec 14, 2022
Abby's Left Hand Modifiers Dictionary

Abby's Left Hand Modifiers Dictionary Design This dictionary is inspired by and

12 Dec 08, 2022
Dependency injection lib for Python 3.8+

PyDI Dependency injection lib for python How to use To define the classes that should be injected and stored as bean use decorator @component @compone

Nikita Antropov 2 Nov 09, 2021
A library for interacting with Path of Exile game and economy data, and a unique loot filter generation framework.

wraeblast A library for interfacing with Path of Exile game and economy data, and a set of item filters geared towards trade league players. Filter Ge

David Gidwani 29 Aug 28, 2022
Retrying library for Python

Tenacity Tenacity is an Apache 2.0 licensed general-purpose retrying library, written in Python, to simplify the task of adding retry behavior to just

Julien Danjou 4.3k Jan 05, 2023
A sys-botbase client for remote control automation of Nintendo Switch consoles. Based on SysBot.NET, written in python.

SysBot.py A sys-botbase client for remote control automation of Nintendo Switch consoles. Based on SysBot.NET, written in python. Setup: Download the

7 Dec 16, 2022
RapidFuzz is a fast string matching library for Python and C++

RapidFuzz is a fast string matching library for Python and C++, which is using the string similarity calculations from FuzzyWuzzy

Max Bachmann 1.7k Jan 04, 2023
Blender 2.93 addon for loading Quake II MD2 files

io_mesh_md2 is a Blender 2.93 addon for importing Quake II MD2 files.

Joshua Skelton 11 Aug 31, 2022
Conveniently measures the time of your loops, contexts and functions.

Conveniently measures the time of your loops, contexts and functions.

Maciej J Mikulski 79 Nov 15, 2022
tade is a discussion/forum/link aggregator application. It provides three interfaces: a regular web page, a mailing list bridge and an NNTP server

tade is a discussion/forum/link aggregator application. It provides three interfaces: a regular web page, a mailing list bridge and an NNTP server

Manos Pitsidianakis 23 Nov 04, 2022
This two python programs can convert km to miles and miles to km

km-to-miles These two little python programs can convert kilometers to miles and miles to kilometers Needed Python3 or a online python compiler with t

Chandula Janith 3 Jan 30, 2022
Obsidian tools - a Python package for analysing an Obsidian.md vault

obsidiantools is a Python package for getting structured metadata about your Obsidian.md notes and analysing your vault.

Mark Farragher 153 Jan 04, 2023
Hot reloading for Python

Hot reloading for Python

Olivier Breuleux 769 Jan 03, 2023
A set of Python scripts to surpass human limits in accomplishing simple tasks.

Human benchmark fooler Summary A set of Python scripts with Selenium designed to surpass human limits in accomplishing simple tasks available on https

Bohdan Dudchenko 3 Feb 10, 2022
Simple script to export contacts from telegram into vCard file

Telegram Contacts Exporter Simple script to export contacts from telegram into vCard file Getting Started Prerequisites You must to put your Telegram

Pere Antoni 1 Oct 17, 2021
Playing with python imports and inducing those pesky errors.

super-duper-python-imports In this repository we are playing with python imports and inducing those pesky ImportErrors. File Organization project │

James Kelsey 2 Oct 14, 2021
This tool analyzes the json files generated by stream-lnd-htlcs to find hidden channel demand.

analyze_lnd_htlc Introduction Rebalancing channels is an important part of running a Lightning Network node. While it would be great if all channels c

Marimox 4 Dec 08, 2022
Display your calendar on the wallpaper.

wallCal Have your calendar appear as the wallpaper. disclaimer Use at your own risk. Don't blame me if you miss a meeting :-) Some parts of the script

7 Jun 14, 2022