Thoughts Rules Engine

Thoughts is a lightweight rules engine.

What's New

Mon Dec-14, 2020 Release (0.0.7)

Addded a sample for calculating Winograd Schema information with light natural language parsing. Also now the "then" portions of rules will push items to the top of the agenda, and in order. This will help favor new rules to finish their forward chaining behaviors sooner.

See winograd_1.json in https://github.com/hofmanniac/thoughts/tree/master/samples/academic/winograd for an example.

Sun Dec-13, 2020 Release (0.0.6)

You can now #tokenize a string and apply an assertion for every token in the string. You can also now use #lookup, to locate a matching fact in the context, which will then assert the matching fact. This is useful in parsing natural language, where you want to assert each word (token) in a sentence, lookup the corresponding lemma, and then match against a set of rules.

See sequence_nlp.json in https://github.com/hofmanniac/thoughts/tree/master/samples for an example in action.

Moved apply_unification from engine into thoughts.unification. Seemed the more natural spot!

Sat Dec-12, 2020 Release (0.0.5)

You can now create sequence-based rules, which wait for multiple assertions in sequence before firing. See the How to Use section below for more details.

Sun Nov-29, 2020 Release (0.0.4)

In this release, you can now load and save .json files into Context Items. See #load-json and #save-json in the Commands section below for more information.

Sat Nov-28, 2020 Release (0.0.3)

In this release, you can now load custom plugins for use in the "then" portion of rules. See "load_plugin" in the Engine Methods section below for more information.

How To Use

Add a .json file that contains your rules:

[
    {   "when": "hello",
        "then": {"#output": "hello, world!"}
    }
]  

See the samples folder in the GitHub project (https://github.com/hofmanniac/thoughts) for examples on various rules and commands.

Import the engine

from thoughts.rules_engine import RulesEngine

Start a new engine and load the rule file above

engine = RulesEngine()
engine.load_rules("rules.json")

Alternatively, you can create a manual rule without loading a file

rule = {"when": "what time is it", "then": {"#output": "time to get a new watch"}}
engine.add_rule(rule)

Define and run assertions

assertion = "hello"
engine.run_assert(assertion)

Assertions will match the "when" portion of rules, based on a unification algorithm:

• Strings will match direct string matches, "when": "hello" will match "hello"
• Strings will match using variables, "when": "my dog is ?name" will match "my dog is fido"
• Dictionaries will match a dictionary, "when": {"name": "fido"} will match {"name": "fido"}

If the assertion matches, the "then" portion will fire

Rules will "forward chain" - the "then" portion of rules will cause the engine to match against rules [ { "when": "hello", "then": {"user-intent": "greet"} },

    {   "when": {"user-intent": "greet"},
        "then": {"#output": "hello, world"}
    }
]

You can have more than one command (action) in the "then" portion

{   "when": "hello",
    "then": [{"#output": "hello there"}, 
            {"#output": "nice to meet you"}]
}

You can store "item" knowledge (facts)

{"item": "user", "name": "jeremy", "dog": "fido"}

You can reference the items (facts) and their properties in your rules, using $itemname.property syntax

{"when": "what is my name",
"then": [{"#output": "your name is $user.name"}
}

You can use sequence-based rules (chart parsing)

This is useful for natural-language type parsing where a rule needs to wait on input before firing the consequent (then) portion. In the example below, when {"cat": "art", "lemma": "the"} is asserted, the rule will match the first constituent and add the rule as an arc to the active arcs. The new arc will "wait" for another consituent with {"cat": "n", "lemma": "..."} to be asserted before matching and firing the "then" portion. Be sure to place the constituents within an array / list [] within the "when" portion.

{   "when": [
        {"cat" :"art", "lemma": "?det"},
        {"cat" :"n", "lemma": "?entity"}],
    "then": 
        {"cat": "np", "entity": "?entity", "art": "?art"}
}

Note - The active rules (arcs) will remain in memory until you clear them using engine.clear_arcs(). This is useful to assert one constituent at a time into the engine to inspect the results.

You can add your own or pip installed modules as plugins!

To do this, use load_plugin() and pass in a moniker and the "dot" path of the module. This module should already have been pip installed in the environment so that the runtime can load it, or could be a standalone module in your project.

from thoughts.rules_engine import RulesEngine
engine = RulesEngine()
engine.load_plugin("#my-module", "my_module")

Then in your rules, you can use this as a command in the "then" rules.

Engine Methods

add_rule(rule)

Adds a rule into memory.

clear_arcs()

Clears all active arcs (sequence rules in-progress) from memory.

load_rules(file)

Loads a .json rules file into memory.

load_plugin(moniker, module_namespace)

Loads a plugin (Python module), which can be used in then "then "portion of rules. Whichever module you use will need to have a process function and that function will need to take two arguments - a dict and a thoughts.Context object.

my_module.py

def process(command, context): 
    ...your logic here
    ...by convention, you can put the most relevant parameter feature into the head #my-module moniker,
    ...for example text = command["#my-module"]

Your custom module has access to the Context object, which contains all of the loaded rules and items from command that ran previously.

run_assert(assertion)

Evaluates the assertion against the loaded rules. Essentially, the evaluation will attempt to match the assertion against the "when" portion of all loaded rules.

If a rule matches, then the engine will add the "then" portion of the rule to the engine's evaluation agenda, substituting any unification variables that were determined during the "when" matching stage into the "then" items, and then evaluting them one at a time.

As each command is evaluated for assertion, the system will also substitute any values from the Context Items that are indicated in the command item.

run_console()

Runs a console input loop. Each item entered will be passed into the engine's run_assert(assertion) function for evaluation.

Entering "#log" will display the debug log.

Entering "#items" will display the Context Items.

Entering "#clear_arcs" will clear any active sequence-based rules (arcs) from memory.

Entering "#exit" will exit the console loop. Note that the "#exit" command is also passed in as an assertion one last time, in case you want to handle the exit event first in any rules.

Commands

You can use commands in the "then" portion of your rules. The engine will run the commands if the "when" portion matches.

In this version, the following commands are available.

#output

• Will echo the text to the console (using print)
• Optional: specifiy a "rate" to slow output the contents to the console {"#output": "hello, world"}

#prompt

• Will ask for input and store into an item {"#input": "what is your name", "into": "username"}

#read-rss

• Will read the specified rss feed into an item {"#read-rss": "https://rss-feed.rss", "into": "rss"}

#load-json

• Will read in a .json file into Context Items {"#load-json": "filename.json", "into": "item-name"}

#lookup

• Will match (through unification) items in the context. If found, will assert the matching item. {"#lookup": {"lemma": "dog"}}

#save-json

• Will save a Context Item into a .json file {"#save-json": "filename.json", "from": "item-name"}

#tokenize

• Will split a string into tokens (separated by spaces) and then assert each into the form specified in the "assert" argument {"#tokenize": "?text", "assert": {"#lookup": {"lemma": "#"}}}

Examples

See the samples folder in the GitHub project (https://github.com/hofmanniac/thoughts) for examples on various rules and commands.

read an rss feed and output it to the console at a readable rate

[
    {"when": "rss digg.top",
    "then": [{"#read-rss": "$?feed", "into": "rss"},
            {"#output": "$rss.title", "rate": 0.0225}]
    },

    {"item": "digg", "top": "https://digg.com/rss/top.rss"}
]