alvan-text-predictor 1.1.1

The text prediction module for ALVAN 5.

A5-Server-Text-Prediction-Module

Module to predict which action a query best fits

To Install

pip install alvan-text-predictor

To Use

from prediction import predictor as pd

predictor = pd.Predictor('intents.json')

predictor.query('turn the lights off')

Predictor(intents_filename: str, pickle_file: str="data.pickle", intents_cache_file: str="intents_cache") -> Predictor

The constructor for the Predictor class takes in 1 required and 2 optional parameters: intents_filename, pickle_file, and intents_cache_file

Parameter	Type	Required	Default	Description
intents_filename	str	☑	""	Filename for the intents file, used to train the model.
pickle_file	str	☐	"data.pickle"	Filename for the file that contains/will contain the serialized model-training output
intents_cache_file	str	☐	"intents_cache"	Filename for the intents cache. This is used to detect if any changes have been made to the original intents file. If the contents of this file differ from the file passed into the intents_filename, the model will be retrained and the contents of this file and the pickle_file will be overwritten with the contents of the intents_filename file, and the serialized output of the training.

query(query)

The query method takes in a query, compares it to the trained intents, and replies with an object indicating which intent it most closely matches.

The output has the following shape:

{intent tag, intent context_set, intent responses, intent flags}

Example

predictor.query('Turn on the lights') -> {'lights_on', None, 4, {"followup": 0}}

based on example below

Intents File

The intents file is a json file with the following shape:

{"intents": [
    <List of Objects representing intents with the following shape:
    {
        "tag":<String - human-readable descriptor>,
        "patterns": [<List of Strings - various ways a user could phrase their query],
        "responses": <Int - unique id of the intent>,
        "context_set":<Object - currently unused metadata>,
        "flags": <Object with boolean values for flags used by the consumer>
    }
    >
]}

Example:

{"intents": [
    {
        "tag":"lights_on",
        "patterns": [
            "turn on the lights",
            "it is too dark in here",
            "I want the lights on",
            "I want it brighter in here"
        ],
        "responses":4,
        "context_set": None,
        "flags": {"followup": 0}
    },
    {
        "tag":"lights_off",
        "patterns": [
            "turn off the lights",
            "it is too bright in here",
            "I want the lights off",
            "I want it darker in here"
        ],
        "responses":5,
        "context_set": None,
        "flags": {"followup": 0}
    },
]}

Corequisites

nltk must be installed alongside.

pip install nltk