Intentclassification

Install

pip install intentclassification

About

Intentclassification AI in python, easy to use intentclassifier package.

Docs:

• Intentclassification
  • Install
  • About
  • Docs:
  • How to use
    • Import:
    • intents.json Structure:
    • Initialize Intentclassifier class:
    • Train/fit model:
    • Load model:
    • Run/predict model:
    • Handle triggers:
    • Example:

How to use

Import:

from intentclassification import IntentClassifier, handleTriggers

intents.json Structure:

{
    "intents": [
        {
            "tag": "lights_on",
            "patterns": ["Turn the light on", "Turn on the light", "Lights on"]
        },
        {
            "tag": "lights_off",
            "patterns": ["Turn the light off", "Turn off the light", "Lights off"]
        },
        {
            "tag": "fan_on",
            "patterns": ["Turn the fan on", "Turn on the fan", "Fan on"]
        },
        {
            "tag": "fan_off",
            "patterns": ["Turn the fan off", "Turn off the fan", "Fan off"]
        }
    ]
}

Initialize Intentclassifier class:

#Intents path is the path to the intents.json file
#Output path is the path to the folder the model will be saved in
IC = IntentClassifier(intents_path="./intents.json", output_path="./intentclassification")

Train/fit model:

First fit the model Aka. train the model on the data in intents.json. Save the model after fitting

IC.fit_model()
IC.save_model()

Load model:

Loading a previously fitted/trained model. When you have a fitted model you dont have to refit everytime you run the program, just simply load the model like so. The model will be loaded through the output path that is given to the IntentClassifier class when initialized.

IC.load_model()

Run/predict model:

result = IC.predict(input_text="Turn the lights on!")

Handle triggers:

Start by setting up the functions that correspond to the specified intents in intents.json. In this function you can do whatever you need if the prediction thinks that this function should be called, then it will run the code.

The Function takes 1 arguments which is passed along from the handleTriggers function, which will be explained later. If anything inside the function is returned it will be parsed through the handleTriggers() function.

def example_function(args):
    print("Arguments:", args)
    #Run whatever code is relevant
    return True

Now you can create a list containing all the trigger_functions. Remember not to call the functions inside the list.

trigger_functions = [example_function]

Parsing arguments is easy, start by creating a tuple of arguments.

arguments = ("Argument!", ["Array argument"])

Now everything can be parsed to the handleTriggers function. The function takes 4 arguments.

The prediction argument is the result from IC.predict(). The probability_threshold argument is a float value between 0 and 1, the closer to 1 the higher certainty the intent needs to run a corresponding function. The trigger_functions argument is the list of trigger functions created earlier. And the args is the tuple of arguments that we want to parse to the trigger functions.

returned_value = handleTriggers(prediction=result, probability_threshold=0.75, trigger_functions=trigger_functions, args=arguments)

The returned_value is the value returned from any of the trigger functions. In this example it will be True if the example function is predicted and executed.

Example:

Example files can be found here

from intentclassification import IntentClassifier, handleTriggers

#Initialise intentclassifier (Intents path being the path to the intents.json file) (Output path being the path to the folder the model will be saved in)
IC = IntentClassifier(intents_path="./intents.json", output_path="./intentclassification")

#Fit the model Aka. train the model on the data in intents.json
IC.fit_model()
#Save the model after fitting
IC.save_model()

#Loading a previously fitted/trained model
#When you have a trained model you dont have to retrain everytime you run the program, just simply load the model like so
IC.load_model()

#Using the model is as simple as parsing a sentence to the predict function
result = IC.predict(input_text="Turn the lights on!")

#The predict function will return a json object including the name of the intent and the probability of that intent
print(result) # Example: {"intent": "lights_on", "probability": "0.999256"}

#Now how about we use this intent for something useful?
#We can define a function that will run whenever the lights_on intent reaches a certain probability
#Remember to give the functions and the intents the same name otherwise the triggerhandler wont work
#Remember always to pass the arguments from handleTriggers into the trigger functions
def lights_on(args):
    print("Arguments:", args)
    print("Turning lights on!")
    #Run code for turning on your light here!
    #If anything in here is returned it will be parsed through the handleTriggers() function
    return "Lights have been turned on!"

def lights_off(args):
    print("Arguments:", args)
    print("Turning lights off!")
    #Run code for turning on your light here!
    #If anything in here is returned it will be parsed through the handleTriggers() function
    return "Lights have been turned off!"

def fan_on(args):
    print("Arguments:", args)
    print("Turning fan on!")
    #Run code for turning on your light here!
    #If anything in here is returned it will be parsed through the handleTriggers() function
    return "Fan has been turned on!"

def fan_off(args):
    print("Arguments:", args)
    print("Turning fan off!")
    #Run code for turning on your light here!
    #If anything in here is returned it will be parsed through the handleTriggers() function
    return "Fan has been turned off!"

#Create an array with all the trigger functions, this is how the handleTriggers() function accesses them.
trigger_functions = [lights_on, lights_off, fan_on, fan_off]

#Handletriggers takes 2 arguments, the result from the IC.predict() function and a probability threshold value for activating the trigger functions
#The probability_threshold can be changed to accomodate any missfires, the threshold might have to be changed depending on the size of intents.json
#The handleTriggers function also takes a tuple for the arguments, this tuple can be filled with any arguments that the functions in trigger_functions can then access
arguments = ("Argument!", ["Array argument"])
returned_value = handleTriggers(prediction=result, probability_threshold=0.75, trigger_functions=trigger_functions, args=arguments)

#The returned value will be None by default depending on if the triggered trigger function returns anything, otherwise it will be the returned value
print(returned_value)

# The printed results of this full program will be the following:
# {'intent': 'lights_on', 'probability': '0.9999784'}
# Arguments: ('Argument!', ['Array argument'])
# Turning lights on!
# Lights have been turned on!