1. Basic Bot Guide#

We start by importing the basic bot from botiverse.bots and gui for testing.

from botiverse.bots import BasicBot
from botiverse import chat_gui

The following is a sample from a dataset of the type which the basic bot can deal with, you can find the whole dataset in the User Guide folder.

1.1. Dataset Sample#

{
  "tag": "admissions",
  "patterns": [
    "How can I apply to the university?",
    "What are the admission requirements?",
    "Tell me about the application process"
  ],
  "responses": [
    "To apply to our university....",
    "To begin your application process, please refer to the 'Admissions'..."
  ]
}

1.1.1. Read the Data#

We start by making an instance of the basic chatbot. machine is the underlying machine learning model to use. The basic bot supports nn and svm for now or any custom model that satisfies a similar fit and predict interface (as found in the documentation). repr is the representation to use for the data. The basic bot supports tf-idf, glove, tf-idf-glove, bow for now or any custom representation that satisfies a similartransform and transform_list interface

bot = BasicBot(machine='nn', repr='glove')

After making an instance of the bot with a neural network model and with the GloVe as the word representation, we read let the chatbot read the data.

bot.read_data('dataset.json')

1.1.2. Train and save the chatbot#

The chatbot can be trained at this point on the data has read. Generally, the train method can take all the hyperparameters that the fit method of the underlying model can take. In our case for SVM or NN it just uses good defaults.

bot.train()
bot.save('BasicBot')

Train Acc: 1.0: 100%|██████████| 240/240 [00:01<00:00, 183.84it/s]

1.1.3. Infer#

Once the chatbot has been trained, the infer method can be called to get a response to a given input. In general, infer also take a confidence parameter to decide when the chatbot should say “I don’t know”

bot.infer("Hello there! ")

"The university campus provides a wide range of facilities to support academic, social, and recreational needs. From well-equipped classrooms and libraries to sports facilities and student clubs, there's something for everyone. Check out our website's 'Campus Life' section to discover more about the available amenities."

1.1.4. Deploy the Chatbot#

To demonstrate working in a real backend, we consider first loading the bot from its file. The dataset is necessary as it includes the responses and is too small anyway.

newbot = BasicBot(machine='nn', repr='glove')
newbot.load('BasicBot', 'dataset.json')

We can run the GUI by running the following. If server is not set to true then a much more humble version of the GUI will run locally using Python’s input() and print() functions.

chat_gui('Basic Bot', bot.infer, server=True)

1.1.5. Using a Custom Model and Transform#

In this we demonstrate how a custom model or representation can be used with the basic bot. We start by importing a neural network and GloVe that satisfy the expected interfaces.

from botiverse.models import NeuralNet
from botiverse.preprocessors import GloVe

We make instances of both to later pass to the chatbot

model = NeuralNet(structure=[50, 12, 8], activation='sigmoid')
transform = GloVe()

After passing them to the basic bot, we follow up with the normal pipeline.

bot = BasicBot(machine=model, repr=transform)
bot.read_data('dataset.json')

Notice how the train method is flexible to take all the parameters needed by the model’s fit.

bot.train(batch_size=1, epochs=30, λ = 0.02, eval_train=True, val_split=0.0)

Train Acc: 0.71: 100%|██████████| 30/30 [00:00<00:00, 220.55it/s]

newbot.infer("Thanks")

'Glad I could help! If you have any more questions, feel free to ask.'

Basic Bot Guide

Contents