toxicity_classification/app.py

import pickle

import gradio as gr
from keras.models import load_model
from keras.preprocessing.sequence import pad_sequences

from src.config import *
from src.data_cleaning import clean_text
# -------------------------------------------------------------------------

lstm_model = load_model(MODEL_LOCATION)

with open(TOKENIZER_LOCATION, 'rb') as handle:
    tokenizer = pickle.load(handle)

# -------------------------------------------------------------------------


def make_prediction(input_comment):

    input_comment = clean_text(input_comment)
    input_comment = input_comment.split(" ")

    sequences = tokenizer.texts_to_sequences(input_comment)
    sequences = [[item for sublist in sequences for item in sublist]]

    padded_data = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH)
    result = lstm_model.predict(padded_data, len(padded_data), verbose=1)

    return \
        {
            "Toxic": str(result[0][0]),
            "Very Toxic": str(result[0][1]),
            "Obscene": str(result[0][2]),
            "Threat": str(result[0][3]),
            "Insult": str(result[0][4]),
            "Hate": str(result[0][5]),
            "Neutral": str(result[0][6])
        }


comment = gr.inputs.Textbox(lines=20, placeholder="Enter your comment!!")

title = "Toxic Comment Classifier"
description = "This application uses a Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN) " \
              "model to predict the inappropriateness of a comment"

gr.Interface(fn=make_prediction,
             inputs=comment,
             outputs="label",
             title=title,
             description=description).launch()