utils.py

import re
import pickle
import numpy as np
import pandas as pd


tfidf = pickle.load(open('models/tfidf.sav', 'rb'))
svc_sentiment = pickle.load(open('models/sentiment_model.sav', 'rb'))
tfidf_sentiment = pickle.load(open('models/tfidf_sentiment.sav', 'rb'))
svc = pickle.load(open('models/svc_model.sav', 'rb'))

labels = [
       'Product quality', 'Knowledge',
       'Appointment', 'Service etiquette', 'Waiting time',
       'Repair speed', 'Repair cost', 'Repair quality', 'Warranty',
       'Product replacement', 'Loan sets']

sample_file = pd.read_csv('sample.csv').to_csv(index=False).encode('utf-8')

print('utils imported!')

def get_single_prediction(text):
    
    # manipulate data into a format that we pass to our model
    text = text.lower().strip() #lower case

    # Vectorise text and store in new dataframe. Sentence vector = average of word vectors
    text_vectors = tfidf.transform(list(text))

    # Make topic predictions
    results = svc.predict_proba(text_vectors).squeeze().round(2)
    pred_prob = pd.DataFrame({'topic': labels, 'probability': results}).sort_values('probability', ascending=True)
    
    # Make sentiment predictions
    text_vectors_sentiment = tfidf_sentiment.transform(list(text))
    
    results_sentiment = svc_sentiment.predict_proba(text_vectors).squeeze().round(2)
    pred_prob_sentiment = pd.DataFrame({'sentiment': ['Negative', 'Positive'], 'probability': results_sentiment}).sort_values('probability', ascending=True)
    

    return (pred_prob, pred_prob_sentiment)

def get_multiple_predictions(csv):
    
    df = pd.read_csv(csv)
    df.columns = ['sequence']

    df['sequence_clean'] = df['sequence'].str.lower().str.strip()

    # Remove rows with blank string
    invalid = df[(pd.isna(df['sequence_clean'])) | (df['sequence_clean'] == '')]
    invalid.drop(columns=['sequence_clean'], inplace=True)
    
    # Drop rows with blank string
    df.dropna(inplace=True)
    df = df[df['sequence_clean'] != ''].reset_index(drop=True)
    
    # Vectorise text and get topic predictions 
    text_vectors = tfidf.transform(df['sequence_clean'])
    pred_results = pd.DataFrame(svc.predict(text_vectors), columns = labels)

    # Vectorise text and get sentiment predictions 
    text_vectors_sentiment = tfidf_sentiment.transform(df['sequence_clean'])
    pred_results_sentiment = pd.DataFrame(svc_sentiment.predict(text_vectors_sentiment), columns = ['sentiment'])    
    
    # Join back to original sequence
    final_results = df.join(pred_results).join(pred_results_sentiment)
    final_results['others'] = final_results[labels].max(axis=1)
    final_results['others'] = final_results['others'].apply(lambda x: 1 if x == 0 else 0)
    
    final_results.drop(columns=['sequence_clean'], inplace=True)
    
    # Append invalid rows
    if len(invalid) == 0:
        return final_results.to_csv(index=False).encode('utf-8')
    else:
        return pd.concat([final_results, invalid]).reset_index(drop=True).to_csv(index=False).encode('utf-8')