app-carlo.py

import streamlit as st
import pandas as pd
import plotly.graph_objects as go
import joblib
import nltk
nltk.download('vader_lexicon')
from nltk.sentiment.vader import SentimentIntensityAnalyzer
from lime.lime_text import LimeTextExplainer

# Add a hero image
st.image('images/hero.jpg', use_column_width=True)

# Load time series data
df_resampled = pd.read_csv('data/reviews_processed.csv')
df_resampled.set_index('date', inplace=True)

# Add app title
st.title('Time Series Analysis Customer Review for Sandbar')

# User input for the time frame selection and sentiment analysis (positive/negative)
st.subheader('Select a Time Frame')
time_frame = st.slider('Time Frame (Months)',
                       min_value=1,
                       max_value=(len(df_resampled)),
                       step=1)

# Resample data according to the user-selected time frame
resampled_data = df_resampled['stars'].rolling(window=time_frame).mean()

# Plot the time series data
fig = go.Figure()

fig.add_trace(go.Scatter(
    x=df_resampled.index,
    y=df_resampled['stars'],
    mode='lines',
    name='Monthly Average'
))

# Add moving averages to the plot as lines on top
fig.add_trace(go.Scatter(
    x=df_resampled.index,
    y=resampled_data,
    mode='lines',
    name=f'{time_frame}-Month Moving Average'
))

# Add title and labels to the plot
fig.update_layout(
    title = f'Average Star Rating Over Time with {time_frame}-Monthly Moving Average',
    xaxis_title = 'Time',
    yaxis_title = 'Average Star Rating',
)

# Show the plot
st.plotly_chart(fig, use_container_width=True)

# Load the Naive Bayes model and TF-IDF vectors for sentiment analysis
naiveBayesModel = joblib.load('models/naive-bayes_model.pkl')
vectorizerTFIdf = joblib.load('models/vectorizer.pkl')

from sklearn.utils.validation import check_is_fitted

# Load vectorizer and model
vectorizerTFIdf = joblib.load('models/vectorizer.pkl')
try:
    check_is_fitted(vectorizerTFIdf, attributes=['idf_'])
    st.write("Vectorizer loaded and is fitted.")
except:
    st.write("Error: Vectorizer is not properly fitted.")

# Instantiate VADER
vader = SentimentIntensityAnalyzer()

# Initialize LIME text explainer
lime_explainer = LimeTextExplainer(class_names=['Positive', 'Neutral', 'Negative'])

# Function to get the predictions from VADER and Naive Bayes
def get_model_prediction(text):
    
    # VADER prediction
    vader_scores = vader.polarity_scores(text)
    vader_sentiment = max(vader_scores, key=vader_scores.get)
    
    # Naive Bayes prediction
    naiveBayesVectorizer = vectorizerTFIdf.transform([text])
    naiveBayesPrediction = naiveBayesModel.predict(naiveBayesVectorizer)[0]
    
    return {
        'VADER': vader_sentiment,
        'Naive Bayes': naiveBayesPrediction
    }, vader_scores
    
# Function to predict probabilities and explain them with LIME
def predict_proba(texts):
    return naiveBayesModel.predict_proba(vectorizerTFIdf.transform(texts))

# Sentiment analysis with LIME
st.header('Sentiment Analysis')

# User text input
user_input = st.text_area('Enter text for sentiment analysis:')

# Predict sentiment
if st.button('Analyze'):
    
    if user_input:
        
        # Get predictions
        predictions, vaderScores = get_model_prediction(user_input)
        
        # Display the predictions
        st.write(f'VADER Sentiment: {predictions["VADER"]}')
        st.write(f'Naive Bayes Sentiment: {predictions["Naive Bayes"]}')
        
        # Visualize model confidence
        fig = go.Figure()
        
        # Add VADER confidence
        fig.add_trace(go.Bar(
            x=list(vaderScores.keys()),
            y=list(vaderScores.values()),
            name='VADER Scores'
        ))
        
        # Add Naive Bayes confidence
        fig.add_trace(go.Bar(
            x=['Naive Bayes'],
            y=[1 if predictions['Naive Bayes'] == 'Positive' else 0],
            name='Naive Bayes Score'
        ))
        
        fig.update_layout(
            title = 'Model Sentiment Comparison',
            xaxis_title = 'Models',
            yaxis_title = 'Confidence Levels',
        )
        
        st.plotly_chart(fig, use_container_width=True)
        
        # LIME Explainer
        st.subheader('LIME Explanation for Naive Bayes')
        exp = lime_explainer.explain_instance(
            user_input,
            predict_proba,
            num_features=10
        )
        
        exp_html = exp.as_html()       
        
        # Display LIME Explanation
        st.components.v1.html(exp_html)
        
    else:
        st.write('Please provide text for sentiment analysis')