pages/Project_1.1_-_LLM.py

import pandas as pd
import streamlit as st
import datetime
import socket

from src.functions_llm import load_model, compare_label_results
from src.functions_db import connect_to_db, Activity, LLM


list_of_models = [
    "1mpreccable/10k_trained_bert",
    "naloui/results",
    "another model from hub"
] 

session, user, activity, llm, _ = connect_to_db(address="sqlite:///src/databases/main.db") # connect to the database


## --------------
## SIDEBAR PARAMS
## --------------

st.sidebar.title("App parameters")

model_url = st.sidebar.selectbox("Choose your model", list_of_models)
st.sidebar.divider()
imported_df = st.sidebar.file_uploader("Upload your df in format csv or xlsx", type=["csv", 'xlsx'])

## --------------
## MAINBAR PARAMS
## --------------
tab1, tab2 = st.tabs(["LLM", "DB_Extraction"])

tab1.title("LLM project")

# Display model description
if model_url:
    tab1.write(f"Selected model: **{model_url}**")

# Sentence for analysis
input_sentence = tab1.text_area("Enter a sentence for sentiment analysis")
button_launch = tab1.button('Launch Prediction')

#logic for single sentence inputed by user manually
if model_url and input_sentence:
    if model_url == "another model from hub":
        model_url = tab1.text_input("Please provide the model URL here:")
    if button_launch:

        # Add activity to the database
        activity.add_activity(
            session, 
            "admin", 
            datetime.date.today(), 
            datetime.datetime.now(), 
            socket.gethostbyname(socket.gethostname()),
            "LLM project"
        )

        sentiment_analyzer = load_model(model_url)
        
        if sentiment_analyzer:
            # Get sentiment prediction
            result = sentiment_analyzer(input_sentence)
            
            # Display result
            if result:
                sentiment = result[0]['label']
                score = result[0]['score']
                tab1.write(f"Sentiment: **{sentiment}** with a confidence score of **{score:.2f}**")

                if sentiment == 'LABEL_1' or sentiment == 'POSITIVE':
                    tab1.write("Positive")  
                elif sentiment == 'LABEL_0' or sentiment == 'NEGATIVE':
                    tab1.write("Negative")

                # Option to download results
                results_df = pd.DataFrame([{'Sentence': input_sentence, 'Sentiment': sentiment, 'Score': score}])
                csv = results_df.to_csv(index=False).encode('utf-8')
                tab1.download_button(
                    label="Download results as CSV",
                    data=csv,
                    file_name='sentiment_analysis_results.csv',
                    mime='text/csv',
                )
                
                # Add LLM history to the database
                llm.add_llm(
                    session,
                    "admin", 
                    input_sentence,
                    sentiment,
                    model_url,
                    "LLM project"
                )

#logic for file uploaded by user
if model_url and imported_df:
    if imported_df.name.endswith('csv'):
        df = pd.read_csv(imported_df, encoding='unicode_escape')
        # print(df.head())
        df = df[:500]  # Limit to 500 rows while testing

        name_of_the_column = tab1.selectbox("Select the column for sentiment analysis", df.columns)
        column_with_real_results = tab1.selectbox("Select the column with real results", df.columns)

        sentiment_analyzer = load_model(model_url)
        if sentiment_analyzer and name_of_the_column and button_launch and column_with_real_results:
            text_data = df[name_of_the_column].tolist()

            # Initialize progress bar
            progress_bar = tab1.progress(0)
            results = []
            for i, text in enumerate(text_data):
                result = sentiment_analyzer(text)
                results.append(result[0])
                # Update progress bar
                progress_bar.progress((i + 1) / len(text_data))
            
            
            df['Sentiment'] = [res['label'] for res in results]
            df['Score'] = [res['score'] for res in results]

            df['Sentiment_Label'] = df['Sentiment'].apply(lambda x: 'Positive' if x in ['LABEL_1', 'POSITIVE'] else 'Negative')
            
            tab1.write(df)

            # here add logic for calculating % of positive and negative reviews and compare with the real results
            result = compare_label_results(df, 'Sentiment_Label', column_with_real_results)
            tab1.write(f"Accuracy: {result:.2%}")

            # Option to download results
            csv = df.to_csv(index=False).encode('utf-8')
            tab1.download_button(
                label="Download results as CSV",
                data=csv,
                file_name='sentiment_analysis_results.csv',
                mime='text/csv',
            )
            #need to add % of positive and negative reviews and compare with the real results
    elif imported_df.name.endswith('xlsx'):
        df = pd.read_excel(imported_df)
        # print(df.head())
        df = df[:500] # Limit to 500 rows while testing

        name_of_the_column = tab1.selectbox("Select the column for sentiment analysis", df.columns)
        column_with_real_results = tab1.selectbox("Select the column with real results", df.columns)
        
        sentiment_analyzer = load_model(model_url)
        if sentiment_analyzer and name_of_the_column and button_launch and column_with_real_results:
            text_data = df[name_of_the_column].tolist()
            # Initialize progress bar
            progress_bar = tab1.progress(0)
            results = []
            for i, text in enumerate(text_data):
                result = sentiment_analyzer(text)
                results.append(result[0])
                # Update progress bar
                progress_bar.progress((i + 1) / len(text_data))
            
            df['Sentiment'] = [res['label'] for res in results]
            df['Score'] = [res['score'] for res in results]

            df['Sentiment_Label'] = df['Sentiment'].apply(lambda x: 'Positive' if x in ['LABEL_1', 'POSITIVE'] else 'Negative')
            
            # here add logic for calculating % of positive and negative reviews and compare with the real results
            result = compare_label_results(df, 'Sentiment_Label', column_with_real_results)
            tab1.write(f"Accuracy: {result:.2%}")
            
            tab1.write(df)
            

            # Option to download results
            csv = df.to_csv(index=False).encode('utf-8')
            tab1.download_button(
                label="Download results as CSV",
                data=csv,
                file_name='sentiment_analysis_results.csv',
                mime='text/csv',
            )
else:
    tab1.write("Please upload a file in csv or xlsx format")


################################################################################
# TAB 2 DB 
################################################################################

llm_histories = session.query(LLM).all()

llm_histories_df = pd.DataFrame([{
    'id': llm_history.id,
    'input_sentence': llm_history.input_sentence,
    'output_label': llm_history.output_label,
    'model_url': llm_history.model_url,
    'type_of_activity': llm_history.type_of_activity,
    'user_id': llm_history.user_id,
} for llm_history in llm_histories])

tab2.write("LLM history:")
tab2.data_editor(llm_histories_df)