app_matching_page.py

import streamlit as st
import pandas as pd
import io
import xlsxwriter
from scipy.sparse import load_npz
import pickle
from sentence_transformers import SentenceTransformer
from modules.multimatch_result_table import show_multi_table
from modules.singlematch_result_table import show_single_table
from modules.allprojects_result_table import show_all_projects_table
from functions.filter_multi_project_matching import filter_multi
from functions.filter_single_project_matching import filter_single
from functions.filter_all_project_matching import filter_all_projects
from functions.multi_project_matching import calc_multi_matches
from functions.same_country_filter import same_country_filter
from functions.single_project_matching import find_similar
import gc

# Catch DATA
# Load Similarity matrix
@st.cache_data
def load_sim_matrix():
    """
    !!! Similarities when matches between same orgas are allowed
    """
    loaded_matrix = load_npz("src/extended_similarities.npz")
    return loaded_matrix

# Load Non Similar Orga Matrix
def load_nonsameorga_sim_matrix():
    """
    !!! Similarities when matches between same orgas are NOT allowed
    """
    loaded_matrix = load_npz("src/extended_similarities_nonsimorga.npz")
    return loaded_matrix

# Load Projects DFs
@st.cache_data
def load_projects():
    def fix_faulty_descriptions(description): # In some BMZ projects there are duplicate descriptions 
        if description and ';' in description:
            parts = description.split(';')
            if len(parts) == 2 and parts[0].strip() == parts[1].strip():
                return parts[0].strip()
        return description

    orgas_df = pd.read_csv("src/projects/project_orgas.csv")
    region_df = pd.read_csv("src/projects/project_region.csv")
    sector_df = pd.read_csv("src/projects/project_sector.csv")
    status_df = pd.read_csv("src/projects/project_status.csv")
    texts_df = pd.read_csv("src/projects/project_texts.csv")

    projects_df = pd.merge(orgas_df, region_df, on='iati_id', how='inner')
    projects_df = pd.merge(projects_df, sector_df, on='iati_id', how='inner')
    projects_df = pd.merge(projects_df, status_df, on='iati_id', how='inner')
    projects_df = pd.merge(projects_df, texts_df, on='iati_id', how='inner')



    # Add regions (should have been done in the preprocessing instead of here, so is just a quick fix to be able to add the region filter)
    region_lookup_df = pd.read_csv('src/codelists/regions.csv', usecols=['alpha-2', 'region', 'sub-region'])

    projects_df['country_code'] = projects_df['country'].str.replace(';', '').str.strip()
    # Replace empty values in the 'country_code' column with 'Unknown'
    projects_df['country_code'] = projects_df['country_code'].fillna('Unknown')

    region_lookup_df['alpha-2'] = region_lookup_df['alpha-2'].str.strip()
    projects_df = pd.merge(projects_df, region_lookup_df[['alpha-2', 'region', 'sub-region']], left_on='country_code', right_on='alpha-2', how='left')
    
    projects_df.rename(columns={'region': 'continent', 'sub-region': 'region'}, inplace=True)
    projects_df['continent'] = projects_df['continent'].fillna('Unknown')
    projects_df['region'] = projects_df['region'].fillna('Unknown')


    # Fix faulty descriptions for BMZ projects
    bmz_mask = projects_df['orga_abbreviation'].str.lower() == 'bmz'
    projects_df.loc[bmz_mask, 'description_main'] = projects_df.loc[bmz_mask, 'description_main'].apply(fix_faulty_descriptions)

    # Add Project Link column
    projects_df['Project Link'] = projects_df['iati_id'].apply(
        lambda x: f'https://d-portal.org/ctrack.html#view=act&aid={x}'
    )

    # Create necessary columns for consistency
    projects_df['crs_3_code_list'] = projects_df['crs_3_name'].apply(
        lambda x: [""] if pd.isna(x) else (str(x).split(";")[:-1] if str(x).endswith(";") else str(x).split(";"))
    )
    projects_df['crs_5_code_list'] = projects_df['crs_5_name'].apply(
        lambda x: [""] if pd.isna(x) else (str(x).split(";")[:-1] if str(x).endswith(";") else str(x).split(";"))
    )
    projects_df['sdg_list'] = projects_df['sgd_pred_code'].apply(
        lambda x: [""] if pd.isna(x) else (str(x).split(";")[:-1] if str(x).endswith(";") else str(x).split(";"))
    )
    
    # Ensure country_flag is set to None if country_name is missing or "NA"
    projects_df['country_flag'] = projects_df.apply(
        lambda row: None if pd.isna(row['country_name']) or row['country_name'] == "NA" else row['country_flag'],
        axis=1
    )

    iati_search_list = [f'{row.iati_id}' for row in projects_df.itertuples()]
    title_search_list = [f'{row.title_main} ({row.orga_abbreviation.upper()})' for row in projects_df.itertuples()]

    return projects_df, iati_search_list, title_search_list


# Load CRS 3 data
@st.cache_data
def getCRS3():
    # Read in CRS3 CODELISTS
    crs3_df = pd.read_csv('src/codelists/crs3_codes.csv')
    CRS3_CODES = crs3_df['code'].tolist()
    CRS3_NAME = crs3_df['name'].tolist()
    CRS3_MERGED = {f"{name} - {code}": code for name, code in zip(CRS3_NAME, CRS3_CODES)}
    return CRS3_MERGED

# Load CRS 5 data
@st.cache_data
def getCRS5():
    # Read in CRS3 CODELISTS
    crs5_df = pd.read_csv('src/codelists/crs5_codes.csv')
    CRS5_CODES = crs5_df['code'].tolist()
    CRS5_NAME = crs5_df['name'].tolist()
    CRS5_MERGED = {code: [f"{name} - {code}"] for name, code in zip(CRS5_NAME, CRS5_CODES)}
    return CRS5_MERGED

# Load SDG data
@st.cache_data
def getSDG():
    # Read in SDG CODELISTS
    sdg_df = pd.read_csv('src/codelists/sdg_goals.csv')
    SDG_NAMES = sdg_df['name'].tolist()
    return SDG_NAMES

@st.cache_data
def getCountry():
    # Read in countries from codelist
    country_df = pd.read_csv('src/codelists/country_codes_ISO3166-1alpha-2.csv')
    
    # Read in regions from codelist, keeping only the relevant columns
    region_lookup_df = pd.read_csv('src/codelists/regions.csv', usecols=['alpha-2', 'region', 'sub-region'])
    
    # Strip quotes from the 'Alpha-2 code' column in country_df
    country_df['Alpha-2 code'] = country_df['Alpha-2 code'].str.replace('"', '').str.strip()
    
    # Ensure no leading/trailing spaces in the 'alpha-2' column in region_lookup_df
    region_lookup_df['alpha-2'] = region_lookup_df['alpha-2'].str.strip()
    
    # Merge country and region dataframes on 'Alpha-2 code' from country_df and 'alpha-2' from region_lookup_df
    merged_df = pd.merge(country_df, region_lookup_df, how='left', left_on='Alpha-2 code', right_on='alpha-2')
    
    # Handle any missing regions or sub-regions
    merged_df['region'] = merged_df['region'].fillna('Unknown')
    merged_df['sub-region'] = merged_df['sub-region'].fillna('Unknown')
    
    # Extract necessary columns as lists
    COUNTRY_CODES = merged_df['Alpha-2 code'].tolist()
    COUNTRY_NAMES = merged_df['Country'].tolist()
    REGIONS = merged_df['region'].tolist()
    SUB_REGIONS = merged_df['sub-region'].tolist()
    
    # Create the original COUNTRY_OPTION_LIST without regions
    COUNTRY_OPTION_LIST = [f"{COUNTRY_NAMES[i]} ({COUNTRY_CODES[i]})" for i in range(len(COUNTRY_NAMES))]

    # Create a hierarchical filter structure for sub-regions
    sub_region_hierarchy = {}
    sub_region_to_region = {}
    for i in range(len(SUB_REGIONS)):
        sub_region = SUB_REGIONS[i]
        country = COUNTRY_CODES[i]
        region = REGIONS[i]
        if sub_region not in sub_region_hierarchy:
            sub_region_hierarchy[sub_region] = []
        sub_region_hierarchy[sub_region].append(country)
        
        # Map sub-regions to regions
        sub_region_to_region[sub_region] = region

    # Sort the subregions by regions
    sorted_sub_regions = sorted(sub_region_hierarchy.keys(), key=lambda x: sub_region_to_region[x])
    
    return COUNTRY_OPTION_LIST, sorted_sub_regions

# Call the function to load and display the country data
COUNTRY_OPTION_LIST, REGION_OPTION_LIST = getCountry()


# Load Sentence Transformer Model
@st.cache_resource
def load_model():
    model = SentenceTransformer('all-MiniLM-L6-v2')
    return model

# Load Embeddings
@st.cache_data
def load_embeddings_and_index():
    # Load embeddings
    with open("src/embeddings.pkl", "rb") as fIn:
        stored_data = pickle.load(fIn)
    embeddings = stored_data["embeddings"]
    return embeddings

# USE CACHE FUNCTIONS
sim_matrix = load_sim_matrix() # For similarities when matches between same orgas are allowed
nonsameorgas_sim_matrix = load_nonsameorga_sim_matrix()  #For similarities when matches between same orgas are NOT allowed
projects_df, iati_search_list, title_search_list = load_projects()

CRS3_MERGED = getCRS3()
CRS5_MERGED = getCRS5()
SDG_NAMES = getSDG()

# LOAD MODEL FROM CACHE FOR SEMANTIC SEARCH
model = load_model()
embeddings = load_embeddings_and_index()



##################################

def show_landing_page():
    st.title("Development Project Synergy Finder")

    st.subheader("About")
    st.markdown("""
    Multiple international organizations have projects in the same field and region. These projects could collaborate or learn from each other to increase their impact if they were aware of one another. The Project Synergy Finder facilitates the search for similar projects across different development organizations and banks in three distinct ways. Note that this app is a prototype, results may be incomplete or inaccurate.   """)
    st.markdown("<br><br>", unsafe_allow_html=True)  # Add two line breaks

    st.subheader("Pages")
    st.markdown("""
    1. **📊 All Projects**: Displays all projects included in the analysis.
        *Example Use Case*: Show all World Bank and African Development Bank projects in East Africa working towards the Sustainable Development Goal of achieving gender equality.
                

    2. **🎯 Single-Project Matching**: Finds the top similar projects to a selected one.
        *Example Use Case*: Show projects in Eastern Europe that are similar to the "Second Irrigation and Drainage Improvement Project" by the World Bank.
                

    3. **🔍 Multi-Project Matching**: Searches for matching pairs of projects.
        *Example Use Case*: Show pairs of similar projects in the "Energy Policy" sector from different organizations within the same country.
    """)
    st.markdown("<br><br>", unsafe_allow_html=True)  # Add two line breaks

    st.subheader("Data")
    st.markdown("""
    **IATI Data**: The data is sourced from the [IATI d-portal](https://d-portal.org/), providing project-level information. The International Aid Transparency Initiative (IATI) aims to enhance transparency and effectiveness in development cooperation by making data publicly accessible.
    
    **Data Update**: The data is updated irregularly, with the last retrieval on 10th May 2024.
    
    **Project Data**: Data from projects labeled as active during the last data retrieval are included. The data includes Project Title, Description, URL, Country, and Sector classification (CRS). The CRS5 and CRS3 classifications organize development cooperation into categories, with the 5-digit level providing more specific details within the broader 3-digit categories.
    
    **Organizations**: The tool currently includes projects from the following organizations:
    - **IAD**: Inter-American Development Bank
    - **ADB**: Asian Development Bank
    - **AfDB**: African Development Bank
    - **EIB**: European Investment Bank
    - **WB**: World Bank
    - **WBTF**: World Bank Trust Fund
    - **BMZ**: Federal Ministry for Economic Cooperation and Development (Germany)
    - **KfW**: KfW Development Bank (Germany)
    - **GIZ**: Deutsche Gesellschaft für Internationale Zusammenarbeit (Germany)
    - **AA**: German Federal Foreign Office (Germany)
    
    **Additional Data**: The Sustainable Development Goals (SDGs) are 17 UN goals aimed at achieving global sustainability, peace, and prosperity by 2030. The SDG categorization in this tool is AI-predicted based on project descriptions and titles using a [SDG Classifier](https://huggingface.co/jonas/bert-base-uncased-finetuned-sdg) trainded on the OSDG dataset.
    """)


##################################


def show_all_projects_page():
    # Define the page size at the beginning
    page_size = 30

    def reset_pagination():
        st.session_state.current_end_idx_all = page_size


    col1, col2, col3 = st.columns([10, 1, 10])
    with col1:
        st.subheader("Project Filter")

    st.session_state.crs5_option_disabled = True
    col1, col2, col3 = st.columns([10, 1, 10])
    with col1:
        # CRS 3 SELECTION
        crs3_option = st.multiselect(
            'CRS 3',
            CRS3_MERGED,
            placeholder="Select a CRS 3 code",
            on_change=reset_pagination,
            key='crs3_all_projects_page'
        )

        # CRS 5 SELECTION
        # Only enable crs5 select field when crs3 code is selected
        if crs3_option:
            st.session_state.crs5_option_disabled = False

        # Define list of crs5 codes depending on crs3 codes
        crs5_list = [txt[0].replace('"', "") for crs3_item in crs3_option for code, txt in CRS5_MERGED.items() if str(code)[:3] == str(crs3_item)[-3:]]

        # crs5 select field
        crs5_option = st.multiselect(
            'CRS 5',
            crs5_list,
            placeholder="Select a CRS 5 code",
            disabled=st.session_state.crs5_option_disabled,
            on_change=reset_pagination,
            key='crs5_all_projects_page'
        )

        # SDG SELECTION
        sdg_option = st.selectbox(
            label='Sustainable Development Goal (AI-predicted)',
            index=None,
            placeholder="Select a SDG",
            options=SDG_NAMES[:-1],
            on_change=reset_pagination,
            key='sdg_all_projects_page'
        )

    with col3:
        # REGION SELECTION
        region_option = st.multiselect(
            'Regions',
            REGION_OPTION_LIST,
            placeholder="All regions selected",
            on_change=reset_pagination,
            key='regions_all_projects_page'
        )

        # COUNTRY SELECTION
        country_option = st.multiselect(
            'Countries',
            COUNTRY_OPTION_LIST,
            placeholder="All countries selected",
            on_change=reset_pagination,
            key='country_all_projects_page'
        )

        # ORGA SELECTION
        orga_abbreviation = projects_df["orga_abbreviation"].unique()
        orga_full_names = projects_df["orga_full_name"].unique()
        orga_list = [f"{orga_full_names[i]} ({orga_abbreviation[i].upper()})" for i in range(len(orga_abbreviation))]

        orga_option = st.multiselect(
            'Organizations',
            orga_list,
            placeholder="All organizations selected",
            on_change=reset_pagination,
            key='orga_all_projects_page'
        )

    # CRS CODE LIST
    crs3_list = [i[-3:] for i in crs3_option]
    crs5_list = [i[-5:] for i in crs5_option]

    # SDG CODE LIST
    if sdg_option is not None:
        sdg_str = sdg_option.split(".")[0]
    else:
        sdg_str = ""

    # COUNTRY CODES LIST
    country_code_list = [option[-3:-1] for option in country_option]

    # ORGANIZATION CODES LIST
    orga_code_list = [option.split("(")[1][:-1].lower() for option in orga_option]

    st.write("-----")

    # FILTER DF WITH SELECTED FILTER OPTIONS
    filtered_df = filter_all_projects(projects_df, country_code_list, orga_code_list, crs3_list, crs5_list, sdg_str, region_option)
    if isinstance(filtered_df, pd.DataFrame) and len(filtered_df) != 0:
        # Implement pagination
        if 'current_end_idx_all' not in st.session_state:
            st.session_state.current_end_idx_all = page_size

        end_idx = st.session_state.current_end_idx_all

        paginated_df = filtered_df.iloc[:end_idx]

        col1, col2 = st.columns([7, 3])
        with col1:
            st.subheader("Filtered Projects")
        with col2:
            # Add a download button for the paginated results
            def to_excel(df, sheet_name):
                # Rename columns
                df = df.rename(columns={
                    "iati_id": "IATI Identifier",
                    "title_main": "Title",
                    "orga_abbreviation": "Organization",
                    "description_main": "Description",
                    "country_name": "Country",
                    "sdg_list": "SDG List",
                    "crs_3_code_list": "CRS 3 Codes",
                    "crs_5_code_list": "CRS 5 Codes",
                    "Project Link": "Project Link"
                })
                output = io.BytesIO()
                writer = pd.ExcelWriter(output, engine='xlsxwriter')
                df.to_excel(writer, index=False, sheet_name=sheet_name)
                writer.close()
                processed_data = output.getvalue()
                return processed_data

            # Direct download buttons
            columns_to_include = ["iati_id", "title_main", "orga_abbreviation", "description_main", "country_name", "sdg_list", "crs_3_code_list", "crs_5_code_list", "Project Link"]

            with st.expander("Excel Download"):
                # First 15 Results Button
                df_to_download_15 = filtered_df[columns_to_include].head(15)
                excel_data_15 = to_excel(df_to_download_15, "Sheet1")
                st.download_button(label="First 30 Projects", data=excel_data_15, file_name="First_15_All_Projects_Filtered.xlsx", mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet")

                # All Results Button
                df_to_download_all = filtered_df[columns_to_include]
                excel_data_all = to_excel(df_to_download_all, "Sheet1")
                st.download_button(label="All", data=excel_data_all, file_name="All_All_Projects_Filtered.xlsx", mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet")
                    
        show_all_projects_table(projects_df, paginated_df)

        st.write(f"Showing 1 to {min(end_idx, len(filtered_df))} of {len(filtered_df)} projects")

        # Center the buttons and place them close together
        col1, col2, col3, col4, col5 = st.columns([2, 1, 1, 1, 2])
        with col2:
            if st.button('Show More', key='show_more'):
                st.session_state.current_end_idx_all = min(end_idx + page_size, len(filtered_df))
                st.experimental_rerun()
        with col4:
            if st.button('Show Less', key='show_less') and end_idx > page_size:
                st.session_state.current_end_idx_all = max(end_idx - page_size, page_size)
                st.experimental_rerun()

    else:
        st.write("-----")
        col1, col2, col3 = st.columns([1, 1, 1])
        with col2:
            st.write("  ")
            st.markdown("<span style='color: red'>There are no results for the applied filter. Try another filter!</span>", unsafe_allow_html=True)

    del crs3_list, crs5_list, sdg_str, filtered_df
    gc.collect()



##################################

def show_single_matching_page():
    # Define the page size at the beginning
    page_size = 15

    def reset_pagination():
        st.session_state.current_end_idx_single = page_size

    with st.expander("Explanation"):
        st.caption("""
                    Single Project Matching enables you to choose an individual project using either the project IATI ID or title, to display projects most similar to it.
                
                    **Similarity Score**:
                    - Similarity ranges from 0 to 100 (identical projects score 100%), and is calculated based on           
                        - Text similarity of project description and title (MiniLMM & Cosine Similiarity).
                        - Matching of SDGs (AI-predicted).
                        - Matching of CRS-3 & CRS-5 sector codes.
                    - Components are weighted to give a normalized score. 

                    Note that this app is a prototype, results may be incomplete or inaccurate.
                    """)

    col1, col2, col3 = st.columns([10, 1, 10])
    with col1:
        st.subheader("Reference Project")
        st.caption("""
                    Select a reference project either by its title or IATI ID. 
                    """)
    with col3:
        st.subheader("Filters for Similar Projects")
        st.caption("""
                    The filters are applied to find the similar projects and are independend of the selected reference project.
                """)

    col1, col2, col3 = st.columns([10, 1, 10])
    with col1:
        search_option = st.selectbox(
            label='Search with project title or IATI ID',
            index=0,
            placeholder=" ",
            options=["Search with IATI ID", "Search with project title"],
            on_change=reset_pagination,
            key='search_option_single'
        )

        if search_option == "Search with IATI ID":
            search_list = iati_search_list
        else:
            search_list = title_search_list

        project_option = st.selectbox(
            label='Search for a project',
            index=None,
            placeholder=" ",
            options=search_list,
            on_change=reset_pagination,
            key='project_option_single'
        )

    with col3:
        orga_abbreviation = projects_df["orga_abbreviation"].unique()
        orga_full_names = projects_df["orga_full_name"].unique()
        orga_list = [f"{orga_full_names[i]} ({orga_abbreviation[i].upper()})" for i in range(len(orga_abbreviation))]

        # REGION SELECTION
        region_option_s = st.multiselect(
            'Regions',
            REGION_OPTION_LIST,
            placeholder="All regions selected",
            on_change=reset_pagination,
            key='regions_single_projects_page'
        )

        country_option_s = st.multiselect(
            'Countries ',
            COUNTRY_OPTION_LIST,
            placeholder="All countries selected ",
            on_change=reset_pagination,
            key='country_option_single'
        )
        orga_option_s = st.multiselect(
            'Organizations',
            orga_list,
            placeholder="All organizations selected ",
            on_change=reset_pagination,
            key='orga_option_single'
        )

        different_orga_checkbox_s = st.checkbox("Only matches between different organizations ", value=True, on_change=reset_pagination, key='different_orga_checkbox_single')

    st.write("-----")

    if project_option:
        selected_project_index = search_list.index(project_option)
        country_code_list = [option[-3:-1] for option in country_option_s]
        orga_code_list = [option.split("(")[1][:-1].lower() for option in orga_option_s]

        TOP_X_PROJECTS = 1000
        with st.spinner('Please wait...'):
            filtered_df_s = filter_single(projects_df, country_code_list, orga_code_list, region_option_s)

        if isinstance(filtered_df_s, pd.DataFrame) and len(filtered_df_s) != 0:
            if different_orga_checkbox_s:
                with st.spinner('Please wait...'):
                    top_projects_df = find_similar(selected_project_index, nonsameorgas_sim_matrix, filtered_df_s, TOP_X_PROJECTS)
            else:
                with st.spinner('Please wait...'):
                    top_projects_df = find_similar(selected_project_index, sim_matrix, filtered_df_s, TOP_X_PROJECTS)

            # Implement show more, show less, and show all functionality
            if 'current_end_idx_single' not in st.session_state:
                st.session_state.current_end_idx_single = page_size

            end_idx = st.session_state.current_end_idx_single

            paginated_df = top_projects_df.iloc[:end_idx]

            # Add a download button for the paginated results
            def to_excel(df, sheet_name):
                # Rename columns
                df = df.rename(columns={
                    "similarity": "Similarity Score",
                    "iati_id": "IATI Identifier",
                    "title_main": "Title",
                    "orga_abbreviation": "Organization",
                    "description_main": "Description",
                    "country_name": "Country",
                    "sdg_list": "SDG List",
                    "crs_3_code_list": "CRS 3 Codes",
                    "crs_5_code_list": "CRS 5 Codes",
                    "Project Link": "Project Link"
                })
                output = io.BytesIO()
                writer = pd.ExcelWriter(output, engine='xlsxwriter')
                df.to_excel(writer, index=False, sheet_name=sheet_name)
                writer.close()
                processed_data = output.getvalue()
                return processed_data

            # Direct download buttons
            columns_to_include = ["similarity", "iati_id", "title_main", "orga_abbreviation", "description_main", "country_name", "sdg_list", "crs_3_code_list", "crs_5_code_list", "Project Link"]

            col1, col2 = st.columns([15, 5])
            with col2:
                with st.expander("Excel Download"):
                    # First 15 Results Button
                    df_to_download_15 = top_projects_df[columns_to_include].head(15)
                    excel_data_15 = to_excel(df_to_download_15, "Sheet1")
                    st.download_button(label="Download first 15 projects", data=excel_data_15, file_name="First_15_Single_Project_Matching_Results.xlsx", mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet")
                    df_to_download_all = top_projects_df[columns_to_include]
                    excel_data_all = to_excel(df_to_download_all, "Sheet1")
                    st.download_button(label="Download All", data=excel_data_all, file_name="All_Single_Project_Matching_Results.xlsx", mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet")
                
            show_single_table(selected_project_index, projects_df, paginated_df)

            st.write(f"Showing 1 to {min(end_idx, len(top_projects_df))} of {len(top_projects_df)} projects")

            # Center the buttons and place them close together
            col1, col2, col3, col4, col5 = st.columns([2, 1, 1, 1, 2])
            with col2:
                if st.button('Show More'):
                    st.session_state.current_end_idx_single = min(end_idx + page_size, len(top_projects_df))
                    st.experimental_rerun()
            with col3:
                if st.button('Show Less') and end_idx > page_size:
                    st.session_state.current_end_idx_single = max(end_idx - page_size, page_size)
                    st.experimental_rerun()
            with col4:
                if st.button('Show All'):
                    st.session_state.current_end_idx_single = len(top_projects_df)
                    st.experimental_rerun()

        else:
            st.write("-----")
            col1, col2, col3 = st.columns([1, 1, 1])
            with col2:
                st.write("  ")
                st.markdown("<span style='color: red'>There are no results for this filter!</span>", unsafe_allow_html=True)
    gc.collect()


##################################
def show_multi_matching_page():
    # Define the page size at the beginning
    page_size = 30

    def reset_pagination():
        st.session_state.current_end_idx_multi = page_size

    with st.expander("Explanation"):
        st.caption("""
        Multi-Project Matching enables to find collaboration opportunities by identifying matching (=similar) projects.

        **How It Works**:
        - Filter projects by CRS sector, SDG, country, and organization.
        - Each match displays two similar projects side-by-side.

        **Similarity Score**:
        - Similarity ranges from 0 to 100 (Identical projects score 100%), and is calculated based on           
        - Text similarity of project description and title (MiniLMM & Cosine Similiarity).
        - Matching of SDGs (AI-predicted).
        - Matching of CRS-3 & CRS-5 sector codes.
        - Components are weighted to give a normalized score.

        Note that this app is a prototype, results may be incomplete or inaccurate.
        """)
    col1, col2, col3 = st.columns([10, 1, 10])
    with col1:
        st.subheader("Sector Filters")
        st.caption("""
            At least one sector filter must be applied to see results.
        """)
    with col3:
        st.subheader("Additional Filters")

    st.session_state.crs5_option_disabled = True
    col1, col2, col3 = st.columns([10, 1, 10])
    with col1:
        crs3_option = st.multiselect(
            'CRS 3',
            CRS3_MERGED,
            placeholder="Select a CRS 3 code",
            on_change=reset_pagination,
            key='crs3_multi_projects_page'
        )

        if crs3_option:
            st.session_state.crs5_option_disabled = False

        crs5_list = [txt[0].replace('"', "") for crs3_item in crs3_option for code, txt in CRS5_MERGED.items() if str(code)[:3] == str(crs3_item)[-3:]]

        crs5_option = st.multiselect(
            'CRS 5',
            crs5_list,
            placeholder="Select a CRS 5 code",
            disabled=st.session_state.crs5_option_disabled,
            on_change=reset_pagination,
            key='crs5_multi_projects_page'
        )

        sdg_option = st.selectbox(
            label='Sustainable Development Goal (AI-predicted)',
            index=None,
            placeholder="Select a SDG",
            options=SDG_NAMES[:-1],
            on_change=reset_pagination,
            key='sdg_multi_projects_page'
        )

        query = ""

    with col3:
        region_option = st.multiselect(
            'Regions',
            REGION_OPTION_LIST,
            placeholder="All regions selected",
            on_change=reset_pagination,
            key='regions_multi_projects_page'
        )
        country_option = st.multiselect(
            'Countries',
            COUNTRY_OPTION_LIST,
            placeholder="All countries selected",
            on_change=reset_pagination,
            key='country_multi_projects_page'
        )

        orga_abbreviation = projects_df["orga_abbreviation"].unique()
        orga_full_names = projects_df["orga_full_name"].unique()
        orga_list = [f"{orga_full_names[i]} ({orga_abbreviation[i].upper()})" for i in range(len(orga_abbreviation))]

        orga_option = st.multiselect(
            'Organizations',
            orga_list,
            placeholder="All organizations selected",
            on_change=reset_pagination,
            key='orga_multi_projects_page'
        )

        identical_country_checkbox = st.checkbox("Only matches where country is identical", value=True, on_change=reset_pagination, key='identical_country_checkbox_multi')
        different_orga_checkbox = st.checkbox("Only matches between different organizations", value=True, on_change=reset_pagination, key='different_orga_checkbox_multi')
        filtered_country_only_checkbox = st.checkbox("Only matches between filtered countries", value=True, on_change=reset_pagination, key='filtered_country_only_checkbox_multi')
        filtered_orga_only_checkbox = st.checkbox("Only matches between filtered organisations", value=True, on_change=reset_pagination, key='filtered_orga_only_checkbox_multi')


    # CRS CODE LIST
    crs3_list = [i[-3:] for i in crs3_option]
    crs5_list = [i[-5:] for i in crs5_option]

    # SDG CODE LIST
    sdg_str = sdg_option.split(".")[0] if sdg_option else ""

    # COUNTRY CODES LIST
    country_code_list = [option[-3:-1] for option in country_option]

    # ORGANIZATION CODES LIST
    orga_code_list = [option.split("(")[1][:-1].lower() for option in orga_option]

    # Handle case where no organizations are selected but the checkbox is checked
    if filtered_orga_only_checkbox and not orga_code_list:
        orga_code_list = projects_df["orga_abbreviation"].unique().tolist()

    # FILTER DF WITH SELECTED FILTER OPTIONS
    TOP_X_PROJECTS = 2000
    filtered_df = filter_multi(projects_df, crs3_list, crs5_list, sdg_str, country_code_list, orga_code_list, region_option, query, model, embeddings, TOP_X_PROJECTS)
    if isinstance(filtered_df, pd.DataFrame) and len(filtered_df) != 0:
        # FIND MATCHES
        # If only same country checkbox is activated
        if filtered_country_only_checkbox:
            with st.spinner('Please wait...'):
                compare_df = same_country_filter(projects_df, country_code_list)
        else:
            compare_df = projects_df

        if filtered_orga_only_checkbox:
            compare_df = compare_df[compare_df['orga_abbreviation'].isin(orga_code_list)]

        # if show only different orgas checkbox is activated
        with st.spinner('Please wait...'):
            p1_df, p2_df = calc_multi_matches(filtered_df, compare_df, nonsameorgas_sim_matrix if different_orga_checkbox else sim_matrix, TOP_X_PROJECTS, identical_country=identical_country_checkbox)

        # Sort by similarity before pagination
        p1_df = p1_df.sort_values(by='similarity', ascending=False)
        p2_df = p2_df.sort_values(by='similarity', ascending=False)

        # Implement pagination
        if 'current_end_idx_multi' not in st.session_state:
            st.session_state.current_end_idx_multi = page_size

        end_idx = st.session_state.current_end_idx_multi

        paginated_p1_df = p1_df.iloc[:end_idx]
        paginated_p2_df = p2_df.iloc[:end_idx]

        if not paginated_p1_df.empty and not paginated_p2_df.empty:
            col1, col2 = st.columns([10, 2])
            with col1:
                st.subheader("Matched Projects")
            with col2:
                # Add a download button for the paginated results
                def to_excel(p1_df, p2_df, sheet_name):
                    # Rename columns
                    p1_df = p1_df.rename(columns={
                        "similarity": "Similarity Score",
                        "iati_id": "IATI Identifier",
                        "title_main": "Title",
                        "orga_abbreviation": "Organization",
                        "description_main": "Description",
                        "country_name": "Country",
                        "sdg_list": "SDG List",
                        "crs_3_code_list": "CRS 3 Codes",
                        "crs_5_code_list": "CRS 5 Codes",
                        "Project Link": "Project Link"
                    })
                    p2_df = p2_df.rename(columns={
                        "similarity": "Similarity Score",
                        "iati_id": "IATI Identifier",
                        "title_main": "Title",
                        "orga_abbreviation": "Organization",
                        "description_main": "Description",
                        "country_name": "Country",
                        "sdg_list": "SDG List",
                        "crs_3_code_list": "CRS 3 Codes",
                        "crs_5_code_list": "CRS 5 Codes",
                        "Project Link": "Project Link"
                    })
                    
                    combined_df = pd.concat([p1_df, pd.DataFrame([{}]), p2_df], ignore_index=True)
                    combined_df.fillna('', inplace=True)
                    
                    empty_row = pd.DataFrame([{}])
                    combined_df_list = []
                    
                    for idx in range(0, len(p1_df), 2):
                        combined_df_list.append(p1_df.iloc[[idx]])
                        combined_df_list.append(p2_df.iloc[[idx]])
                        combined_df_list.append(empty_row)
                    
                    combined_df = pd.concat(combined_df_list, ignore_index=True)
                    
                    output = io.BytesIO()
                    writer = pd.ExcelWriter(output, engine='xlsxwriter')
                    combined_df.to_excel(writer, index=False, sheet_name=sheet_name)
                    writer.close()
                    processed_data = output.getvalue()
                    return processed_data

                # Direct download buttons
                columns_to_include = ["similarity", "iati_id", "title_main", "orga_abbreviation", "description_main", "country_name", "sdg_list", "crs_3_code_list", "crs_5_code_list", "Project Link"]

                with st.expander("Excel Download"):
                    # First 15 Results Button
                    p1_df_to_download_15 = p1_df[columns_to_include].head(30)
                    p2_df_to_download_15 = p2_df[columns_to_include].head(30)
                    excel_data_15 = to_excel(p1_df_to_download_15, p2_df_to_download_15, "Sheet1")
                    st.download_button(label="First 15 Matches", data=excel_data_15, file_name="First_15_Multi_Projects_Matching_Results.xlsx", mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet")

                    # All Results Button
                    p1_df_to_download_all = p1_df[columns_to_include]
                    p2_df_to_download_all = p2_df[columns_to_include]
                    excel_data_all = to_excel(p1_df_to_download_all, p2_df_to_download_all, "Sheet1")
                    st.download_button(label="All", data=excel_data_all, file_name="All_Multi_Projects_Matching_Results.xlsx", mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet")

            show_multi_table(paginated_p1_df, paginated_p2_df)

            st.write(f"Showing 1 to {min(end_idx // 2, len(p1_df) // 2)} of {len(p1_df) // 2} matches")

            # Center the buttons and place them close together
            col1, col2, col3, col4, col5 = st.columns([2, 1, 1, 1, 2])
            with col2:
                if st.button('Show More', key='show_more_button'):
                    st.session_state.current_end_idx_multi = min(end_idx + page_size, len(p1_df))
                    st.experimental_rerun()
            with col3:
                if st.button('Show Less', key='show_less_button') and end_idx > page_size:
                    st.session_state.current_end_idx_multi = max(end_idx - page_size, page_size)
                    st.experimental_rerun()
            with col4:
                if st.button('Show All', key='show_all_button'):
                    st.session_state.current_end_idx_multi = len(p1_df)
                    st.experimental_rerun()

            del p1_df, p2_df
        else:
            st.write("-----")
            col1, col2, col3 = st.columns([1, 1, 1])
            with col2:
                st.write("  ")
                st.markdown("<span style='color: red'>There are no results for the applied filter. Try another filter!</span>", unsafe_allow_html=True)

    else:
        st.write("-----")
        col1, col2, col3 = st.columns([1, 1, 1])
        with col2:
            st.write("  ")
            st.markdown("<span style='color: red'>There are no results for the applied filter. Try another filter!</span>", unsafe_allow_html=True)

    del crs3_list, crs5_list, sdg_str, filtered_df
    gc.collect()