Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
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 | |
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 | |
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 | |
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 | |
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 | |
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 | |
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 | |
def load_model(): | |
model = SentenceTransformer('all-MiniLM-L6-v2') | |
return model | |
# Load Embeddings | |
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. """) | |
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 detailed project-level information. IATI (International Aid Transparency Initiative) is a global initiative to improve the transparency of development and humanitarian resources and their results to address poverty and crises. | |
**Data Update**: The data is updated irregularly, with the last retrieval on 10th May 2024. | |
**Project Data**: Includes Project Title, Description, URL, Country, and Sector classification (CRS). The CRS5 and CRS3 classifications organize development aid 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', | |
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. | |
""") | |
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. | |
""") | |
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', | |
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() |