app.py

import gradio as gr
import pandas as pd
import geopandas as gpd
import plotly.express as px
import numpy as np

# 1. Load Data
# ---------------------------------------------------------
try:
    df = pd.read_parquet('Russia_regions_data.parquet')
    gdf = gpd.read_parquet('Russia_regions_geo_simplified.geoparquet')
except Exception as e:
    print(f"Error loading files: {e}")
    # Fallback dummy data
    df = pd.DataFrame(columns=['section', 'subsection', 'indicator_name', 'comment', 'year', 
                               'indicator_value', 'rel_indicator_value', 'indicator_unit', 'positive',
                               'indicator_code', 'Region', 'object_oktmo', 'ISO'])
    gdf = gpd.GeoDataFrame(columns=['ISO', 'geometry'])

# Ensure geometries are in standard lat/lon
gdf = gdf.to_crs(epsg=4326)

# 2. Prepare the Table Data (The "Menu")
# ---------------------------------------------------------
# Filter for unique indicator_code
df_unique = df.drop_duplicates(subset=['indicator_code']).copy()

# Define columns (added indicator_value and rel_indicator_value, hidden positive)
display_columns = [
    'section', 'subsection', 'indicator_name', 'comment', 
    'year', 'indicator_value', 'rel_indicator_value', 'indicator_unit'
]
hidden_link_key = 'indicator_code'

# Create the display dataframe
df_display = df_unique[display_columns].reset_index(drop=True)

# Calculate optimal column widths based on content with min/max constraints
def calculate_column_widths(df, columns):
    """Calculate column widths based on content length with constraints"""
    widths = []
    for col in columns:
        # Get max of header length and content lengths
        header_len = len(str(col))
        if len(df) > 0:
            content_lengths = df[col].astype(str).str.len()
            max_content_len = content_lengths.max()
            avg_content_len = content_lengths.mean()
            
            # Use average of max and mean, bounded by min/max
            optimal_len = (max_content_len + avg_content_len) / 2
            final_len = max(header_len, min(optimal_len, 50))  # Cap at 50 chars
        else:
            final_len = header_len
        
        # Scale to pixel width (roughly 8 pixels per character)
        # Minimum width of 100px, maximum of 400px
        width = max(100, min(int(final_len * 8), 400))
        widths.append(width)
    
    return widths

column_widths = calculate_column_widths(df_display, display_columns)

# Create styling for smaller font
def create_styling(df):
    """Create styling array for smaller font"""
    num_rows = len(df)
    num_cols = len(df.columns)
    # Apply small font style to all cells
    styling = [["font-size: 10px; white-space: normal; word-wrap: break-word;" for _ in range(num_cols)] for _ in range(num_rows)]
    return styling

table_styling = create_styling(df_display)

# 3. Utility Functions
# ---------------------------------------------------------
def remove_outliers(series):
    """Remove outliers using IQR method, replacing them with NaN"""
    Q1 = series.quantile(0.01)
    Q3 = series.quantile(0.99)
    IQR = Q3 - Q1
    lower_bound = Q1 - 2 * IQR
    upper_bound = Q3 + 2 * IQR
    return series.where((series >= lower_bound) & (series <= upper_bound), np.nan)

def should_use_log_scale(values):
    """
    Heuristic to decide if logarithmic scale should be used.
    Returns True if:
    1. All values are positive
    2. The ratio of max to min is greater than 100
    3. The distribution is highly skewed (skewness > 2)
    """
    # Remove NaN values for analysis
    clean_values = values.dropna()
    
    if len(clean_values) == 0:
        return False
    
    # Check if all values are positive
    if (clean_values <= 0).any():
        return False
    
    # Check ratio of max to min
    max_val = clean_values.max()
    min_val = clean_values.min()
    
    if min_val == 0:
        return False
    
    ratio = max_val / min_val
    
    # Check skewness
    skewness = clean_values.skew()
    
    # Use log scale if ratio is high and distribution is skewed
    return ratio > 100 and skewness > 2

def format_value(value):
    """Format value for tooltip: integers without decimals, floats with max 3 decimals"""
    if pd.isna(value):
        return "N/A"
    if isinstance(value, (int, np.integer)) or value == int(value):
        return f"{int(value)}"
    else:
        return f"{value:.3f}".rstrip('0').rstrip('.')

# 4. Define App Logic
# ---------------------------------------------------------
def create_ranking_map():
    """Create the default map showing overall rankings"""
    # Filter for the overall ranking indicator
    df_ranking = df[df['indicator_code'] == 'OVERALL_RANKING'].copy()
    
    if df_ranking.empty:
        return px.choropleth(title="Overall Ranking not found")
    
    # Ensure one row per region
    df_ranking = df_ranking.drop_duplicates(subset=['ISO'], keep='first')
    
    # Use rel_indicator_value (which is the same as indicator_value for ranking)
    df_ranking['color_value'] = df_ranking['rel_indicator_value']
    
    # Merge with Geometry
    merged = gdf.merge(df_ranking, on='ISO', how='inner')
    
    if merged.empty:
        return px.choropleth(title="No ranking data available")

    # Color scale: blue (lower/better) to red (higher/worse)
    fig = px.choropleth_map(
        merged,
        geojson=merged.geometry,
        locations=merged.index, 
        color='color_value',
        color_continuous_scale="RdYlBu_r",  # Blue (low/good) to Red (high/bad)
        map_style="satellite-streets",
        zoom=2,
        center={"lat": 60, "lon": 90},
        opacity=0.6,
        labels={'color_value': 'Overall Ranking'}
    )

    # Format values for tooltip
    merged['formatted_value'] = merged['rel_indicator_value'].apply(format_value)

    # Tooltip Configuration
    fig.update_traces(
        customdata=merged[['Region', 'formatted_value']],
        hovertemplate=(
            "<b>Region:</b> %{customdata[0]}<br>"
            "<b>Overall Ranking:</b> %{customdata[1]}"
            "<extra></extra>"
        )
    )
    
    fig.update_layout(
        margin={"r":0,"t":0,"l":0,"b":0},
        height=800
    )
    
    return fig

def update_map(select_data: gr.SelectData, current_table_data):
    """
    Triggered when a cell in the table is clicked.
    select_data.index is a tuple (row, col) or int depending on version.
    """
    if select_data is None:
        return create_ranking_map()

    # Handle index format (it often comes as [row, col] or just row index)
    # We safely extract the row index
    if isinstance(select_data.index, (list, tuple)):
        row_index = select_data.index[0]
    else:
        row_index = select_data.index
    
    # Get the row data from the visible table
    # In newer Gradio, current_table_data is a DataFrame
    if isinstance(current_table_data, pd.DataFrame):
        selected_row = current_table_data.iloc[row_index]
    else:
        return create_ranking_map()
    
    # Find the corresponding unique indicator code
    match = df_unique[
        (df_unique['indicator_name'] == selected_row['indicator_name']) &
        (df_unique['section'] == selected_row['section'])
    ]
    
    if match.empty:
        return px.choropleth(title="Indicator not found")
        
    selected_code = match.iloc[0][hidden_link_key]
    selected_unit = match.iloc[0]['indicator_unit']
    selected_positive = match.iloc[0]['positive']
    
    # Filter main data for this indicator
    df_filtered = df[df['indicator_code'] == selected_code].copy()
    
    # Ensure one row per region (remove duplicates, keep first occurrence)
    df_filtered = df_filtered.drop_duplicates(subset=['ISO'], keep='first')
    
    # Remove outliers (replace with NaN) - use rel_indicator_value
    df_filtered['value_clean'] = remove_outliers(df_filtered['rel_indicator_value'])
    
    # Calculate Rankings based on positive column
    df_filtered_for_ranking = df_filtered.dropna(subset=['value_clean']).copy()
    
    # If P: higher is better (ascending=False), if N or other: lower is better (ascending=True)
    ascending = (selected_positive != 'P')
    df_filtered_for_ranking = df_filtered_for_ranking.sort_values('value_clean', ascending=ascending).reset_index(drop=True)
    df_filtered_for_ranking['Ranking'] = range(1, len(df_filtered_for_ranking) + 1)
    
    # Merge rankings back
    df_filtered = df_filtered.merge(
        df_filtered_for_ranking[['ISO', 'Ranking']], 
        on='ISO', 
        how='left'
    )
    
    # Decide if we should use log scale
    use_log = should_use_log_scale(df_filtered['value_clean'])
    
    # Create color scale values (log if needed)
    if use_log:
        df_filtered['color_value'] = np.log10(df_filtered['value_clean'])
        color_label = f"{selected_unit} (log scale)"
    else:
        df_filtered['color_value'] = df_filtered['value_clean']
        color_label = selected_unit

    # Merge with Geometry
    merged = gdf.merge(df_filtered, on='ISO', how='inner')
    
    if merged.empty:
        return px.choropleth(title="No data for this indicator")

    # Determine color scale based on positive column
    # If P: red (low/bad) to blue (high/good)
    # If N or other: blue (low/good) to red (high/bad)
    if selected_positive == 'P':
        color_scale = "RdYlBu"  # Blue (low/bad) to Red (high/good)
    else:
        color_scale = "RdYlBu_r"  # Red (low/good) to Blue (high/bad) reversed

    # Construct Map with appropriate color scale
    fig = px.choropleth_map(
        merged,
        geojson=merged.geometry,
        locations=merged.index, 
        color='color_value',
        color_continuous_scale=color_scale,  # Red (low/good) to Blue (high/bad)
        map_style="satellite-streets",
        zoom=2,
        center={"lat": 60, "lon": 90},
        opacity=0.6,
        labels={'color_value': 'Overall Ranking'}
    )

    # Format values for tooltip
    merged['formatted_value'] = merged['rel_indicator_value'].apply(format_value)
    merged['formatted_ranking'] = merged['Ranking'].apply(lambda x: str(int(x)) if pd.notna(x) else "N/A")

    # Tooltip Configuration
    fig.update_traces(
        customdata=merged[['formatted_ranking', 'Region', 'indicator_name', 'formatted_value']],
        hovertemplate=(
            "<b>Rank:</b> %{customdata[0]}<br>"
            "<b>Region:</b> %{customdata[1]}<br>"
            "<b>Indicator Name:</b> %{customdata[2]}<br>"
            "<b>Relative Value:</b> %{customdata[3]}"
            "<extra></extra>"
        )
    )
    
    fig.update_layout(
        margin={"r":0,"t":0,"l":0,"b":0},
        height=800
    )
    
    return fig

# 5. Build Gradio UI
# ---------------------------------------------------------
with gr.Blocks(title="Russian Regions Analytics") as demo:
    gr.Markdown("## Russian Regional Indicators")
    
    with gr.Row():
        # Initialize with ranking map
        map_plot = gr.Plot(label="Regional Distribution", value=create_ranking_map())
    
    with gr.Row():
        # Prepare table value with styling metadata
        table_value = {
            "data": df_display.values.tolist(),
            "headers": display_columns,
            "metadata": {
                "styling": table_styling
            }
        }
        
        table = gr.DataFrame(
            value=table_value, 
            label="Select an Indicator",
            datatype=["str", "str", "str", "str", "number", "number", "number", "str"],
            interactive=True,
            max_height=700,
            column_widths=column_widths
        )

    # Wire the selection event
    table.select(
        fn=update_map,
        inputs=[table],
        outputs=[map_plot]
    )
    
    # Footer with data sources
    gr.Markdown("""
    ---
    ### Data Sources & Attribution
    
    **Statistical Data:** [Tochno.st Regional Datasets](https://tochno.st/datasets/regions_collection)  
    **Administrative Boundaries:** [geoBoundaries](https://www.geoboundaries.org/countryDownloads.html)  
    **Regional Codes:** [Codes of Subjects of the Russian Federation](https://en.everybodywiki.com/Codes_of_subjects_of_the_Russian_Federation)  
    **Translation Model:** [Tencent HY-MT1.5-7B](https://huggingface.co/tencent/HY-MT1.5-7B)
    """)

if __name__ == "__main__":
    demo.launch()