app.py

import gradio as gr
import pandas as pd
import numpy as np
import plotly.express as px
from sklearn.ensemble import RandomForestClassifier
from sklearn.preprocessing import OneHotEncoder
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from pathlib import Path
import tempfile

DATA_PATH = Path("synthetic_delivery_data.csv")

NUMERIC_COLS = [
    "distance_km", "package_weight_kg", "delivery_time_hours",
    "expected_time_hours", "delivery_rating", "delivery_cost"
]
CAT_COLS = [
    "delivery_partner", "package_type", "vehicle_type", "delivery_mode",
    "region", "weather_condition", "delayed", "delivery_status"
]

CUSTOM_CSS = """
.gradio-container {max-width: 1280px !important; margin: auto;}
.metric-card {background: linear-gradient(135deg, #ffffff, #f7f8fb); border: 1px solid #e8e8ef; border-radius: 18px; padding: 18px; box-shadow: 0 8px 24px rgba(0,0,0,.05);} 
.metric-label {font-size: 13px; color: #5f6470; margin-bottom: 6px;}
.metric-value {font-size: 30px; font-weight: 800; color: #111827;}
.insight-box {background: #111827; color: white; border-radius: 18px; padding: 20px; line-height: 1.55;}
.small-muted {color: #6b7280; font-size: 13px;}
"""


def _clean_time_column(series):
    """Convert either normal numbers or timestamp-looking duration strings into numeric hours."""
    if pd.api.types.is_numeric_dtype(series):
        return pd.to_numeric(series, errors="coerce")
    s = series.astype(str)
    # Handles values like 1970-01-01 00:00:00.000000008 by extracting last part.
    extracted = s.str.split(".").str[-1]
    return pd.to_numeric(extracted, errors="coerce")


def load_and_prepare(file_obj=None):
    if file_obj is None:
        df = pd.read_csv(DATA_PATH)
    else:
        df = pd.read_csv(file_obj.name)

    df = df.copy()
    df.columns = df.columns.str.strip().str.lower()
    df = df.drop_duplicates()

    required_minimum = ["distance_km", "vehicle_type", "weather_condition", "delivery_mode", "region"]
    missing_required = [c for c in required_minimum if c not in df.columns]
    if missing_required:
        raise gr.Error(f"Your file is missing these required columns: {missing_required}")

    for col in ["delivery_time_hours", "expected_time_hours"]:
        if col in df.columns:
            df[col] = _clean_time_column(df[col])

    for col in NUMERIC_COLS:
        if col in df.columns:
            df[col] = pd.to_numeric(df[col], errors="coerce")
            df[col] = df[col].fillna(df[col].median())

    for col in CAT_COLS:
        if col in df.columns:
            df[col] = df[col].astype(str).str.strip().str.lower()
            if df[col].isna().any():
                df[col] = df[col].fillna(df[col].mode()[0])

    # If expected/delivery time are not reliable or missing, rebuild them with business logic.
    df = create_synthetic_time_logic(df)
    df["delay_hours"] = (df["delivery_time_hours"] - df["expected_time_hours"]).round(2)
    df["calculated_delay"] = np.where(df["delay_hours"] > 0, "yes", "no")
    df["delay_score"] = df["delay_hours"].apply(delay_score)
    df["performance_label"] = df["delay_score"].apply(performance_label)
    df["distance_category"] = pd.cut(
        df["distance_km"],
        bins=[0, 50, 150, 300, float("inf")],
        labels=["short", "medium", "long", "very long"],
        include_lowest=True,
    ).astype(str)
    return df


def create_synthetic_time_logic(df):
    df = df.copy()
    for col in ["vehicle_type", "weather_condition", "delivery_mode", "region"]:
        df[col] = df[col].astype(str).str.strip().str.lower()

    vehicle_adjustment = {"bike": 1.2, "van": 0.5, "truck": 0.8, "ev van": 0.4}
    weather_adjustment = {"clear": 0.0, "cloudy": 0.2, "foggy": 0.6, "rainy": 0.8, "stormy": 1.2, "cold": 0.2, "hot": 0.2, "windy": 0.3}
    mode_adjustment = {"same day": 0.3, "express": 0.2, "two day": 0.7, "standard": 0.5}
    region_adjustment = {"central": 0.6, "north": 0.3, "south": 0.3, "east": 0.4, "west": 0.4}

    expected = (
        df["distance_km"] / 45
        + df["vehicle_type"].map(vehicle_adjustment).fillna(0.5)
        + df["weather_condition"].map(weather_adjustment).fillna(0.3)
        + df["delivery_mode"].map(mode_adjustment).fillna(0.4)
        + df["region"].map(region_adjustment).fillna(0.3)
    ).clip(lower=0.5)

    vehicle_mult = {"bike": 1.05, "van": 0.95, "truck": 1.02, "ev van": 0.97}
    weather_mult = {"clear": 0.95, "cloudy": 1.00, "foggy": 1.05, "rainy": 1.10, "stormy": 1.20, "cold": 1.02, "hot": 1.02, "windy": 1.03}
    mode_mult = {"same day": 1.05, "express": 1.02, "two day": 0.97, "standard": 1.00}
    region_mult = {"central": 1.08, "north": 1.00, "south": 1.01, "east": 1.02, "west": 1.03}

    actual = (
        expected
        * df["vehicle_type"].map(vehicle_mult).fillna(1)
        * df["weather_condition"].map(weather_mult).fillna(1)
        * df["delivery_mode"].map(mode_mult).fillna(1)
        * df["region"].map(region_mult).fillna(1)
    ).clip(lower=0.5)

    ratio = actual / expected
    balanced_actual = np.where(
        ratio < 0.98, expected * 0.95,
        np.where(ratio < 1.05, expected * 1.00,
                 np.where(ratio < 1.15, expected * 1.10, expected * 1.25))
    )
    df["expected_time_hours"] = expected.round(2)
    df["delivery_time_hours"] = pd.Series(balanced_actual).round(2)
    return df


def delay_score(delay):
    if delay <= 0: return 5
    if delay <= 2: return 4
    if delay <= 5: return 3
    if delay <= 8: return 2
    return 1


def performance_label(score):
    return {5: "excellent", 4: "good", 3: "average", 2: "poor", 1: "critical"}.get(int(score), "unknown")


def filter_df(df, vehicle, weather, mode, region):
    out = df.copy()
    filters = {"vehicle_type": vehicle, "weather_condition": weather, "delivery_mode": mode, "region": region}
    for col, selected in filters.items():
        if selected and "all" not in selected:
            out = out[out[col].isin(selected)]
    return out


def kpi_html(df):
    total = len(df)
    delay_rate = (df["calculated_delay"].eq("yes").mean() * 100) if total else 0
    avg_delay = df["delay_hours"].mean() if total else 0
    avg_score = df["delay_score"].mean() if total else 0
    cost = df["delivery_cost"].mean() if "delivery_cost" in df.columns and total else 0
    return f"""
    <div style='display:grid;grid-template-columns:repeat(4,minmax(0,1fr));gap:14px;'>
      <div class='metric-card'><div class='metric-label'>Deliveries analyzed</div><div class='metric-value'>{total:,.0f}</div></div>
      <div class='metric-card'><div class='metric-label'>Delay rate</div><div class='metric-value'>{delay_rate:.1f}%</div></div>
      <div class='metric-card'><div class='metric-label'>Average delay hours</div><div class='metric-value'>{avg_delay:.2f}</div></div>
      <div class='metric-card'><div class='metric-label'>Avg. delay score</div><div class='metric-value'>{avg_score:.2f}/5</div></div>
    </div>
    <p class='small-muted'>Average delivery cost in filtered data: {cost:,.2f}</p>
    """


def group_summary(df, col):
    return (
        df.groupby(col, observed=False)
        .agg(
            deliveries=(col, "size"),
            delay_rate=("calculated_delay", lambda x: round((x.eq("yes").mean() * 100), 2)),
            avg_delay_hours=("delay_hours", "mean"),
            avg_delay_score=("delay_score", "mean"),
            avg_distance_km=("distance_km", "mean"),
        )
        .round(2)
        .sort_values(["delay_rate", "avg_delay_hours"], ascending=False)
        .reset_index()
    )


def make_charts(df):
    by_vehicle = group_summary(df, "vehicle_type")
    by_weather = group_summary(df, "weather_condition")
    by_region = group_summary(df, "region")
    by_mode = group_summary(df, "delivery_mode")

    fig_vehicle = px.bar(by_vehicle, x="vehicle_type", y="delay_rate", text="delay_rate", title="Delay Risk by Vehicle Type")
    fig_weather = px.bar(by_weather, x="weather_condition", y="avg_delay_hours", text="avg_delay_hours", title="Average Delay Hours by Weather")
    fig_region = px.bar(by_region, x="region", y="delay_rate", text="delay_rate", title="Delay Rate by Region")
    fig_mode = px.bar(by_mode, x="delivery_mode", y="avg_delay_score", text="avg_delay_score", title="Performance Score by Delivery Mode")
    fig_scatter = px.scatter(df.sample(min(len(df), 2000), random_state=42), x="distance_km", y="delay_hours", color="weather_condition", hover_data=["vehicle_type", "delivery_mode", "region"], title="Distance vs Delay Hours")

    for fig in [fig_vehicle, fig_weather, fig_region, fig_mode, fig_scatter]:
        fig.update_layout(template="plotly_white", height=430, margin=dict(l=40, r=20, t=60, b=40))
    return fig_vehicle, fig_weather, fig_region, fig_mode, fig_scatter


def train_feature_importance(df):
    model_cols = ["vehicle_type", "weather_condition", "delivery_mode", "region", "distance_category", "distance_km", "package_weight_kg"]
    model_cols = [c for c in model_cols if c in df.columns]
    X = df[model_cols]
    y = df["calculated_delay"].eq("yes").astype(int)
    cat = [c for c in model_cols if X[c].dtype == "object" or str(X[c].dtype) == "category"]
    num = [c for c in model_cols if c not in cat]
    pre = ColumnTransformer([("cat", OneHotEncoder(handle_unknown="ignore"), cat), ("num", "passthrough", num)])
    clf = RandomForestClassifier(n_estimators=80, random_state=42, max_depth=7)
    pipe = Pipeline([("pre", pre), ("clf", clf)])
    pipe.fit(X, y)
    names = list(pipe.named_steps["pre"].get_feature_names_out())
    importances = pipe.named_steps["clf"].feature_importances_
    imp = pd.DataFrame({"factor": names, "importance": importances}).sort_values("importance", ascending=False).head(12)
    imp["factor"] = imp["factor"].str.replace("cat__", "", regex=False).str.replace("num__", "", regex=False)
    fig = px.bar(imp.sort_values("importance"), x="importance", y="factor", orientation="h", title="AI Model: Most Important Delay-Risk Drivers")
    fig.update_layout(template="plotly_white", height=470, margin=dict(l=120, r=20, t=60, b=40))
    return fig, imp


def auto_insights(df):
    if len(df) == 0:
        return "<div class='insight-box'>No data available for the selected filters.</div>"
    summaries = {c: group_summary(df, c) for c in ["vehicle_type", "weather_condition", "delivery_mode", "region", "distance_category"] if c in df.columns}
    worst = {k: v.iloc[0] for k, v in summaries.items() if len(v) > 0}
    best = {k: v.sort_values(["delay_rate", "avg_delay_hours"], ascending=True).iloc[0] for k, v in summaries.items() if len(v) > 0}
    top_risk_text = "<br>".join([f"• <b>{k.replace('_',' ').title()}</b>: highest risk = <b>{row[k]}</b> ({row['delay_rate']:.1f}% delay rate, {row['avg_delay_hours']:.2f} avg delay hours)" for k, row in worst.items()])
    best_text = "<br>".join([f"• <b>{k.replace('_',' ').title()}</b>: best performer = <b>{row[k]}</b> ({row['delay_rate']:.1f}% delay rate)" for k, row in best.items()])
    delay_rate = df["calculated_delay"].eq("yes").mean() * 100
    recommendation = "Prioritize operational buffers for the highest-risk combinations, especially where bad weather, central routes, same-day delivery, or slower vehicle types overlap."
    if delay_rate > 35:
        recommendation += " The current filtered scenario has a high delay rate, so management should add contingency capacity and proactively communicate expected delays to customers."
    else:
        recommendation += " The current filtered scenario is relatively manageable, so management can focus on monitoring and selective process improvements."
    return f"""
    <div class='insight-box'>
    <h3>AI-enhanced executive interpretation</h3>
    <p><b>Business challenge:</b> Which operational factors create the highest delivery-delay risk, and what should management do?</p>
    <p><b>Highest-risk factors found in the filtered data:</b><br>{top_risk_text}</p>
    <p><b>Best-performing conditions:</b><br>{best_text}</p>
    <p><b>Management action:</b> {recommendation}</p>
    <p><b>Qualitative interpretation:</b> Delay risk is not only a numeric issue. It affects customer trust, service reliability, driver planning, and cost control. The dashboard therefore combines quantitative KPIs with qualitative business recommendations.</p>
    </div>
    """


def update_dashboard(file_obj, vehicle, weather, mode, region):
    df = load_and_prepare(file_obj)
    fdf = filter_df(df, vehicle, weather, mode, region)
    if len(fdf) == 0:
        raise gr.Error("Your filters produced no rows. Select fewer filters.")
    figs = make_charts(fdf)
    model_fig, imp = train_feature_importance(fdf)
    sample = fdf.head(15)
    tables = [group_summary(fdf, c) for c in ["vehicle_type", "weather_condition", "region", "delivery_mode", "distance_category"]]
    return (kpi_html(fdf), auto_insights(fdf), *figs, model_fig, *tables, sample)


def choices_from_data(file_obj=None):
    df = load_and_prepare(file_obj)
    return [
        gr.update(choices=sorted(df["vehicle_type"].dropna().unique().tolist()), value=[]),
        gr.update(choices=sorted(df["weather_condition"].dropna().unique().tolist()), value=[]),
        gr.update(choices=sorted(df["delivery_mode"].dropna().unique().tolist()), value=[]),
        gr.update(choices=sorted(df["region"].dropna().unique().tolist()), value=[]),
    ]


def simulate_delivery(distance, weight, vehicle, weather, mode, region):
    row = pd.DataFrame({
        "distance_km": [distance], "package_weight_kg": [weight], "vehicle_type": [vehicle],
        "weather_condition": [weather], "delivery_mode": [mode], "region": [region]
    })
    row = create_synthetic_time_logic(row)
    row["delay_hours"] = (row["delivery_time_hours"] - row["expected_time_hours"]).round(2)
    row["delay_score"] = row["delay_hours"].apply(delay_score)
    row["performance_label"] = row["delay_score"].apply(performance_label)
    risk = "HIGH RISK" if row.loc[0, "delay_hours"] > 0 else "LOW RISK"
    return f"""
    ### Simulation Result
    - Expected delivery time: **{row.loc[0, 'expected_time_hours']:.2f} hours**
    - Predicted actual delivery time: **{row.loc[0, 'delivery_time_hours']:.2f} hours**
    - Predicted delay: **{row.loc[0, 'delay_hours']:.2f} hours**
    - Delay score: **{row.loc[0, 'delay_score']}/5**
    - Performance label: **{row.loc[0, 'performance_label'].title()}**
    - Risk classification: **{risk}**
    """


def download_summary(file_obj, vehicle, weather, mode, region):
    df = load_and_prepare(file_obj)
    fdf = filter_df(df, vehicle, weather, mode, region)
    summary = {
        "rows_analyzed": len(fdf),
        "delay_rate_percent": round(fdf["calculated_delay"].eq("yes").mean() * 100, 2),
        "average_delay_hours": round(fdf["delay_hours"].mean(), 2),
        "average_delay_score": round(fdf["delay_score"].mean(), 2),
    }
    lines = ["Delivery Delay Risk Executive Summary", "", "KPIs:"]
    for k, v in summary.items():
        lines.append(f"- {k.replace('_', ' ').title()}: {v}")
    lines += ["", "Highest-risk groups:"]
    for c in ["vehicle_type", "weather_condition", "delivery_mode", "region", "distance_category"]:
        tab = group_summary(fdf, c)
        row = tab.iloc[0]
        lines.append(f"- {c}: {row[c]} | delay rate {row['delay_rate']}% | avg delay {row['avg_delay_hours']}h")
    lines += ["", "Recommended actions:", "- Add operational buffers for high-risk weather and region combinations.", "- Match faster vehicle types to same-day and express deliveries.", "- Use the simulator before accepting risky delivery promises.", "- Monitor delay score weekly as an operational KPI."]
    tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".txt", mode="w", encoding="utf-8")
    tmp.write("\n".join(lines))
    tmp.close()
    return tmp.name


base_df = load_and_prepare(None)
vehicle_choices = sorted(base_df["vehicle_type"].dropna().unique().tolist())
weather_choices = sorted(base_df["weather_condition"].dropna().unique().tolist())
mode_choices = sorted(base_df["delivery_mode"].dropna().unique().tolist())
region_choices = sorted(base_df["region"].dropna().unique().tolist())

with gr.Blocks(theme=gr.themes.Soft(primary_hue="indigo", neutral_hue="slate"), css=CUSTOM_CSS, title="Delivery Delay Risk Dashboard") as demo:
    gr.Markdown("""
    # 🚚 Delivery Delay Risk Intelligence Dashboard
    **AI-enhanced operations dashboard for identifying delivery delay risk factors and management actions.**

    Upload your CSV or use the included dataset. The app cleans the data, generates realistic delivery-time logic, calculates delay risk, visualizes operational drivers, simulates new deliveries, and creates an executive summary.
    """)

    with gr.Row():
        file_input = gr.File(label="Optional: upload your real-world/found delivery CSV", file_types=[".csv"])
        refresh_btn = gr.Button("Load / refresh data", variant="primary")

    with gr.Accordion("Filters", open=True):
        with gr.Row():
            vehicle_filter = gr.Dropdown(vehicle_choices, label="Vehicle type", multiselect=True)
            weather_filter = gr.Dropdown(weather_choices, label="Weather condition", multiselect=True)
            mode_filter = gr.Dropdown(mode_choices, label="Delivery mode", multiselect=True)
            region_filter = gr.Dropdown(region_choices, label="Region", multiselect=True)

    kpis = gr.HTML()
    insights = gr.HTML()

    with gr.Tab("Interactive dashboard"):
        with gr.Row():
            fig_vehicle = gr.Plot()
            fig_weather = gr.Plot()
        with gr.Row():
            fig_region = gr.Plot()
            fig_mode = gr.Plot()
        fig_scatter = gr.Plot()

    with gr.Tab("AI risk-driver model"):
        model_fig = gr.Plot()
        gr.Markdown("This section trains a simple Random Forest model inside the app to estimate which factors are most important for predicting delays.")

    with gr.Tab("Summary tables"):
        with gr.Row():
            vehicle_table = gr.Dataframe(label="Vehicle performance")
            weather_table = gr.Dataframe(label="Weather performance")
        with gr.Row():
            region_table = gr.Dataframe(label="Region performance")
            mode_table = gr.Dataframe(label="Delivery mode performance")
        distance_table = gr.Dataframe(label="Distance category performance")
        sample_table = gr.Dataframe(label="Cleaned sample data")

    with gr.Tab("Delivery risk simulator"):
        with gr.Row():
            sim_distance = gr.Slider(1, 500, value=120, label="Distance km")
            sim_weight = gr.Slider(0.1, 60, value=10, label="Package weight kg")
        with gr.Row():
            sim_vehicle = gr.Dropdown(vehicle_choices, value=vehicle_choices[0], label="Vehicle")
            sim_weather = gr.Dropdown(weather_choices, value=weather_choices[0], label="Weather")
            sim_mode = gr.Dropdown(mode_choices, value=mode_choices[0], label="Mode")
            sim_region = gr.Dropdown(region_choices, value=region_choices[0], label="Region")
        sim_btn = gr.Button("Simulate delivery risk", variant="primary")
        sim_output = gr.Markdown()

    with gr.Tab("Download executive summary"):
        gr.Markdown("Generate a short text summary for your presentation/report.")
        download_btn = gr.Button("Create executive summary file")
        download_file = gr.File(label="Download summary")

    outputs = [kpis, insights, fig_vehicle, fig_weather, fig_region, fig_mode, fig_scatter, model_fig, vehicle_table, weather_table, region_table, mode_table, distance_table, sample_table]
    refresh_btn.click(update_dashboard, inputs=[file_input, vehicle_filter, weather_filter, mode_filter, region_filter], outputs=outputs)
    file_input.change(choices_from_data, inputs=[file_input], outputs=[vehicle_filter, weather_filter, mode_filter, region_filter])
    sim_btn.click(simulate_delivery, inputs=[sim_distance, sim_weight, sim_vehicle, sim_weather, sim_mode, sim_region], outputs=sim_output)
    download_btn.click(download_summary, inputs=[file_input, vehicle_filter, weather_filter, mode_filter, region_filter], outputs=download_file)
    demo.load(update_dashboard, inputs=[file_input, vehicle_filter, weather_filter, mode_filter, region_filter], outputs=outputs)

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