import gradio as gr
import joblib
import numpy as np
import pandas as pd
import plotly.graph_objects as go

# -----------------------------
# Load trained model
# -----------------------------
MODEL_PATH = "models/influencer_roi_model.joblib"
model = joblib.load(MODEL_PATH)

# IMPORTANT: These must match the training feature columns (X.columns)
# We'll load them from a saved list if you created one.
# If you didn't save feature names yet, do Step 2.1 below.
FEATURES_PATH = "models/feature_columns.joblib"
feature_columns = joblib.load(FEATURES_PATH)


# -----------------------------
# Helper: create input row
# -----------------------------
def build_input_row(platform, influencer_category, campaign_type,
                    engagements, estimated_reach, campaign_duration_days,
                    assumed_cost_usd):
    """
    Converts UI inputs into a single-row DataFrame with exact feature columns.
    We one-hot encode platform/category/type to match training.
    """

    # Base numeric features
    base = {
        "engagements": engagements,
        "estimated_reach": estimated_reach,
        "campaign_duration_days": campaign_duration_days,
    }

    # Build a raw dataframe with the same pre-encoded columns you had before encoding:
    raw = pd.DataFrame([{
        **base,
        "platform": platform,
        "influencer_category": influencer_category,
        "campaign_type": campaign_type
    }])

    # One-hot encode like training (drop_first=True)
    encoded = pd.get_dummies(
        raw,
        columns=["platform", "influencer_category", "campaign_type"],
        drop_first=True
    )

    # Align to training columns (missing columns -> 0, extra -> drop)
    for col in feature_columns:
        if col not in encoded.columns:
            encoded[col] = 0
    encoded = encoded[feature_columns]

    return encoded


# -----------------------------
# Main prediction function
# -----------------------------
def predict_sales_and_roi(platform, influencer_category, campaign_type,
                          engagements, estimated_reach, campaign_duration_days,
                          assumed_cost_usd):
    """
    Predict product_sales and compute an *estimated ROI* using user-provided cost.
    ROI = (Predicted Sales - Cost) / Cost
    """

    # Guard against zero/negative cost
    if assumed_cost_usd <= 0:
        return "⚠️ Please enter a campaign cost greater than 0.", None

    X_input = build_input_row(
        platform, influencer_category, campaign_type,
        engagements, estimated_reach, campaign_duration_days,
        assumed_cost_usd
    )

    # Predict sales
    pred_sales = float(model.predict(X_input)[0])

    # Compute ROI (as %)
    roi = (pred_sales - assumed_cost_usd) / assumed_cost_usd
    roi_percent = roi * 100.0

    # Label ROI quality
    if roi_percent >= 50:
        roi_label = "🟢 Strong ROI"
    elif roi_percent >= 0:
        roi_label = "🟡 Moderate ROI"
    else:
        roi_label = "🔴 Negative ROI"

    # Build Plotly chart: Cost vs Predicted Sales
    fig = go.Figure()
    fig.add_bar(name="Campaign Cost", x=["Campaign"], y=[assumed_cost_usd])
    fig.add_bar(name="Predicted Sales", x=["Campaign"], y=[pred_sales])
    fig.update_layout(
        barmode="group",
        title="Spend vs Predicted Return",
        xaxis_title="",
        yaxis_title="USD",
        height=420
    )

    # Suggestions (simple, user-friendly)
    tips = []
    if engagements < 2000:
        tips.append("Increase engagement (better content hook, CTA, posting time).")
    if estimated_reach < 10000:
        tips.append("Improve reach (cross-post, collab posts, hashtags, boosted content).")
    if campaign_duration_days < 7:
        tips.append("Consider longer campaigns (more touchpoints often improves conversions).")

    if not tips:
        tips_text = "Your inputs look strong. Focus on creative quality + audience match."
    else:
        tips_text = " • " + "\n • ".join(tips)

    result_md = f"""
# 📈 Influencer ROI Prediction

### Predicted Financial Return
- **Predicted Product Sales:** `${pred_sales:,.2f}`
- **Campaign Cost:** `${assumed_cost_usd:,.2f}`
- **Estimated ROI:** **{roi_percent:,.2f}%**  — {roi_label}

### What this means
This prediction helps brands estimate whether a campaign is likely to be profitable **before spending money**.

### Suggestions to improve ROI
{tips_text}
"""

    return result_md, fig


# -----------------------------
# UI (modern dashboard feel)
# -----------------------------
with gr.Blocks(title="Influencer ROI Prediction Dashboard") as app:
    with gr.Tab("Home"):
        gr.Markdown("""
# ✨ Influencer ROI Prediction Dashboard

This app predicts **financial return (product sales)** from influencer campaign inputs using **Regression ML**.

### Why it matters
Marketing teams can use this to:
- Compare influencer platforms (Instagram vs TikTok vs YouTube)
- Estimate sales impact before launching a campaign
- Run **what-if scenarios** (change budget, duration, engagement)

➡️ Go to the **Predict** tab to try it.
""")

    with gr.Tab("Predict"):
        gr.Markdown("## 🧠 Predict Sales + Estimated ROI")

        with gr.Row():
            with gr.Column(scale=2):
                platform = gr.Dropdown(
                    choices=["Instagram", "YouTube", "TikTok", "Twitter", "Facebook"],
                    value="Instagram",
                    label="Platform"
                )
                influencer_category = gr.Dropdown(
                    choices=["Micro", "Macro", "Mega", "Nano"],
                    value="Micro",
                    label="Influencer Category"
                )
                campaign_type = gr.Dropdown(
                    choices=["Product Launch", "Brand Awareness", "Seasonal Promo", "Giveaway", "Affiliate"],
                    value="Brand Awareness",
                    label="Campaign Type"
                )

                engagements = gr.Number(value=5000, label="Engagements (likes/comments/etc.)")
                estimated_reach = gr.Number(value=50000, label="Estimated Reach")
                campaign_duration_days = gr.Slider(1, 60, value=14, step=1, label="Campaign Duration (days)")
                assumed_cost_usd = gr.Number(value=1000, label="Campaign Cost (USD)")

                predict_btn = gr.Button("🚀 Predict ROI", variant="primary")

            with gr.Column(scale=3):
                output_md = gr.Markdown()
                output_plot = gr.Plot()

        predict_btn.click(
            fn=predict_sales_and_roi,
            inputs=[platform, influencer_category, campaign_type,
                    engagements, estimated_reach, campaign_duration_days,
                    assumed_cost_usd],
            outputs=[output_md, output_plot]
        )

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