import gradio as gr
import pandas as pd
import joblib
import re
import matplotlib.pyplot as plt

products = pd.read_csv("products_master.csv")
model    = joblib.load("rf_model.joblib")

def shorten(name, max_len=85):
    # Base product = everything before first comma
    base = name.split(",")[0].strip()

    # Extract storage like "16 GB", "32 GB"
    storage_match = re.search(r"(\d+)\s*GB", name)
    storage = storage_match.group(0) if storage_match else ""

    # Detect colour / variant (longer phrases first)
    colors = [
        "Marine Blue", "Pink Kid-Proof Case", "Blue Kid-Proof Case",
        "Green Kid-Proof Case", "Tangerine", "Magenta",
        "Black", "Blue", "Pink", "Green", "White", "Yellow", "Red"
    ]
    found_color = ""
    for c in colors:
        if c.lower() in name.lower():
            found_color = c
            break

    # Detect ads vs ad-free
    n_lower = name.lower()
    if "without special offer" in n_lower or "no special offer" in n_lower:
        offers = "Ad-Free"
    elif "special offer" in n_lower:
        offers = "with Ads"
    else:
        offers = ""

    extras = [x for x in [storage, found_color, offers] if x]
    label = base + (" – " + " / ".join(extras) if extras else "")

    if len(label) > max_len:
        label = label[:max_len-1] + "…"
    return label

# Build (label, value) pairs and de-duplicate any remaining collisions
raw = [(shorten(n), n) for n in products["name"].tolist()]
seen = {}
dropdown_choices = []
for label, full in raw:
    if label in seen:
        seen[label] += 1
        label = f"{label} (v{seen[label]})"
    else:
        seen[label] = 1
    dropdown_choices.append((label, full))
dropdown_choices.sort(key=lambda x: x[0])

def make_recommendation(product_name):
    row = products[products["name"] == product_name].iloc[0]

    fig, ax = plt.subplots(figsize=(5, 3))
    counts = [row["pct_positive"], row["pct_neutral"], row["pct_negative"]]
    ax.bar(["Positive", "Neutral", "Negative"],
           counts, color=["#2ecc71", "#95a5a6", "#e74c3c"], edgecolor="black")
    ax.set_ylabel("% of reviews"); ax.set_title("Review sentiment breakdown"); ax.set_ylim(0, 1)

    action = row["model_prediction"]
    emoji  = {"Raise": "⬆️", "Hold": "➡️", "Drop": "⬇️"}[action]

    summary = f"""
### {shorten(row["name"], max_len=110)}

**Brand:** {row["brand"]}  
**Reviews analyzed:** {int(row["n_reviews"]):,}  
**Average rating:** {row["avg_rating"]:.2f} / 5  
**Average sentiment (VADER):** {row["avg_compound"]:.3f}

---

### {emoji} Recommendation: **{action} the price**

| Metric | Current | Recommended |
|---|---|---|
| Price | ${row["current_price"]:.2f} | ${row["recommended_price"]:.2f} |
| Profit per unit | ${row["current_profit_per_unit"]:.2f} | ${row["recommended_profit_per_unit"]:.2f} |

**Estimated monthly profit change:** **${row["monthly_profit_change"]:,.2f}**
"""
    return summary, fig


with gr.Blocks(title="Amazon Electronics Pricing Recommender") as demo:
    gr.Markdown("# 🛒 Amazon Electronics Pricing Recommender")
    gr.Markdown("Pick a product to see its sentiment breakdown and a price recommendation based on a Random Forest model trained on real reviews.")

    product_dropdown = gr.Dropdown(
        choices=dropdown_choices,
        label="Select a product",
        value=dropdown_choices[0][1],
    )
    btn = gr.Button("Get Recommendation", variant="primary")

    with gr.Row():
        output_text  = gr.Markdown()
        output_chart = gr.Plot()

    btn.click(make_recommendation, inputs=product_dropdown, outputs=[output_text, output_chart])
    demo.load (make_recommendation, inputs=product_dropdown, outputs=[output_text, output_chart])

demo.launch()