# Hosted in https://huggingface.co/spaces/ArturoNereu/emojilator
import gradio as gr
import spacy
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np

# Load the NLP model, downloaded via python
nlp = spacy.load("en_core_web_sm")
embedder = SentenceTransformer("all-MiniLM-L6-v2")

# Basic emoji dictionary
emoji_map = {
    # Food
    "pizza": "🍕",
    "burger": "🍔",
    "fries": "🍟",
    "sushi": "🍣",
    "coffee": "☕",
    "ice cream": "🍦",
    "cake": "🎂",
    "beer": "🍺",
    "wine": "🍷",
    "apple": "🍎",
    "banana": "🍌",
    "taco": "🌮",
    "chocolate": "🍫",

    # Animals
    "dog": "🐶",
    "cat": "🐱",
    "lion": "🦁",
    "monkey": "🐒",
    "panda": "🐼",
    "elephant": "🐘",
    "fish": "🐟",
    "bird": "🐦",
    "horse": "🐴",

    # Places & activities
    "beach": "🏖️",
    "mountain": "⛰️",
    "city": "🏙️",
    "school": "🏫",
    "work": "💼",
    "party": "🎉",
    "swim": "🏊",
    "run": "🏃",
    "travel": "✈️",
    "camping": "🏕️",
    "movie": "🎬",

    # Objects
    "rocket": "🚀",
    "book": "📚",
    "phone": "📱",
    "computer": "💻",
    "car": "🚗",
    "bicycle": "🚲",
    "guitar": "🎸",
    "music": "🎵",

    # Emotions
    "happy": "😄",
    "sad": "😢",
    "angry": "😡",
    "love": "❤️",
    "laugh": "😂",
    "sleep": "😴"
}

# Predefined sentences to transform
sentences = [
    "I love pizza and burgers",
    "My dog and cat are so cute",
    "Let's go swim at the beach",
    "The rocket launch was amazing",
    "I am reading a good book",
    "We are going camping in the mountains",
    "Happy birthday! I brought cake",
    "I need coffee and chocolate",
    "I want to travel and see new cities"
]

# Generate the embeddings on the emojis' names
emoji_names = list(emoji_map.keys())
emoji_embeddings = embedder.encode(emoji_names)

def translate_to_emoji(text):
    doc = nlp(text)
    keywords = [token.text for token in doc if token.pos_ in ["NOUN", "VERB"]]

    word_embeddings = embedder.encode(keywords)
    result = text

    for i, word in enumerate(keywords):
        sims = cosine_similarity([word_embeddings[i]], emoji_embeddings)[0]
        best_idx = np.argmax(sims)
        best_emoji = emoji_map[emoji_names[best_idx]]

        # Thee 0.5 is just a threshold to make sure the replacement makes some sense
        if sims[best_idx] > 0.5:
            result = result.replace(word, f"{word} {best_emoji}")

    return result

# Gradio UI
with gr.Blocks(theme="soft") as app:
    gr.Markdown("# Emojilator 🦖")
    gr.Markdown("AI-powered emoji translator: make your text expressive with semantic emoji matching!")

    with gr.Row():
        with gr.Column(scale=1):
            gr.Image("Frida_Dino.png", elem_id="mascot", show_label=False)
        with gr.Column(scale=2):
            gr.Markdown("""
            ### About this Space 🧠
            This is a simple Space to help build intuition around the concept of embeddings.

            Most of the magic behind LLMs for developers revolves around understanding this concept: converting data into a numeric representation, or a vector.

            For example, the word apple could become [8348832, 3423943, 923493294]. If we use the 🍎 emoji, its embedding might be [432432, 43423, 3423423]. When we calculate the distance between them, they're close!

            In this demo, we use that distance to replace words with relevant emojis. What's interesting is that it's not only about exact matches—if you write "I want ICE CREAM", you might get 🍦, not 🧊 + 🧴.

            This means we're capturing meaning (semantics), not just character-to-character equality.

            This is why "ice cream" gives 🍦 instead of 🧊 + 🧴.
            """)

            # Dropdown instead of textbox
            phrase_selector = gr.Dropdown(
                label="Select a phrase",
                choices=sentences,
                value=sentences[0]
            )

            output_box = gr.Textbox(label="Text with emojis", lines=3)

    # Automatically translate when a phrase is selected
    phrase_selector.change(translate_to_emoji, inputs=phrase_selector, outputs=output_box)

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