import gradio as gr
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import numpy as np

BASE_MODEL = "AlekseyDorkin/xlm-roberta-en-ru-emoji"
TOP_N = 5

model = AutoModelForSequenceClassification.from_pretrained(BASE_MODEL)
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)

def preprocess(text):
    new_text = []
    for t in text.split(" "):
        t = '@user' if t.startswith('@') and len(t) > 1 else t
        t = 'http' if t.startswith('http') else t
        new_text.append(t)
    return " ".join(new_text)
    
def get_top_emojis(text, top_n=TOP_N):
    preprocessed = preprocess(text)
    inputs = tokenizer(preprocessed)
    preds = model(**inputs)
    scores = torch.nn.functional.softmax(preds, axis=-1).detach().numpy()
    ranking = np.argsort(scores)
    ranking = ranking[::-1][:top_n]
    emojis = [model.config.id2label[index] for index in ranking]
    return emojis
    
    
gradio_ui = gr.Interface(
    fn=get_top_emojis,
    title="Predicting review scores from customer reviews",
    description="Enter some review text about an Amazon product and check what the model predicts for it's star rating.",
    inputs=[
        gr.inputs.Textbox(lines=5, label="Paste some text here"),
    ],
    outputs=[
        gr.outputs.Textbox(label=f"№{i}") for i in range(TOP_N)
    ],
    examples=[
        ["Awesome!"], ["Круто!"], ["lol"]
    ],
)

gradio_ui.launch(debug=True)