import gradio as gr
import pandas as pd
import numpy as np
import tensorflow as tf
from sklearn.preprocessing import StandardScaler
from keras.saving import register_keras_serializable

@register_keras_serializable()
class SimplifiedTFT_Iter3(tf.keras.Model):
    ...
# Load your trained TFT model
model = tf.keras.models.load_model("tft_model.keras", compile=False)

# Load scalers if saved separately (optional), or define here again

def predict_from_csv(file):
    df = pd.read_csv(file.name)
    
    # Perform the same preprocessing as during training
    # This must match what you did before model.fit()

    # For demo, let's assume the last N rows have the correct shape
    input_data = np.expand_dims(df.tail(1).values, axis=0)
    
    # Make prediction
    pred = model.predict(input_data)
    return f"Prediction: {pred.flatten()[0]}"

# Gradio interface
gr.Interface(fn=predict_from_csv, inputs="file", outputs="text").launch()