import gradio as gr
import pandas as pd
import numpy as np
from scipy.fft import fft, fftfreq
from sklearn.preprocessing import MinMaxScaler
from tensorflow.keras.models import Sequential, load_model
import requests

# --- Pre-trained Model (Simple LSTM) ---
def build_model():
    model = Sequential([
        tf.keras.layers.LSTM(32, input_shape=(30, 1)),
        tf.keras.layers.Dense(1)
    ])
    model.compile(loss='mse', optimizer='adam')
    return model

# --- Core Functions ---
def analyze_data(data_url, prediction_days=30):
    try:
        # 1. Fetch data
        df = pd.read_csv(data_url)
        dates = df.columns[4:]  # COVID data format
        values = df.drop(columns=['Province/State', 'Country/Region', 'Lat', 'Long']).sum(axis=0)[4:].values.astype(float)
        
        # 2. Detect cycles
        N = len(values)
        yf = fft(values)
        xf = fftfreq(N, 1)[:N//2]
        dominant_freq = xf[np.argmax(np.abs(yf[0:N//2]))]
        cycle_days = int(1/dominant_freq)
        
        # 3. Make predictions (simplified)
        scaler = MinMaxScaler()
        scaled = scaler.fit_transform(values.reshape(-1, 1))
        
        model = build_model()
        model.fit(scaled[:-10], scaled[10:], epochs=5, verbose=0)  # Quick training
        
        preds = model.predict(scaled[-30:].reshape(1, 30, 1))
        preds = scaler.inverse_transform(preds).flatten().tolist()
        
        # 4. Generate insights
        insights = [
            f"Dominant cycle: {cycle_days} days",
            f"Next {prediction_days}-day trend: {'↑ Upward' if preds[-1] > preds[0] else '↓ Downward'}",
            "Action: Monitor closely around cycle peaks"
        ]
        
        # Simple plot
        plot = pd.DataFrame({
            'Historical': values,
            'Predicted': [None]*(len(values)) + preds
        }).plot(title="Cases Analysis").figure
        
        return plot, insights
    
    except Exception as e:
        return None, [f"Error: {str(e)}"]

# --- Gradio Interface ---
interface = gr.Interface(
    fn=analyze_data,
    inputs=[
        gr.Textbox(label="Data URL", 
                   value="https://raw.githubusercontent.com/CSSEGISandData/COVID-19/master/data/time_series_covid19_confirmed_global.csv"),
        gr.Number(label="Days to Predict", value=30)
    ],
    outputs=[
        gr.Plot(label="Analysis"),
        gr.JSON(label="Insights")
    ],
    title="DeepSeek Lite Analyzer",
    description="Analyze time-series data from public URLs. Works best with COVID-19 format data."
)

interface.launch()