Update app.py

app.py

@@ -1,109 +1,90 @@
 import gradio as gr
 import pandas as pd
-import numpy as np
+import torch
 import yfinance as yf
 import plotly.graph_objects as go
-from plotly.subplots import make_subplots
-import timesfm
+from chronos import ChronosPipeline
 
-# --- SDK v2.5 INITIALIZATION ---
-# Load the 200M parameter model specifically for PyTorch
-model = timesfm.TimesFM_2p5_200M_torch.from_pretrained(
-    "google/timesfm-2.5-200m-pytorch"
+# --- LOAD AMAZON CHRONOS-2 ---
+# This model treats time-series as a language, making it more flexible
+pipeline = ChronosPipeline.from_pretrained(
+    "amazon/chronos-t5-base",
+    device_map="cpu",  # Use "cuda" if you have a GPU
+    torch_dtype=torch.float32,
 )
 
-# FIXED: Parameters are now moved into ForecastConfig
-model.compile(timesfm.ForecastConfig(
-    max_context=1024,      # v2.5 now supports up to 16k
-    max_horizon=256,
-    normalize_inputs=True,
-    infer_is_positive=True
-))
-
-def generate_professional_chart(df, forecast_data, ticker, is_backtest=False):
-    fig = make_subplots(rows=2, cols=1, shared_xaxes=True, 
-                        vertical_spacing=0.03, row_heights=[0.8, 0.2])
-
+def get_pro_chart(df, forecast_samples, ticker, is_backtest=False):
+    # Chronos provides a distribution (multiple possible futures)
+    # We take the median (50th percentile) for the main line
+    low, median, high = np.quantile(forecast_samples, [0.1, 0.5, 0.9], axis=0)
+    
+    fig = go.Figure()
     hist_df = df[:-30] if is_backtest else df
     
-    # Historical Trend
-    fig.add_trace(go.Scatter(x=hist_df.index, y=hist_df['Close'], name='Market Price',
-                             line=dict(color='#2962FF', width=2)), row=1, col=1)
+    # 1. Historical Data
+    fig.add_trace(go.Scatter(x=hist_df.index, y=hist_df['Close'], name='History', line=dict(color='#2962FF')))
 
-    # Reality Check (Dotted)
-    if is_backtest:
-        fig.add_trace(go.Scatter(x=df.index[-30:], y=df['Close'][-30:], name='Actual Path',
-                                 line=dict(color='#787b86', dash='dot')), row=1, col=1)
-
-    # AI Forecast Line
-    fc_dates = pd.date_range(start=hist_df.index[-1], periods=len(forecast_data)+1, freq='B')[1:]
-    fig.add_trace(go.Scatter(x=fc_dates, y=forecast_data, name='AI Forecast',
-                             line=dict(color='#F23645', width=3)), row=1, col=1)
+    # 2. Future Forecast (Median)
+    last_date = hist_df.index[-1]
+    fc_dates = pd.date_range(start=last_date, periods=len(median)+1, freq='B')[1:]
+    
+    # Confidence Interval (The "Cloud")
+    fig.add_trace(go.Scatter(
+        x=list(fc_dates) + list(fc_dates)[::-1],
+        y=list(high) + list(low)[::-1],
+        fill='toself', fillcolor='rgba(242, 54, 69, 0.1)',
+        line=dict(color='rgba(255,255,255,0)'), name='Confidence Range'
+    ))
+    
+    fig.add_trace(go.Scatter(x=fc_dates, y=median, name='AI Median Forecast', line=dict(color='#F23645', width=3)))
 
-    # Volume Bars
-    fig.add_trace(go.Bar(x=df.index, y=df['Volume'], name='Volume',
-                         marker_color='rgba(120, 123, 134, 0.3)'), row=2, col=1)
+    if is_backtest:
+        fig.add_trace(go.Scatter(x=df.index[-30:], y=df['Close'][-30:], name='Actual', line=dict(color='#787b86', dash='dot')))
 
-    fig.update_layout(template='plotly_dark', paper_bgcolor='#131722', plot_bgcolor='#131722',
-                      margin=dict(l=50, r=50, t=30, b=30), hovermode="x unified")
-    fig.update_yaxes(side='right', gridcolor='#2a2e39')
-    fig.update_xaxes(gridcolor='#2a2e39')
+    fig.update_layout(template='plotly_dark', paper_bgcolor='#131722', plot_bgcolor='#131722', margin=dict(l=10, r=10, t=40, b=10))
     return fig
 
-def run_terminal(ticker, horizon, mode):
+def analyze(ticker, horizon, mode):
     try:
-        # 1. Data Ingestion
-        data = yf.download(ticker, period="2y")
-        if data.empty: return None, "⚠️ SYMBOL NOT FOUND", None
+        df = yf.download(ticker, period="2y")
+        if df.empty: return None, "Symbol not found", None
         
-        train_data = data[:-30] if mode == "Backtest Mode" else data
+        context_data = df[:-30] if mode == "Backtest" else df
+        context_tensor = torch.tensor(context_data['Close'].values)
         
-        # 2. SDK Inference (Array-based input for v2.5)
-        inputs = [train_data['Close'].values]
-        point_forecast, _ = model.forecast(
-            inputs=inputs, 
-            horizon=30 if mode == "Backtest Mode" else horizon
-        )
-        prediction = point_forecast[0]
-
-        # 3. Financial Metrics
-        pct_change = ((prediction[-1] - train_data['Close'].iloc[-1]) / train_data['Close'].iloc[-1]) * 100
+        # Chronos Inference
+        # h = horizon, num_samples = 20 (to get a confidence range)
+        forecast = pipeline.predict(context_tensor, horizon if mode != "Backtest" else 30, num_samples=20)
+        forecast_np = forecast.numpy()[0] # Shape: [samples, horizon]
+        median_fc = np.median(forecast_np, axis=0)
         
-        if mode == "Future Forecast":
-            signal = f"<h2 style='color: {'#00ff88' if pct_change > 0 else '#ff4444'}; text-align: center;'>" \
-                     f"{'BULLISH' if pct_change > 0 else 'BEARISH'} ({pct_change:+.2f}%)</h2>"
-        else:
-            # Calculate Backtest Accuracy
-            accuracy = 100 - abs((data['Close'].iloc[-1] - prediction[-1]) / data['Close'].iloc[-1] * 100)
-            signal = f"<h2 style='color: #FFD700; text-align: center;'>AI ACCURACY: {accuracy:.1f}%</h2>"
-
-        # 4. Generate Report
-        report_path = f"{ticker}_AI_Report.csv"
-        pd.DataFrame({'Forecast_Price': prediction}).to_csv(report_path, index=False)
+        # Results & Signals
+        change = ((median_fc[-1] - context_data['Close'].iloc[-1]) / context_data['Close'].iloc[-1]) * 100
+        signal = f"<h2 style='color: {'#00ff88' if change > 0 else '#ff4444'};'>{'BULLISH' if change > 0 else 'BEARISH'} ({change:+.2f}%)</h2>"
         
-        chart = generate_professional_chart(data, prediction, ticker, (mode == "Backtest Mode"))
-        return chart, signal, report_path
-
+        csv_path = "forecast.csv"
+        pd.DataFrame(forecast_np.T).to_csv(csv_path)
+        
+        return get_pro_chart(df, forecast_np, ticker, mode == "Backtest"), signal, csv_path
     except Exception as e:
-        return None, f"<div style='color:red;'>API Error: {str(e)}</div>", None
+        return None, f"Error: {str(e)}", None
 
-# --- TERMINAL UI ---
-with gr.Blocks(title="G-TIMES QUANT 2.5", theme=gr.themes.Base()) as demo:
-    gr.HTML("<div style='background-color:#131722; padding:20px; border-bottom:3px solid #2962FF; text-align:center;'>"
-            "<h1 style='color:white; letter-spacing:3px;'>G-TIMES <span style='color:#2962FF;'>QUANT 2.5</span></h1></div>")
-    
+import numpy as np # Needed for quantiles
+
+# --- GRADIO UI ---
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.HTML("<h1 style='text-align:center; color:#2962FF;'>CHRONOS-2 ANALYTICS</h1>")
     with gr.Row():
         with gr.Column(scale=1):
-            ticker_in = gr.Textbox(label="TICKER", value="TSLA")
-            mode_in = gr.Radio(["Future Forecast", "Backtest Mode"], label="STRATEGY", value="Future Forecast")
-            days_in = gr.Slider(7, 128, value=30, label="HORIZON")
-            btn = gr.Button("RUN QUANT ANALYSIS", variant="primary")
-            status_out = gr.HTML()
-            file_out = gr.File(label="CSV EXPORT")
-            
+            t_in = gr.Textbox(label="Ticker", value="AAPL")
+            m_in = gr.Radio(["Future Forecast", "Backtest"], label="Mode", value="Future Forecast")
+            h_in = gr.Slider(7, 90, value=30, label="Days")
+            btn = gr.Button("RUN AI MODEL", variant="primary")
+            msg = gr.HTML()
+            file = gr.File(label="Export Data")
         with gr.Column(scale=4):
-            plot_out = gr.Plot()
+            plot = gr.Plot()
 
-    btn.click(run_terminal, [ticker_in, days_in, mode_in], [plot_out, status_out, file_out])
+    btn.click(analyze, [t_in, h_in, m_in], [plot, msg, file])
 
 demo.launch()