Update app.py

app.py

@@ -4,42 +4,38 @@ import numpy as np
 import yfinance as yf
 import plotly.graph_objects as go
 from plotly.subplots import make_subplots
-from timesfm import TimesFm, TimesFmHparams, TimesFmCheckpoint
+import timesfm
 
-# --- JAX INITIALIZATION ---
-# Using the native JAX/Flax checkpoint
-tfm = TimesFm(
-    hparams=TimesFmHparams(
-        backend="cpu", # JAX handles CPU/GPU automatically
-        per_core_batch_size=32,
-        horizon_len=128,
-        context_len=512,
-        num_layers=20,
-        model_dims=1280,
-    ),
-    checkpoint=TimesFmCheckpoint(
-        huggingface_repo_id="google/timesfm-1.0-200m" # Original JAX weights
-    ),
+# --- UPDATED INITIALIZATION (v2.5 API) ---
+# We use the new torch-specific class and ForecastConfig object
+model = timesfm.TimesFM_2p5_200M_torch.from_pretrained(
+    "google/timesfm-2.5-200m-pytorch"
 )
 
+# Configure the model using the new ForecastConfig
+model.compile(timesfm.ForecastConfig(
+    max_context=512,
+    max_horizon=128,
+    normalize_inputs=True,
+    infer_is_positive=True
+))
+
 def get_financial_plot(df, forecast_df, ticker, is_backtest=False):
     fig = make_subplots(rows=2, cols=1, shared_xaxes=True, 
                         vertical_spacing=0.05, row_heights=[0.75, 0.25])
 
-    # Plotting logic remains consistent for professional look
     display_df = df[:-30] if is_backtest else df
     
-    # 1. Price
+    # 1. Historical Price
     fig.add_trace(go.Scatter(x=display_df.index, y=display_df['Close'], name='History', line=dict(color='#2962FF')), row=1, col=1)
 
-    # 2. Backtest Reality (Truth)
+    # 2. Backtest Truth
     if is_backtest:
         fig.add_trace(go.Scatter(x=df.index[-30:], y=df['Close'][-30:], name='Actual', line=dict(color='#787b86', dash='dot')), row=1, col=1)
 
-    # 3. Forecast
-    fc_dates = [display_df.index[-1]] + list(forecast_df['ds'])
-    fc_vals = [display_df['Close'].iloc[-1]] + list(forecast_df['timesfm'])
-    fig.add_trace(go.Scatter(x=fc_dates, y=fc_vals, name='AI Forecast', line=dict(color='#F23645', width=3)), row=1, col=1)
+    # 3. AI Forecast
+    fc_dates = pd.date_range(start=display_df.index[-1], periods=len(forecast_df)+1, freq='D')[1:]
+    fig.add_trace(go.Scatter(x=fc_dates, y=forecast_df, name='AI Forecast', line=dict(color='#F23645', width=3)), row=1, col=1)
 
     # 4. Volume
     fig.add_trace(go.Bar(x=df.index, y=df['Volume'], name='Volume', marker_color='rgba(38, 166, 154, 0.5)'), row=2, col=1)
@@ -50,51 +46,51 @@ def get_financial_plot(df, forecast_df, ticker, is_backtest=False):
 def run_analysis(ticker, horizon, mode):
     try:
         df = yf.download(ticker, period="2y")
-        if df.empty: return None, "⚠️ Ticker Not Found"
+        if df.empty: return None, "⚠️ Ticker Not Found", None
         
         train_df = df[:-30] if mode == "Backtest (Reality Check)" else df
         
-        # Prepare inputs for TimesFM
-        input_df = pd.DataFrame({
-            'unique_id': [ticker],
-            'ds': train_df.index,
-            'y': train_df['Close'].values
-        })
-
-        # Forecast Execution (Native JAX backend)
-        forecast_df, _ = tfm.forecast_on_df(
-            inputs=input_df,
-            freq="D", 
-            value_name="y"
+        # New API uses simple array inputs
+        inputs = [train_df['Close'].values]
+        
+        # Run inference using the v2.5 forecast method
+        point_forecast, _ = model.forecast(
+            inputs=inputs,
+            horizon=horizon if mode != "Backtest (Reality Check)" else 30
         )
-        forecast_df = forecast_df.head(horizon if mode != "Backtest (Reality Check)" else 30)
         
-        # Signal Generation
+        forecast_values = point_forecast[0] # Get results for the first (only) input
+        
+        # Save CSV
+        csv_path = f"{ticker}_forecast.csv"
+        pd.DataFrame({'Forecast': forecast_values}).to_csv(csv_path, index=False)
+        
+        # Signals
         if mode != "Backtest (Reality Check)":
-            pct = ((forecast_df['timesfm'].iloc[-1] - train_df['Close'].iloc[-1]) / train_df['Close'].iloc[-1]) * 100
+            pct = ((forecast_values[-1] - train_df['Close'].iloc[-1]) / train_df['Close'].iloc[-1]) * 100
             signal = f"<h3 style='color: {'#00ff88' if pct > 0 else '#ff4444'};'>{ 'BULLISH' if pct > 0 else 'BEARISH' } ({pct:+.2f}%)</h3>"
         else:
-            acc = 100 - (abs(df['Close'].iloc[-1] - forecast_df['timesfm'].iloc[-1]) / df['Close'].iloc[-1] * 100)
-            signal = f"<h3 style='color: #FFD700;'>Model Confidence: {acc:.1f}%</h3>"
+            acc = 100 - (abs(df['Close'].iloc[-1] - forecast_values[-1]) / df['Close'].iloc[-1] * 100)
+            signal = f"<h3 style='color: #FFD700;'>Accuracy: {acc:.1f}%</h3>"
 
-        return get_financial_plot(df, forecast_df, ticker, (mode == "Backtest (Reality Check)")), signal
+        return get_financial_plot(df, forecast_values, ticker, (mode == "Backtest (Reality Check)")), signal, csv_path
         
     except Exception as e:
-        return None, f"Runtime Error: {str(e)}"
+        return None, f"Runtime Error: {str(e)}", None
 
-# --- UI Layout ---
+# --- UI ---
 with gr.Blocks(theme=gr.themes.Default(), css=".gradio-container {background-color: #000000}") as demo:
-    gr.HTML("<h2 style='color: #2962FF; text-align: center;'>G-TIMES JAX TERMINAL</h2>")
+    gr.HTML("<h2 style='color: #2962FF; text-align: center;'>G-TIMES 2.5 QUANT TERMINAL</h2>")
     with gr.Row():
         with gr.Column(scale=1):
             ticker_in = gr.Textbox(label="SYMBOL", value="NVDA")
             mode_in = gr.Radio(["Future Forecast", "Backtest (Reality Check)"], label="MODE", value="Future Forecast")
             days_in = gr.Slider(5, 128, value=30, label="Days")
-            btn = gr.Button("RUN JAX INFERENCE", variant="primary")
+            btn = gr.Button("RUN ANALYSIS", variant="primary")
             result_box = gr.HTML()
+            file_output = gr.File(label="Download Forecast")
         with gr.Column(scale=4):
             plot_out = gr.Plot()
-
-    btn.click(run_analysis, [ticker_in, days_in, mode_in], [plot_out, result_box])
+    btn.click(run_analysis, [ticker_in, days_in, mode_in], [plot_out, result_box, file_output])
 
 demo.launch()