Update app.py

app.py

@@ -199,31 +199,77 @@ def predict_and_plot(timeframe, limit, epsilon, n_steps, ma):
     return fig
 
 
-def predict_both_plots(limit, epsilon, ma):
-    period = f'{limit}d'
-    n_steps = limit
-    btc_data_1d = fetch_yfinance_data('BTC/USDT', period, interval='1d')
-    btc_data_4h = fetch_yfinance_data('BTC/USDT', period, interval='4h')
-    bch_data_1d = fetch_yfinance_data('BCH/USDT', period, interval='1d')
-    bch_data_4h = fetch_yfinance_data('BCH/USDT', period, interval='4h')
-
-    btc_data_1d, _ = normalize(btc_data_1d)
-    bch_data_1d, _ = normalize(bch_data_1d)
-    btc_data_4h, _ = normalize(btc_data_4h)
-    bch_data_4h, _ = normalize(bch_data_4h)
-
-    preds_1d, label_1d = predictions(model_n1d_cat, btc_data_1d, bch_data_1d, name='1 day', n_steps=n_steps)
-    preds_4h, label_4h = predictions(model_n4h_cat, btc_data_4h, bch_data_4h, name='4 hours', n_steps=n_steps)
-    fig_1d = plot(preds_1d, label=label_1d, timeframe='1d', ma=ma, n_steps=n_steps)
-    fig_4h = plot(preds_4h, label=label_4h, timeframe='4h', ma=ma, n_steps=n_steps)
-    return fig_1d, fig_4h
 
+def make_interactive_fig(y, label, timeframe='1h', ma=5, n_steps=None, line_width=2):
+    # align lengths
+    if n_steps is None or n_steps > len(label):
+        n_steps = len(label)
+    dates = label['timestamp'].iloc[-n_steps:]
+    real  = label['close'].iloc[-n_steps:]
+    preds = 5 * (y[-n_steps:] - 0.5)
+
+    # rolling means
+    real_ma = pd.Series(real.values).rolling(window=ma).mean()
+    pred_ma = pd.Series(preds).rolling(window=ma).mean()
+    diff_ma = (real_ma - pred_ma)
+
+    fig = go.Figure()
+    fig.add_trace(go.Scatter(
+        x=dates, y=pred_ma,
+        mode='lines', name='Predicted Δ',
+        line=dict(width=line_width)
+    ))
+    fig.add_trace(go.Scatter(
+        x=dates, y=real_ma,
+        mode='lines', name='Real Close',
+        line=dict(width=line_width)
+    ))
+    fig.add_trace(go.Scatter(
+        x=dates, y=diff_ma,
+        mode='lines', name='Difference',
+        line=dict(width=line_width)
+    ))
+    # horizontal zero line
+    fig.add_shape(
+        type='line', x0=dates.min(), x1=dates.max(),
+        y0=0, y1=0, line=dict(dash='dash', width=line_width)
+    )
+    fig.update_layout(
+        title=f"BTC {timeframe} Forecast vs. Real",
+        xaxis_title='Timestamp',
+        yaxis_title='Value',
+        hovermode='x unified'
+    )
+    return fig
+
+def predict_both_plots(limit, epsilon, n_steps, ma):
+    period = f'{limit}d'
+    # fetch & normalize both timeframes
+    btc_1d = fetch_yfinance_data('BTC/USDT', period, '1d')
+    bch_1d = fetch_yfinance_data('BCH/USDT', period, '1d')
+    btc_1d, _ = normalize(btc_1d)
+    bch_1d, _ = normalize(bch_1d)
+
+    btc_4h = fetch_yfinance_data('BTC/USDT', period, '4h')
+    bch_4h = fetch_yfinance_data('BCH/USDT', period, '4h')
+    btc_4h, _ = normalize(btc_4h)
+    bch_4h, _ = normalize(bch_4h)
+
+    # generate predictions
+    y1, lbl1 = predictions(model_n1d_cat, btc_1d, bch_1d, '1d', n_steps)
+    y2, lbl2 = predictions(model_n4h_cat, btc_4h, bch_4h, '4h', n_steps)
+
+    # build interactive figures
+    fig1 = make_interactive_fig(y1, lbl1, timeframe='1d', ma=ma, n_steps=n_steps)
+    fig2 = make_interactive_fig(y2, lbl2, timeframe='4h', ma=ma, n_steps=n_steps)
+
+    return fig1, fig2
+    
 with gr.Blocks() as demo:
     with gr.Row():
         with gr.Column(scale=1):
             limit   = gr.Slider(50,500,step=50,value=100,label='Number of points')
             epsilon = gr.Slider(0.1,5.0,step=0.1,value=2.0, label='Epsilon')
-            # n_steps = gr.Slider(50,500,step=50,value=200,label='N_steps')
             ma      = gr.Slider(1,20,step=1,value=5,   label='MA Window')
             run_btn = gr.Button("Run Prediction")