Update app.py

app.py

@@ -1,7 +1,7 @@
 # app.py — DataSynthis_ML_JobTask
 # Evaluate pretrained ARIMA + LSTM on a user-picked historical window inside 2020-10-03..2025-10-03
 
-import os, math, json, pickle, traceback
+import os, math, json, pickle, traceback, warnings
 from typing import Tuple
 import numpy as np
 import pandas as pd
@@ -15,6 +15,9 @@ import matplotlib.pyplot as plt
 from statsmodels.tsa.arima.model import ARIMA
 from sklearn.metrics import mean_squared_error, mean_absolute_error, mean_absolute_percentage_error
 
+# Silence pandas/np deprecation spam in logs
+warnings.simplefilter("ignore", FutureWarning)
+
 # -------------------------
 # Historical universe bounds (DO NOT change)
 # -------------------------
@@ -37,6 +40,14 @@ _LSTM_META  = None
 
 POPULAR = ["AAPL","MSFT","GOOGL","AMZN","TSLA","NVDA","META","NFLX","IBM","ORCL"]
 
+# =========================
+# Small helpers
+# =========================
+def _as_scalar(x) -> float:
+    """Robustly turn a pandas/np value into a plain float without FutureWarnings."""
+    arr = np.asarray(x).reshape(-1,)
+    return float(arr[0])
+
 # =========================
 # Data utilities
 # =========================
@@ -67,7 +78,6 @@ def build_train_test_for_window(df: pd.DataFrame, start_sel: date, end_sel: date
     Train = all history strictly before start_sel (business days).
     Test  = start_sel..end_sel inclusive (business days).
     """
-    # align selection to universe
     start_sel = clip_to_universe(start_sel)
     end_sel   = clip_to_universe(end_sel)
     if start_sel > end_sel:
@@ -111,9 +121,8 @@ def arima_roll_eval(train_s: pd.Series, test_s: pd.Series, order=(2,1,2)):
     preds, trues = [], []
     for i in range(len(test_s)):
         fcast = res.forecast(1)
-        fval = float(np.asarray(fcast).reshape(-1)[0])  # robust scalar
-        preds.append(fval)
-        trues.append(float(test_s.iloc[i]))
+        preds.append(_as_scalar(fcast))
+        trues.append(_as_scalar(test_s.iloc[i]))
         try:
             res = res.append(test_s.iloc[i:i+1])  # update state with realized value
         except Exception:
@@ -202,7 +211,7 @@ def lstm_roll_eval(train_s: pd.Series, test_s: pd.Series):
         yhat_scaled = model.predict(x, verbose=0).reshape(-1)[0]
         yhat = _minmax_unscale_1d([yhat_scaled], vmin, vmax, fr_low, fr_high)[0]
         preds.append(float(yhat))
-        trues.append(float(test_s.iloc[i]))
+        trues.append(_as_scalar(test_s.iloc[i]))
         # update history with the REAL observed value (fair evaluation)
         full_hist = pd.concat([full_hist, test_s.iloc[i:i+1]])
 
@@ -248,6 +257,7 @@ def run_evaluation(ticker_choice: str, start_sel_txt: str, end_sel_txt: str):
         e_sel = clip_to_universe(parse_date(end_sel_txt))
         if s_sel > e_sel:
             raise ValueError("Start date must be before or equal to end date.")
+
         # Fetch universe data and build window train/test
         df = fetch_universe_data(ticker)
         train_s, test_s = build_train_test_for_window(df, s_sel, e_sel)
@@ -287,11 +297,11 @@ def run_evaluation(ticker_choice: str, start_sel_txt: str, end_sel_txt: str):
         fig_lstm  = plot_series(pd.to_datetime(dates), actual, lstm_p,
                                 f"LSTM (lookback={lookback}) — Actual vs Prediction")
 
-        # KPI Markdown
+        # KPI Markdown (fixed format specifiers)
         kpi = (
             f"**Ticker:** {ticker}  |  **Window:** {s_sel} → {e_sel}  \n"
             f"**ARIMA {order}** — RMSE: {ar_tbl['RMSE'][0]:.2f} | MAE: {ar_tbl['MAE'][0]:.2f} | MAPE: {ar_tbl['MAPE (%)'][0]:.2f}%  \n"
-            f"**LSTM (lookback={lookback})** — RMSE: {lstm_tbl['RMSE'][0]:.2f} | MAE: {lstm_tbl['MAE'][0]:.2f} | MAPE: {lstm_tbl['MAPE (%)'][0]::.2f}%"
+            f"**LSTM (lookback={lookback})** — RMSE: {lstm_tbl['RMSE'][0]:.2f} | MAE: {lstm_tbl['MAE'][0]:.2f} | MAPE: {lstm_tbl['MAPE (%)'][0]:.2f}%"
         )
 
         return kpi, fig_arima, fig_lstm, table
@@ -332,6 +342,7 @@ with gr.Blocks(title="DataSynthis_ML_JobTask", theme=gr.themes.Soft()) as demo:
     run_btn = gr.Button("Run Evaluation", variant="primary")
 
     gr.Markdown("### Results")
+    kpi = gr.Mardown if hasattr(gr, "Mardown") else gr.Markdown  # just in case
     kpi = gr.Markdown(label="Model scores (lower = better)", value="")
     with gr.Row():
         chart_arima = gr.Plot(label="ARIMA — Actual vs Predicted")