Update app.py

app.py

@@ -5,13 +5,13 @@ import plotly.graph_objects as go
 from datetime import datetime, timedelta
 import pickle
 import yfinance as yf
+import os
+import re
 from statsmodels.tsa.arima.model import ARIMA
 from prophet import Prophet
 from tensorflow import keras
 from sklearn.preprocessing import MinMaxScaler
 import warnings
-import os
-from datetime import datetime, timedelta
 warnings.filterwarnings('ignore')
 
 # Load your saved models (update paths as needed)
@@ -41,70 +41,87 @@ def load_models():
 arima_model, prophet_model, lstm_model, scaler = load_models()
 SEQ_LENGTH = 60  # Should match your training
 
-
 def fetch_stock_data(ticker, days=365):
-    """
-    Fetch stock data from a local CSV (handles multi-row headers)
-    or downloads from Yahoo Finance if not found.
-    Returns last `days` of data with standardized 'Price' column.
-    """
+    """Fetch stock data from Yahoo Finance"""
     try:
-        filename = f"{ticker.upper()}.csv"
-        file_path = os.path.join(os.getcwd(), filename)
-
-        # ✅ Load local dataset if it exists
-        if os.path.exists(file_path):
-            print(f"📂 Loading local dataset: {filename}")
+        # Prefer local CSV file named <TICKER>.csv in the project root
+        csv_name = f"{ticker.upper()}.csv"
+        workspace_dir = os.path.dirname(__file__)
+        csv_path = os.path.join(workspace_dir, csv_name)
 
-            # Try reading while skipping bad header rows
-            df = pd.read_csv(file_path, skiprows=2)  # skip first two rows ("Ticker", "Date" lines)
-            df.rename(columns=lambda c: c.strip().lower(), inplace=True)
+        if os.path.exists(csv_path):
+            # Read the CSV fully, then detect which column contains dates. Many of
+            # the CSVs here contain extra header/noise rows; reading everything and
+            # filtering non-date rows is more robust than skipping rows beforehand.
+            df_raw = pd.read_csv(csv_path, header=0, dtype=str)
 
-            if 'date' not in df.columns:
-                # Maybe pandas treated first column as unnamed
-                df.rename(columns={df.columns[0]: 'date'}, inplace=True)
+            # Try to detect a date column by checking which column's values look like dates
+            date_col = None
+            for col in df_raw.columns:
+                sample = df_raw[col].astype(str).head(20)
+                matches = sample.str.match(r"^\s*\d{4}-\d{2}-\d{2}")
+                if matches.sum() >= max(1, int(len(sample) * 0.5)):
+                    date_col = col
+                    break
 
-        else:
-            print(f"🌐 Downloading {ticker} data from Yahoo Finance...")
-            end_date = datetime.now()
-            start_date = end_date - timedelta(days=days * 2)
-            df = yf.download(ticker, start=start_date, end=end_date, progress=False)
+            if date_col is None and 'Date' in df_raw.columns:
+                date_col = 'Date'
 
-            if df.empty:
-                return None, f"No data found for ticker: {ticker}"
-
-            df.to_csv(file_path)
-            print(f"💾 Saved downloaded dataset to: {file_path}")
-            df.reset_index(inplace=True)
-            df.rename(columns=lambda c: c.strip().lower(), inplace=True)
-
-        # ✅ Identify columns
-        date_col = next((c for c in df.columns if 'date' in c or 'time' in c), None)
-        close_col = next((c for c in df.columns if 'close' in c), None)
-        price_col = 'price' if 'price' in df.columns else close_col
-
-        if not date_col or not price_col:
-            return None, f"Could not detect date or price/close column in {filename}"
+            if date_col is not None:
+                df_raw[date_col] = pd.to_datetime(df_raw[date_col], errors='coerce')
+                df = df_raw.dropna(subset=[date_col]).copy()
+                df.set_index(date_col, inplace=True)
+            else:
+                # Try parsing the index as dates (if CSV had implicit index)
+                try:
+                    df_raw.index = pd.to_datetime(df_raw.index)
+                    df = df_raw.copy()
+                except Exception:
+                    # Give up and use raw DataFrame — downstream checks will catch issues
+                    df = df_raw.copy()
+            
 
-        # ✅ Clean up and format
-        df[date_col] = pd.to_datetime(df[date_col], errors='coerce')
-        df.dropna(subset=[date_col, price_col], inplace=True)
-        df.set_index(date_col, inplace=True)
-        df.sort_index(inplace=True)
+            # Prefer 'Close' column, fall back to common alternatives
+            if 'Close' in df.columns:
+                df = df[['Close']].copy()
+            elif 'Adj Close' in df.columns:
+                df = df[['Adj Close']].copy()
+                df.columns = ['Close']
+            elif 'Close*' in df.columns:
+                df = df[['Close*']].copy()
+                df.columns = ['Close']
+            else:
+                # Try to find a column that looks like price
+                possible = [c for c in df.columns if 'close' in c.lower() or 'price' in c.lower()]
+                if possible:
+                    df = df[[possible[0]]].copy()
+                    df.columns = ['Close']
+                else:
+                    return None, f"Local CSV found but no 'Close' column in {csv_name}"
 
-        # ✅ Keep last `days` of data
-        end_date = df.index.max()
-        start_date = end_date - timedelta(days=days)
-        df = df.loc[df.index >= start_date]
+            # Coerce to numeric price and drop rows that can't be converted
+            df.columns = ['Price']
+            df['Price'] = pd.to_numeric(df['Price'], errors='coerce')
+            df.dropna(subset=['Price'], inplace=True)
+            # Ensure sorted by date
+            df.sort_index(inplace=True)
+            # Remove index name to avoid printing a duplicate label
+            try:
+                df.index.name = None
+            except Exception:
+                pass
 
-        # ✅ Final column cleanup
-        df = df[[price_col]].copy()
-        df.columns = ['Price']
+            # Slice to the requested window (last `days` days)
+            if days is not None and days > 0:
+                start_dt = df.index.max() - timedelta(days=days - 1)
+                df = df.loc[df.index >= start_dt]
 
-        return df, None
+            if df.empty:
+                return None, f"No data in local CSV for the requested period: {csv_name}"
 
+            return df, None
     except Exception as e:
-        return None, str(e)
+        return None, f"Error fetching stock data: {e}"
 
 def make_arima_forecast(data, days):
     """Make ARIMA forecast"""