Delete backtest_lstm.py
Browse files- backtest_lstm.py +0 -93
backtest_lstm.py
DELETED
|
@@ -1,93 +0,0 @@
|
|
| 1 |
-
|
| 2 |
-
import os
|
| 3 |
-
import sys
|
| 4 |
-
import pandas as pd
|
| 5 |
-
import numpy as np
|
| 6 |
-
import random
|
| 7 |
-
from datetime import datetime, timedelta
|
| 8 |
-
# Add backend directory to sys.path for imports
|
| 9 |
-
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
| 10 |
-
from stock_data import fetch_stock_data, add_technical_indicators
|
| 11 |
-
from model import train_and_predict
|
| 12 |
-
|
| 13 |
-
def evaluate_prediction_accuracy(ticker, start_date, end_date, lookahead_days=5):
|
| 14 |
-
"""
|
| 15 |
-
For a given ticker and date range, run the LSTM model as if in the past, then compare its 5-day forecast to actual closes.
|
| 16 |
-
Returns a DataFrame with prediction, actual, and error for each window.
|
| 17 |
-
"""
|
| 18 |
-
df = fetch_stock_data(ticker)
|
| 19 |
-
df = add_technical_indicators(df)
|
| 20 |
-
df = df[(df['Date'] >= start_date) & (df['Date'] <= end_date)].reset_index(drop=True)
|
| 21 |
-
results = []
|
| 22 |
-
for i in range(len(df) - 60 - lookahead_days):
|
| 23 |
-
window = df.iloc[i:i+60+lookahead_days]
|
| 24 |
-
hist = window.iloc[:60]
|
| 25 |
-
future = window.iloc[60:60+lookahead_days]
|
| 26 |
-
# Skip if not enough future data or if hist is too short after dropna
|
| 27 |
-
if len(future) < lookahead_days or len(hist.dropna()) < 60:
|
| 28 |
-
continue
|
| 29 |
-
try:
|
| 30 |
-
pred = train_and_predict(hist)
|
| 31 |
-
pred_prices = pred['predicted_prices']
|
| 32 |
-
actual_prices = future['Close'].values.tolist()
|
| 33 |
-
errors = [abs(p - a) for p, a in zip(pred_prices, actual_prices)]
|
| 34 |
-
results.append({
|
| 35 |
-
'date': hist.iloc[-1]['Date'],
|
| 36 |
-
'predicted': pred_prices,
|
| 37 |
-
'actual': actual_prices,
|
| 38 |
-
'mae': np.mean(errors),
|
| 39 |
-
'mse': np.mean([(p-a)**2 for p,a in zip(pred_prices, actual_prices)]),
|
| 40 |
-
})
|
| 41 |
-
except Exception as e:
|
| 42 |
-
print(f"Skipping window at {hist.iloc[-1]['Date']} due to error: {e}")
|
| 43 |
-
continue
|
| 44 |
-
return pd.DataFrame(results)
|
| 45 |
-
|
| 46 |
-
if __name__ == "__main__":
|
| 47 |
-
ticker = input("Enter IPO stock ticker (e.g. RELIANCE.NS): ").strip()
|
| 48 |
-
# Fetch all available data for the ticker (max 2 years)
|
| 49 |
-
df_all = fetch_stock_data(ticker, period="2y", interval="1d")
|
| 50 |
-
df_all = add_technical_indicators(df_all)
|
| 51 |
-
if len(df_all) < 70:
|
| 52 |
-
print("Not enough data for backtest (need at least 70 days). Exiting.")
|
| 53 |
-
sys.exit(1)
|
| 54 |
-
# Find all valid window start indices (where 60-row history and 5-row future have no NaNs)
|
| 55 |
-
valid_starts = []
|
| 56 |
-
for i in range(len(df_all) - 65):
|
| 57 |
-
hist = df_all.iloc[i:i+60]
|
| 58 |
-
future = df_all.iloc[i+60:i+65]
|
| 59 |
-
if len(hist.dropna()) == 60 and len(future.dropna()) == 5:
|
| 60 |
-
valid_starts.append(i)
|
| 61 |
-
if not valid_starts:
|
| 62 |
-
print("No valid windows found in the last 2 years for this stock.")
|
| 63 |
-
sys.exit(1)
|
| 64 |
-
# Pick a random valid window
|
| 65 |
-
start_idx = random.choice(valid_starts)
|
| 66 |
-
# Pick a random end index at least 65 days ahead, but not more than 120 days or end of data
|
| 67 |
-
max_end = min(start_idx + 120, len(df_all) - 1)
|
| 68 |
-
end_idx = random.randint(start_idx + 65, max_end)
|
| 69 |
-
start_date = str(df_all.iloc[start_idx]['Date'])[:10]
|
| 70 |
-
end_date = str(df_all.iloc[end_idx]['Date'])[:10]
|
| 71 |
-
print(f"Randomly selected date range: {start_date} to {end_date}")
|
| 72 |
-
df = evaluate_prediction_accuracy(ticker, start_date, end_date)
|
| 73 |
-
if df.empty:
|
| 74 |
-
print("No valid prediction windows found in this range.")
|
| 75 |
-
# Try to print the last available predicted and actual price if possible
|
| 76 |
-
# Use the last 60 rows for prediction and next 5 for actual
|
| 77 |
-
hist = df_all.iloc[start_idx:start_idx+60]
|
| 78 |
-
future = df_all.iloc[start_idx+60:start_idx+65]
|
| 79 |
-
if len(hist.dropna()) == 60 and len(future.dropna()) == 5:
|
| 80 |
-
try:
|
| 81 |
-
pred = train_and_predict(hist)
|
| 82 |
-
pred_prices = pred['predicted_prices']
|
| 83 |
-
actual_prices = future['Close'].values.tolist()
|
| 84 |
-
print(f"Predicted 5-day prices: {pred_prices}")
|
| 85 |
-
print(f"Actual 5-day prices: {actual_prices}")
|
| 86 |
-
except Exception as e:
|
| 87 |
-
print(f"Could not compute prediction for last window: {e}")
|
| 88 |
-
else:
|
| 89 |
-
print(df[['date', 'mae', 'mse']])
|
| 90 |
-
print("Average MAE:", df['mae'].mean())
|
| 91 |
-
print("Average MSE:", df['mse'].mean())
|
| 92 |
-
df.to_csv(f"trained_models/{ticker}_backtest_results.csv", index=False)
|
| 93 |
-
print(f"Results saved to trained_models/{ticker}_backtest_results.csv")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|