train works
This commit is contained in:
Dobromir Popov
@ -209,11 +209,20 @@ class DataInterface:
|
||||
price_changes = (next_close - curr_close) / curr_close
|
||||
|
||||
# Define thresholds for price movement classification
|
||||
threshold = 0.001 # 0.1% threshold
|
||||
threshold = 0.0005 # 0.05% threshold - smaller to encourage more signals
|
||||
y = np.zeros(len(price_changes), dtype=int)
|
||||
y[price_changes > threshold] = 2 # Up
|
||||
y[price_changes < -threshold] = 0 # Down
|
||||
y[(price_changes >= -threshold) & (price_changes <= threshold)] = 1 # Neutral
|
||||
|
||||
# Log the target distribution to understand our data better
|
||||
sell_count = np.sum(y == 0)
|
||||
hold_count = np.sum(y == 1)
|
||||
buy_count = np.sum(y == 2)
|
||||
total_count = len(y)
|
||||
logger.info(f"Target distribution for {self.symbol} {self.timeframes[0]}: SELL: {sell_count} ({sell_count/total_count:.2%}), " +
|
||||
f"HOLD: {hold_count} ({hold_count/total_count:.2%}), BUY: {buy_count} ({buy_count/total_count:.2%})")
|
||||
|
||||
logger.info(f"Created features - X shape: {X.shape}, y shape: {y.shape}")
|
||||
return X, y, timestamps[window_size:]
|
||||
|
||||
@ -295,73 +304,107 @@ class DataInterface:
|
||||
|
||||
def calculate_pnl(self, predictions, actual_prices, position_size=1.0):
|
||||
"""
|
||||
Calculate PnL and win rates based on predictions and actual price movements.
|
||||
Robust PnL calculator that handles:
|
||||
- Action predictions (0=SELL, 1=HOLD, 2=BUY)
|
||||
- Probability predictions (array of [sell_prob, hold_prob, buy_prob])
|
||||
- Single price array or OHLC data
|
||||
|
||||
Args:
|
||||
predictions: Array of predicted actions (0=SELL, 1=HOLD, 2=BUY) or probabilities
|
||||
actual_prices: Array of actual close prices
|
||||
position_size: Position size for each trade
|
||||
predictions: Array of predicted actions or probabilities
|
||||
actual_prices: Array of actual prices (can be 1D or 2D OHLC format)
|
||||
position_size: Position size multiplier
|
||||
|
||||
Returns:
|
||||
tuple: (pnl, win_rate, trades) where:
|
||||
pnl is the total profit and loss
|
||||
win_rate is the ratio of winning trades
|
||||
trades is a list of trade dictionaries
|
||||
tuple: (total_pnl, win_rate, trades)
|
||||
"""
|
||||
# Ensure we have enough prices for the predictions
|
||||
if len(actual_prices) <= 1:
|
||||
logger.error("Not enough price data for PnL calculation")
|
||||
# Convert inputs to numpy arrays if they aren't already
|
||||
try:
|
||||
predictions = np.array(predictions)
|
||||
actual_prices = np.array(actual_prices)
|
||||
except Exception as e:
|
||||
logger.error(f"Error converting inputs: {str(e)}")
|
||||
return 0.0, 0.0, []
|
||||
|
||||
# Validate input shapes
|
||||
if len(predictions.shape) > 2 or len(actual_prices.shape) > 2:
|
||||
logger.error("Invalid input dimensions")
|
||||
return 0.0, 0.0, []
|
||||
|
||||
# Convert OHLC data to close prices if needed
|
||||
if len(actual_prices.shape) == 2 and actual_prices.shape[1] >= 4:
|
||||
prices = actual_prices[:, 3] # Use close prices
|
||||
else:
|
||||
prices = actual_prices
|
||||
|
||||
# Adjust predictions length to match available price data
|
||||
n_prices = len(actual_prices) - 1 # We need current and next price for each prediction
|
||||
if len(predictions) > n_prices:
|
||||
predictions = predictions[:n_prices]
|
||||
elif len(predictions) < n_prices:
|
||||
n_prices = len(predictions)
|
||||
actual_prices = actual_prices[:n_prices + 1] # +1 to include the next price
|
||||
# Handle case where prices is 2D with single column
|
||||
if len(prices.shape) == 2 and prices.shape[1] == 1:
|
||||
prices = prices.flatten()
|
||||
|
||||
# Convert probabilities to actions if needed
|
||||
if len(predictions.shape) == 2 and predictions.shape[1] > 1:
|
||||
actions = np.argmax(predictions, axis=1)
|
||||
else:
|
||||
actions = predictions
|
||||
|
||||
pnl = 0.0
|
||||
trades = 0
|
||||
wins = 0
|
||||
trade_history = []
|
||||
# Ensure we have enough prices
|
||||
if len(prices) < 2:
|
||||
logger.error("Not enough price data")
|
||||
return 0.0, 0.0, []
|
||||
|
||||
# Trim to matching length
|
||||
min_length = min(len(actions), len(prices)-1)
|
||||
actions = actions[:min_length]
|
||||
prices = prices[:min_length+1]
|
||||
|
||||
for i in range(len(predictions)):
|
||||
pred = predictions[i]
|
||||
current_price = actual_prices[i]
|
||||
next_price = actual_prices[i + 1]
|
||||
|
||||
# Calculate price change percentage
|
||||
pnl = 0.0
|
||||
wins = 0
|
||||
trades = []
|
||||
|
||||
for i in range(min_length):
|
||||
current_price = prices[i]
|
||||
next_price = prices[i+1]
|
||||
action = actions[i]
|
||||
|
||||
# Skip HOLD actions
|
||||
if action == 1:
|
||||
continue
|
||||
|
||||
price_change = (next_price - current_price) / current_price
|
||||
|
||||
# Calculate PnL based on prediction
|
||||
if pred == 2: # Buy
|
||||
if action == 2: # BUY
|
||||
trade_pnl = price_change * position_size
|
||||
trades += 1
|
||||
if trade_pnl > 0:
|
||||
wins += 1
|
||||
trade_history.append({
|
||||
'type': 'buy',
|
||||
'price': current_price,
|
||||
'pnl': trade_pnl,
|
||||
'timestamp': self.dataframes[self.timeframes[0]]['timestamp'].iloc[i] if self.dataframes[self.timeframes[0]] is not None else None
|
||||
})
|
||||
elif pred == 0: # Sell
|
||||
trade_type = 'BUY'
|
||||
is_win = price_change > 0
|
||||
elif action == 0: # SELL
|
||||
trade_pnl = -price_change * position_size
|
||||
trades += 1
|
||||
if trade_pnl > 0:
|
||||
wins += 1
|
||||
trade_history.append({
|
||||
'type': 'sell',
|
||||
'price': current_price,
|
||||
'pnl': trade_pnl,
|
||||
'timestamp': self.dataframes[self.timeframes[0]]['timestamp'].iloc[i] if self.dataframes[self.timeframes[0]] is not None else None
|
||||
})
|
||||
trade_type = 'SELL'
|
||||
is_win = price_change < 0
|
||||
else:
|
||||
continue # Invalid action
|
||||
|
||||
pnl += trade_pnl
|
||||
wins += int(is_win)
|
||||
|
||||
pnl += trade_pnl if pred in [0, 2] else 0
|
||||
|
||||
win_rate = wins / trades if trades > 0 else 0.0
|
||||
return pnl, win_rate, trade_history
|
||||
# Track trade details
|
||||
trades.append({
|
||||
'type': trade_type,
|
||||
'entry': current_price,
|
||||
'exit': next_price,
|
||||
'pnl': trade_pnl,
|
||||
'win': is_win,
|
||||
'duration': 1 # In number of candles
|
||||
})
|
||||
|
||||
win_rate = wins / len(trades) if trades else 0.0
|
||||
|
||||
# Add timestamps to trades if available
|
||||
if hasattr(self, 'dataframes') and self.timeframes and self.timeframes[0] in self.dataframes:
|
||||
df = self.dataframes[self.timeframes[0]]
|
||||
if df is not None and 'timestamp' in df.columns:
|
||||
for i, trade in enumerate(trades[:len(df)]):
|
||||
trade['timestamp'] = df['timestamp'].iloc[i]
|
||||
|
||||
return pnl, win_rate, trades
|
||||
|
||||
def get_future_prices(self, prices, n_candles=3):
|
||||
"""
|
||||
|
||||
Reference in New Issue
Block a user