Create MLCryptoForecasterAllAssetsTPSL.py

MLCryptoForecasterAllAssetsTPSL.py

@@ -0,0 +1,124 @@
+import os
+import pandas as pd
+import numpy as np
+from datetime import timedelta
+from binance.client import Client
+from sklearn.model_selection import train_test_split
+from sklearn.ensemble import RandomForestClassifier
+from sklearn.metrics import classification_report
+import ta
+
+# Initialize Binance client (insert API keys if needed)
+client = Client()
+
+# Settings
+interval = Client.KLINE_INTERVAL_4HOUR
+symbols = [s['symbol'] for s in client.get_exchange_info()['symbols']
+           if s['status']=='TRADING' and s['quoteAsset']=='USDT']
+
+def process_symbol(symbol):
+    data_file = f"{symbol}_data_4h_full.csv"
+    # Load or download data
+    if os.path.exists(data_file):
+        df = pd.read_csv(data_file, index_col=0, parse_dates=True)
+        # Normalize volume column name
+        if 'volume' in df.columns:
+            df.rename(columns={'volume':'vol'}, inplace=True)
+        last_ts = df.index[-1]
+        start = (last_ts + timedelta(hours=4)).strftime("%d %B %Y %H:%M:%S")
+        new = client.get_historical_klines(symbol, interval, start)
+        if new:
+            new_df = pd.DataFrame(new, columns=['ts','open','high','low','close','vol',
+                                                 'close_time','quote_av','trades','tb_base_av','tb_quote_av','ignore'])
+            new_df = new_df[['ts','open','high','low','close','vol']].astype({k:float for k in ['open','high','low','close','vol']})
+            new_df['ts'] = pd.to_datetime(new_df['ts'], unit='ms')
+            new_df.set_index('ts', inplace=True)
+            df = pd.concat([df, new_df]).drop_duplicates()
+            df.to_csv(data_file)
+    else:
+        klines = client.get_historical_klines(symbol, interval, "01 Dec 2021")
+        df = pd.DataFrame(klines, columns=['ts','open','high','low','close','vol',
+                                           'close_time','quote_av','trades','tb_base_av','tb_quote_av','ignore'])
+        df = df[['ts','open','high','low','close','vol']].astype({k:float for k in ['open','high','low','close','vol']})
+        df['ts'] = pd.to_datetime(df['ts'], unit='ms')
+        df.set_index('ts', inplace=True)
+        df.to_csv(data_file)
+
+    # Standardize volume if still present as 'volume'
+    if 'volume' in df.columns:
+        df.rename(columns={'volume':'vol'}, inplace=True)
+
+    # Feature Engineering
+    df['rsi'] = ta.momentum.RSIIndicator(df['close'], window=14).rsi()
+    df['macd'] = ta.trend.MACD(df['close']).macd()
+    for s in [10, 20, 50, 100]:
+        df[f'ema_{s}'] = df['close'].ewm(span=s).mean()
+    for w in [10, 20, 50, 100]:
+        df[f'sma_{w}'] = df['close'].rolling(window=w).mean()
+    bb = ta.volatility.BollingerBands(df['close'], window=20, window_dev=2)
+    df['bbw'] = (bb.bollinger_hband() - bb.bollinger_lband()) / bb.bollinger_mavg()
+    df['atr'] = ta.volatility.AverageTrueRange(df['high'], df['low'], df['close'], window=14).average_true_range()
+    df['adx'] = ta.trend.ADXIndicator(df['high'], df['low'], df['close'], window=14).adx()
+    st = ta.momentum.StochasticOscillator(df['high'], df['low'], df['close'], window=14)
+    df['st_k'] = st.stoch()
+    df['st_d'] = st.stoch_signal()
+    df['wr'] = ta.momentum.WilliamsRIndicator(df['high'], df['low'], df['close'], lbp=14).williams_r()
+    df['cci'] = ta.trend.CCIIndicator(df['high'], df['low'], df['close'], window=20).cci()
+    df['mom'] = df['close'] - df['close'].shift(10)
+    ichi = ta.trend.IchimokuIndicator(df['high'], df['low'], window1=9, window2=26, window3=52)
+    df['span_a'] = ichi.ichimoku_a()
+    df['span_b'] = ichi.ichimoku_b()
+    df.dropna(inplace=True)
+
+    # Trend labels
+    df['signal'] = np.select(
+        [(df['close'] > df['span_a']) & (df['close'] > df['span_b']),
+         (df['close'] < df['span_a']) & (df['close'] < df['span_b'])],
+        [1, 0],
+        default=-1
+    )
+
+    # Train/Test
+    features = df.drop(columns=['open', 'high', 'low', 'close', 'vol', 'signal']).columns
+    X, y = df[features], df['signal']
+    Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.2, shuffle=False)
+    mdl = RandomForestClassifier(n_estimators=200, class_weight='balanced', random_state=42)
+    mdl.fit(Xtr, ytr)
+    yp = mdl.predict(Xte)
+    print(f"=== {symbol} ===")
+    print(classification_report(yte, yp, zero_division=0))
+
+    # Backtest for optimal TP/SL
+    def optimize_tp_sl(df, signals, side, pgrid, lgrid):
+        best = (0, 0, -np.inf)
+        prices = df['close'].values
+        idxs = np.where(signals == side)[0]
+        for tp in pgrid:
+            for sl in lgrid:
+                rets = []
+                for i in idxs:
+                    entry = prices[i]
+                    for j in range(i+1, min(i+11, len(prices))):
+                        ret = ((prices[j] - entry) / entry) if side == 1 else ((entry - prices[j]) / entry)
+                        if ret >= tp or ret <= -sl:
+                            rets.append(np.sign(ret) * min(abs(ret), max(tp, sl)))
+                            break
+                if rets:
+                    avg_ret = np.mean(rets)
+                    if avg_ret > best[2]:
+                        best = (tp, sl, avg_ret)
+        return best
+
+    hist_signals = pd.Series(mdl.predict(X), index=X.index)
+    pgrid = np.arange(0.01, 0.1, 0.01)
+    lgrid = np.arange(0.01, 0.1, 0.01)
+    up_tp, up_sl, _ = optimize_tp_sl(df, hist_signals.values, 1, pgrid, lgrid)
+    dn_tp, dn_sl, _ = optimize_tp_sl(df, hist_signals.values, 0, pgrid, lgrid)
+    print(f"Optimal UP TP/SL: +{up_tp*100:.1f}% / -{up_sl*100:.1f}%")
+    print(f"Optimal DN TP/SL: +{dn_tp*100:.1f}% / -{dn_sl*100:.1f}%")
+
+for sym in symbols:
+    try:
+        process_symbol(sym)
+    except Exception as e:
+        print(f"Error {sym}: {e}")