MLCryptoForecasterAllAssetsTPSL_ParisTime.py

import os
import pandas as pd
import numpy as np
import argparse
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
import pytz

# Parse command-line arguments for timeframe
parser = argparse.ArgumentParser(description="Binance Trend Forecaster with adjustable timeframe")
parser.add_argument("--interval", type=str, default="4h",
                    choices=["1m","3m","5m","15m","30m","1h","4h","1d"],
                    help="Time interval for klines (e.g. '1h', '4h', '1d')")
args = parser.parse_args()

# Map user-friendly intervals to Binance API constants
interval_map = {
    "1m": Client.KLINE_INTERVAL_1MINUTE,
    "3m": Client.KLINE_INTERVAL_3MINUTE,
    "5m": Client.KLINE_INTERVAL_5MINUTE,
    "15m": Client.KLINE_INTERVAL_15MINUTE,
    "30m": Client.KLINE_INTERVAL_30MINUTE,
    "1h": Client.KLINE_INTERVAL_1HOUR,
    "4h": Client.KLINE_INTERVAL_4HOUR,
    "1d": Client.KLINE_INTERVAL_1DAY
}
interval = interval_map[args.interval]

# Function to log results to both console and file
# Blank lines are added after each asset block explicitly
def log_results(message, filename="predictions_results.txt"):
    print(message)
    with open(filename, "a") as f:
        f.write(message + "\n")

# Convert UTC timestamp to Europe/Paris timezone
def convert_to_paris_time(utc_time):
    paris_tz = pytz.timezone('Europe/Paris')
    utc_time = utc_time.replace(tzinfo=pytz.utc)
    paris_time = utc_time.astimezone(paris_tz)
    return paris_time.strftime('%Y-%m-%d %H:%M:%S')

# Initialize Binance client
client = Client()

# Settings
result_file = f"predictions_results_{args.interval}.txt"

# Delete the results file if it exists for a fresh start
if os.path.exists(result_file):
    os.remove(result_file)

# Initialize result file header
with open(result_file, "w") as f:
    f.write("Asset,Time,Price,Prediction,Optimal_UP_TP,Optimal_UP_SL,Optimal_DN_TP,Optimal_DN_SL\n")

# Get USDT-quoted trading symbols
symbols = [s['symbol'] for s in client.get_exchange_info()['symbols']
           if s['status']=='TRADING' and s['quoteAsset']=='USDT']

# Optimize take-profit / stop-loss function
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

# Main loop: process each symbol
for symbol in symbols:
    log_results(f"=== {symbol} ({args.interval}) ===", result_file)

    # Load or download historical data
    data_file = f"{symbol}_data_{args.interval}_full.csv"
    if os.path.exists(data_file):
        df = pd.read_csv(data_file, index_col=0, parse_dates=True)
        last_ts = df.index[-1]
        start = (last_ts + timedelta(**{
            'minutes':1 if args.interval=='1m' else 3 if args.interval=='3m' else 5 if args.interval=='5m' else 15 if args.interval=='15m' else 30 if args.interval=='30m' else 60 if args.interval=='1h' else 240 if args.interval=='4h' else 1440
        })).strftime("%d %B %Y %H:%M:%S")
        new = client.get_historical_klines(symbol, interval, start)
        if new:
            new_df = pd.DataFrame(new, columns=[
                'timestamp','open','high','low','close','volume',
                'close_time','quote_av','trades','tb_base_av','tb_quote_av','ignore'
            ])
            new_df = new_df[['timestamp','open','high','low','close','volume']].astype(float)
            new_df['timestamp'] = pd.to_datetime(new_df['timestamp'], unit='ms')
            new_df.set_index('timestamp', inplace=True)
            df = pd.concat([df, new_df]).drop_duplicates()
            df.to_csv(data_file)
    else:
        klines = client.get_historical_klines(symbol, interval, "01 December 2021")
        df = pd.DataFrame(klines, columns=[
            'timestamp','open','high','low','close','volume',
            'close_time','quote_av','trades','tb_base_av','tb_quote_av','ignore'
        ])
        df = df[['timestamp','open','high','low','close','volume']].astype(float)
        df['timestamp'] = pd.to_datetime(df['timestamp'], unit='ms')
        df.set_index('timestamp', inplace=True)
        df.to_csv(data_file)

    # Compute technical indicators
    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)

    # Label signals based on Ichimoku cloud
    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 split
    features = [c for c in df.columns if c not in ['open','high','low','close','volume','signal']]
    X, y = df[features], df['signal']
    Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.2, shuffle=False)
    model = RandomForestClassifier(n_estimators=200, class_weight='balanced', random_state=42)
    model.fit(Xtr, ytr)
    ypr = model.predict(Xte)

    # Log classification report
    report = classification_report(yte, ypr, zero_division=0)
    log_results(f"Classification report for {symbol}:\n{report}", result_file)

    # Predict latest trend with correct feature naming
    latest_df = X.iloc[-1:]
    trend_label = model.predict(latest_df)[0]

    # Convert timestamp to Paris time and fetch price
    pred_time_utc = df.index[-1]
    pred_time = convert_to_paris_time(pred_time_utc)
    pred_price = df['close'].iloc[-1]
    trend_str = {1:'Uptrend',0:'Downtrend',-1:'Neutral'}[trend_label]
    log_results(f"Time: {pred_time}, Price: {pred_price:.2f}, Prediction: {trend_str}", result_file)

    # Optimize TP/SL and log results
    hist_sign = model.predict(X)
    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_sign, 1, pgrid, lgrid)
    dn_tp, dn_sl, _ = optimize_tp_sl(df, hist_sign, 0, pgrid, lgrid)
    log_results(f"Optimal UP TP/SL: +{up_tp*100:.1f}% / -{up_sl*100:.1f}%", result_file)
    log_results(f"Optimal DN TP/SL: +{dn_tp*100:.1f}% / -{dn_sl*100:.1f}%", result_file)

    # Blank line after asset
    with open(result_file, "a") as f:
        f.write("\n")

# End of processing
log_results("All assets processed.", result_file)