From dab0f74fc4501ef07084be71ac3a3c94fbe4647c Mon Sep 17 00:00:00 2001 From: Vittus Mikiassen Date: Sun, 21 Jun 2026 08:32:13 +0200 Subject: [PATCH] Add files via upload --- mt5-xau-lstm-ppo-bot.py | 1957 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 1957 insertions(+) create mode 100644 mt5-xau-lstm-ppo-bot.py diff --git a/mt5-xau-lstm-ppo-bot.py b/mt5-xau-lstm-ppo-bot.py new file mode 100644 index 0000000..7126156 --- /dev/null +++ b/mt5-xau-lstm-ppo-bot.py @@ -0,0 +1,1957 @@ +import pandas as pd +import numpy as np +import os +import pickle +from io import StringIO +import random +from collections import deque +from datetime import datetime, timedelta +import subprocess +import time +import argparse +import threading +import MetaTrader5 as mt5 +import torch +import torch.nn as nn +import torch.nn.functional as F +from torch.distributions import Categorical + +# ready_event = threading.Event() +# process_counter = 0 +# pd.set_option('future.no_silent_downcasting', True) + +ACTIONS = ['hold', 'long', 'short', 'close'] + +# capital = 800 + +def load_last_mb_xauusd(file_path="C:\\Users\\Vittus Mikiassen\\Desktop\\XAU_5m_data.csv", mb=6, delimiter=';', col_names=None): + file_size = os.path.getsize(file_path) + offset = max(file_size - mb * 1024 * 1024, 0) # start position + + with open(file_path, 'rb') as f: + # Seek to approximately 20 MB before EOF + f.seek(offset) + + # Read to the end of file from that offset + data = f.read().decode(errors='ignore') + + # If not at start of file, discard partial first line (incomplete) + if offset > 0: + data = data.split('\n', 1)[-1] + + df = pd.read_csv(StringIO(data), delimiter=delimiter, header=None, engine='python') + + #if col_names: + df.columns = ["Date", "Open", "High", "Low", "Close", "Volume"] + + # Convert columns if needed, e.g.: + df["Date"] = pd.to_datetime(df["Date"], format="%Y.%m.%d %H:%M", errors='coerce') + for col in ["Open", "High", "Low", "Close", "Volume"]: + df[col] = pd.to_numeric(df[col], errors='coerce') + df['Date'] = pd.to_datetime(df['Date']) + df.set_index('Date', inplace=True) + df = df[['Open', 'High', 'Low', 'Close', 'Volume']].copy() + + # df = df.resample('15min').agg({ + # 'Open': 'first', + # 'High': 'max', + # 'Low': 'min', + # 'Close': 'last' + # }).dropna() + + df = df.dropna() + + return df + +def ADX(df, period=14): + """ + Returns +DI, -DI and ADX using Wilder's smoothing. + Columns required: High, Low, Close + """ + high = df['High'] + low = df['Low'] + close = df['Close'] + + # --- directional movement ----------------------------------------- + # plus_dm = (high.diff() > low.diff()) * (high.diff()).clip(lower=0) + # minus_dm = (low.diff() > high.diff()) * (low.diff().abs()).clip(lower=0)̈́ + up = high.diff() + dn = -low.diff() + + plus_dm_array = np.where((up > dn) & (up > 0), up, 0.0) + minus_dm_array = np.where((dn > up) & (dn > 0), dn, 0.0) + + plus_dm = pd.Series(plus_dm_array, index=df.index) # ← wrap + minus_dm = pd.Series(minus_dm_array, index=df.index) # ← wrap + + # --- true range ---------------------------------------------------- + tr = pd.concat([ + (high - low), + (high - close.shift()).abs(), + (low - close.shift()).abs() + ], axis=1).max(axis=1) + + # --- Wilder smoothing --------------------------------------------- + atr = tr.ewm(alpha=1/period, adjust=False).mean() + plus_di = 100 * (plus_dm.ewm(alpha=1/period, adjust=False).mean() / atr) + minus_di = 100 * (minus_dm.ewm(alpha=1/period, adjust=False).mean() / atr) + + dx = 100 * (plus_di - minus_di).abs() / (plus_di + minus_di) + adx = dx.ewm(alpha=1/period, adjust=False).mean() + adx = round(adx , 2) + plus_di = round(plus_di, 2) + minus_di = round(minus_di, 2) + + return adx, plus_di, minus_di + +def STOCH(df, period=14, smooth_d=3): + """ + Returns %K and %D stochastic oscillator. + + Columns required: + High, Low, Close + """ + + high = df['High'] + low = df['Low'] + close = df['Close'] + + # --- highest high / lowest low ------------------------------------ + lowest_low = low.rolling(window=period).min() + highest_high = high.rolling(window=period).max() + + # --- %K ------------------------------------------------------------ + k = 100 * ((close - lowest_low) / (highest_high - lowest_low)) + + # --- %D (smoothed %K) --------------------------------------------- + d = k.rolling(window=smooth_d).mean() + + k = round(k, 2) + d = round(d, 2) + + return k, d + +def EMA(df, period): + return df['Close'].ewm(span=period, adjust=False).mean().round(2) + +def EQH(df, tolerance=1): + swing_high = ( + (df["High"] > df["High"].shift(1)) & + (df["High"] > df["High"].shift(-1)) + ) + + prev_swing_high = df["High"].where(swing_high).ffill().shift(1) + + return ( + swing_high & + (abs(df["High"] - prev_swing_high) <= tolerance) + ).astype(int) + +def EQL(df, tolerance=1): + swing_low = ( + (df["Low"] < df["Low"].shift(1)) & + (df["Low"] < df["Low"].shift(-1)) + ) + + prev_swing_low = df["Low"].where(swing_low).ffill().shift(1) + + return ( + swing_low & + (abs(df["Low"] - prev_swing_low) <= tolerance) + ).astype(int) + +def Indecision(df, threshold=0.2): + body = (df["Close"] - df["Open"]).abs() + candle_range = (df["High"] - df["Low"]).replace(0, 1e-9) + + return (body / candle_range < threshold).astype(int) + +def RejectionBlocks(df, wick_ratio=2.0): + body = (df["Close"] - df["Open"]).abs() + + upper = df["High"] - df[["Open", "Close"]].max(axis=1) + lower = df[["Open", "Close"]].min(axis=1) - df["Low"] + + bullish_rb = ( + (lower > body * wick_ratio) & + (lower > upper) + ).astype(int) + + bearish_rb = ( + (upper > body * wick_ratio) & + (upper > lower) + ).astype(int) + + return bullish_rb, bearish_rb + +def BullishOB(df, multiplier=1.5): + body = (df["Close"] - df["Open"]).abs() + next_body = body.shift(-1) + + bearish = df["Close"] < df["Open"] + next_bullish = df["Close"].shift(-1) > df["Open"].shift(-1) + + return (bearish & + next_bullish & + (next_body >= body * multiplier)).astype(int) + +def BearishOB(df, multiplier=1.5): + body = (df["Close"] - df["Open"]).abs() + next_body = body.shift(-1) + + bullish = df["Close"] > df["Open"] + next_bearish = df["Close"].shift(-1) < df["Open"].shift(-1) + + return (bullish & + next_bearish & + (next_body >= body * multiplier)).astype(int) + +def BullishFVG(df): + return (df["Low"].shift(-1) > df["High"].shift(1)).astype(int) + +def BearishFVG(df): + return (df["High"].shift(-1) < df["Low"].shift(1)).astype(int) + +def BullishMB(df, multiplier=1.5): + body = (df["Close"] - df["Open"]).abs() + + bearish = df["Close"] < df["Open"] + next_bullish = df["Close"].shift(-1) > df["Open"].shift(-1) + displacement = body.shift(-1) >= body * multiplier + + ob = bearish & next_bullish & displacement + + ob_high = df["High"].where(ob).ffill() + ob_low = df["Low"].where(ob).ffill() + + return ((df["Low"] <= ob_high) & + (df["High"] >= ob_low)).astype(int) + +def BearishMB(df, multiplier=1.5): + body = (df["Close"] - df["Open"]).abs() + + bullish = df["Close"] > df["Open"] + next_bearish = df["Close"].shift(-1) < df["Open"].shift(-1) + displacement = body.shift(-1) >= body * multiplier + + ob = bullish & next_bearish & displacement + + ob_high = df["High"].where(ob).ffill() + ob_low = df["Low"].where(ob).ffill() + + return ((df["Low"] <= ob_high) & + (df["High"] >= ob_low)).astype(int) + +def AsiaHigh(df): + # Asia session: 23:00-06:59 GMT + asia = (df.index.hour >= 1) | (df.index.hour < 9) + + # Trading day starts at 23:00 + trade_day = (df.index - pd.Timedelta(hours=24)).date + + asia_high = ( + df["High"] + .where(asia) + .groupby(trade_day) + .transform("max") + .ffill() + ) + + return asia_high + +def AsiaLow(df): + asia = (df.index.hour >= 1) | (df.index.hour < 9) + + trade_day = (df.index - pd.Timedelta(hours=24)).date + + asia_low = ( + df["Low"] + .where(asia) + .groupby(trade_day) + .transform("min") + .ffill() + ) + + return asia_low + +def PDH(df): + day = df.index.date + + daily_high = ( + df["High"] + .groupby(day) + .transform("max") + ) + + pdh = ( + daily_high + .groupby(day) + .first() + .shift(1) + .reindex(day) + .to_numpy() + ) + + return pdh + +def PDL(df): + day = df.index.date + + daily_low = ( + df["Low"] + .groupby(day) + .transform("min") + ) + + pdl = ( + daily_low + .groupby(day) + .first() + .shift(1) + .reindex(day) + .to_numpy() + ) + + return pdl + +def VWAP(df, atr_period=14, atr_multiplier=1.0): + + # print(type(df.index)) + # print(df.index.dtype) + # print(df.index[:5]) + + # -------------------------------------------------- + # Select volume column + # -------------------------------------------------- + if "Volume" in df.columns: + volume = df["Volume"] + elif "tick_volume" in df.columns: + volume = df["tick_volume"] + elif "real_volume" in df.columns: + volume = df["real_volume"] + else: + raise ValueError("No volume column found.") + + # -------------------------------------------------- + # VWAP + # -------------------------------------------------- + typical_price = ( + df["High"] + + df["Low"] + + df["Close"] + ) / 3 + + session = df.index.normalize() + + cum_tpv = (typical_price * volume).groupby(session).cumsum() + cum_volume = volume.groupby(session).cumsum() + + vwap = round(cum_tpv / cum_volume, 2) + + # -------------------------------------------------- + # Session VWAP Standard Deviation + # -------------------------------------------------- + + # Squared distance from VWAP + sq_diff = ((typical_price - vwap) ** 2) * volume + + # Cumulative weighted variance + cum_sq_diff = sq_diff.groupby(session).cumsum() + + variance = cum_sq_diff / cum_volume + stddev = variance.pow(0.5) + + upper = round(vwap + stddev * atr_multiplier, 2) + lower = round(vwap - stddev * atr_multiplier, 2) + + # -------------------------------------------------- + # Derived features + # -------------------------------------------------- + # dist = df["Close"] - vwap + + above = (df["Close"] > vwap).astype(int) + below = (df["Close"] < vwap).astype(int) + + above_upper = (df["Close"] > upper).astype(int) + below_lower = (df["Close"] < lower).astype(int) + + slope = vwap.diff() + + return ( + vwap, + upper, + lower, + # dist, + above, + below, + above_upper, + below_lower, + slope + ) + +def VolumeMA(df, period=14): + + # ----------------------------------------- + # Select volume column + # ----------------------------------------- + if "Volume" in df.columns: + volume = df["Volume"] + elif "tick_volume" in df.columns: + volume = df["tick_volume"] + elif "real_volume" in df.columns: + volume = df["real_volume"] + else: + raise ValueError("No volume column found.") + + return round(volume.rolling(period).mean(), 2) + +def BuyScore(df): + + return ( + (df["EMA7"] > df["EMA21"]).astype(int) * 2 + + (df["EMA_DIFF"] > 0).astype(int) * 1 + + (df["+di"] > df["-di"]).astype(int) * 2 + + (df["adx"] > 20).astype(int) * 1 + + (df["k"] > df["k_smooth"]).astype(int) * 1 + + df["bullish_ob"] * 2 + + df["bullish_mb"] * 1 + + df["bullish_fvg"] * 1 + + df["eql"] * 1 + + df["bullish_rb"] * 1 + + df["above_vwap"] * 1 + + df["vwap_above_upper"] * 1 - + df["below_vwap"] * 1 - + df["vwap_below_lower"] * 1 - + df["bearish_rb"] * 1 - + df["bearish_ob"] * 2 - + df["bearish_fvg"] * 1 - + df["eqh"] * 1 - + (df["k"] >= 80) * 1 + ) + +def SellScore(df): + + return ( + (df["EMA7"] < df["EMA21"]).astype(int) * 2 + + (df["EMA_DIFF"] < 0).astype(int) * 1 + + (df["-di"] > df["+di"]).astype(int) * 2 + + (df["adx"] > 20).astype(int) * 1 + + (df["k"] < df["k_smooth"]).astype(int) * 1 + + df["bearish_ob"] * 2 + + df["bearish_mb"] * 1 + + df["bearish_fvg"] * 1 + + df["eqh"] * 1 + + df["below_vwap"] * 1 + + df["vwap_below_lower"] * 1 - + df["above_vwap"] * 1 - + df["vwap_above_upper"] * 1 - + df["bearish_rb"] * 1 - + df["bullish_rb"] * 1 - + df["bullish_ob"] * 2 - + df["bullish_fvg"] * 1 - + df["eql"] * 1 - + (df["k"] <= 20) * 1 + ) + +def add_indicators(df): + df['adx'], df['+di'], df['-di'] = ADX(df) + + df['k'], df['k_smooth'] = STOCH(df) + + df['EMA7'] = EMA(df, 7) + df['EMA21'] = EMA(df, 21) + df['EMA_DIFF'] = df['EMA7'] - df['EMA21'] + + df["indecision"] = Indecision(df) + + df["bullish_ob"] = BullishOB(df) + df["bearish_ob"] = BearishOB(df) + + df["bullish_fvg"] = BullishFVG(df) + df["bearish_fvg"] = BearishFVG(df) + + df["eqh"] = EQH(df) + df["eql"] = EQL(df) + + df["bearish_mb"] = BearishMB(df) + df["bullish_mb"] = BullishMB(df) + + df["bullish_rb"], df["bearish_rb"] = RejectionBlocks(df) + + df["vwap"], df["vwap_upper"], df["vwap_lower"], df["above_vwap"], df["below_vwap"], df["vwap_above_upper"], df["vwap_below_lower"], df["vwap_slope"] = VWAP(df) + + df["sell_score"] = SellScore(df) + df["buy_score"] = BuyScore(df) + + df["volume_ma"] = VolumeMA(df) + + df["asia_high"] = AsiaHigh(df) + df["asia_low"] = AsiaLow(df) + + df["pdh"] = PDH(df) + df["pdl"] = PDL(df) + + df = df[["Open", "High", "Low", "Close", + "k", "k_smooth", "adx", "+di", "-di", "EMA7", "EMA21", "EMA_DIFF", + "indecision", "bullish_ob", "bearish_ob", "bullish_fvg", "bearish_fvg", "eqh", "eql", "bearish_mb", "bullish_mb", "bullish_rb", "bearish_rb", + "vwap", "vwap_upper", "vwap_lower", "above_vwap", "below_vwap", "vwap_above_upper", "vwap_below_lower", "vwap_slope", + "volume_ma", + "asia_high", "asia_low", + "pdh", "pdl", + "sell_score", "buy_score"]].copy() + # df = df[["Open", "High", "Low", "Close", "EMA_crossover", "macd_zone", "macd_line", "macd_signal", "macd_line_diff", "macd_signal_diff", "macd_line_slope", "macd_signal_line_slope" , "macd_osma", "macd_crossover", "bb_sma", "bb_upper", "bb_lower", "RSI_zone", "ADX_zone", "+DI_val", "-DI_val", "ATR", "order_block_type"]].copy() + + df.dropna(inplace=True) + # print(df.isna().sum()) + # ready_event.set() + return df + +class PPOLSTMNetwork(nn.Module): + def __init__(self, state_size=12, hidden_size=64, action_size=3): + super().__init__() + + self.lstm = nn.LSTM( + input_size=state_size, + hidden_size=hidden_size, + batch_first=True + ) + + self.policy = nn.Sequential( + nn.Linear(hidden_size, 64), + nn.ReLU(), + nn.Linear(64, action_size) + ) + + self.value = nn.Sequential( + nn.Linear(hidden_size, 64), + nn.ReLU(), + nn.Linear(64, 1) + ) + + # self.debug = False + + def forward(self, x): + """ + if self.debug: + print("x shape before lstm:", x.shape) + + if x.shape[1] == 0: + print("ERROR: zero sequence length") + print("x shape:", x.shape) + raise ValueError("Zero sequence length") + """ + + out, _ = self.lstm(x) + h = out[:, -1, :] + + logits = self.policy(h) + value = self.value(h).squeeze(-1) + + return logits, value + +class LSTMPPOAgent: + def __init__( + self, + state_size, + hidden_size, + action_size, + lr=3e-4, + gamma=0.95, + clip_ratio=0.2, + gae_lambda=0.95 + ): + self.state_size = state_size + self.hidden_size = hidden_size + self.action_size = action_size + + self.gamma = gamma + self.clip_ratio = clip_ratio + self.gae_lambda = gae_lambda + + self.train_epochs = 10 + self.batch_size = 64 + self.entropy_coef = 0.01 + self.value_coef = 0.5 + + self.device = torch.device( + "cuda" if torch.cuda.is_available() else "cpu" + ) + + self.model = PPOLSTMNetwork( + state_size, + hidden_size, + action_size + ).to(self.device) + + self.optimizer = torch.optim.Adam( + self.model.parameters(), + lr=lr + ) + + self.trajectory = [] + + def _state_tensor(self, state_seq): + return torch.tensor( + state_seq, + dtype=torch.float32, + device=self.device + ).unsqueeze(0) + + def select_action(self, state_seq, in_position=False, training=False): + + state = self._state_tensor(state_seq) + + # if training is False: + # print("state type:", type(state)) + # print("state shape:", state.shape if hasattr(state, "shape") else "no shape") + + with torch.no_grad(): + logits, value = self.model(state) + + logits = logits.squeeze(0) + + if in_position: + valid_actions = [0] + else: + valid_actions = [0, 1, 2] + + masked_logits = logits.clone() + + for i in range(self.action_size): + if i not in valid_actions: + masked_logits[i] = -1e9 + + probs = torch.softmax(masked_logits, dim=-1) + + dist = Categorical(probs) + + if training: + action = dist.sample() + else: + action = torch.argmax(probs) + + logprob = dist.log_prob(action) + + """ + print( + f"H={probs[0]:.2f} " + f"B={probs[1]:.2f} " + f"S={probs[2]:.2f}" + ) + """ + + return ( + int(action.item()), + float(logprob.item()), + float(value.item()) + ) + + def store_transition( + self, + state_seq, + action, + logprob, + value, + reward, + done + ): + self.trajectory.append( + ( + np.array(state_seq, dtype=np.float32), + action, + logprob, + value, + reward, + done + ) + ) + + def compute_gae(self, rewards, values, dones): + + advantages = [] + gae = 0 + + values = np.append(values, 0.0) + + for t in reversed(range(len(rewards))): + + delta = ( + rewards[t] + + self.gamma * values[t + 1] * (1 - dones[t]) + - values[t] + ) + + gae = ( + delta + + self.gamma + * self.gae_lambda + * (1 - dones[t]) + * gae + ) + + advantages.insert(0, gae) + + return np.array(advantages, dtype=np.float32) + + def train(self): + + if len(self.trajectory) < 32: + return + + states, actions, old_logprobs, values, rewards, dones = zip( + *self.trajectory + ) + + states = np.array(states, dtype=np.float32) + actions = np.array(actions) + old_logprobs = np.array(old_logprobs, dtype=np.float32) + values = np.array(values, dtype=np.float32) + rewards = np.array(rewards, dtype=np.float32) + dones = np.array(dones, dtype=np.float32) + + advantages = self.compute_gae( + rewards, + values, + dones + ) + + returns = advantages + values + + advantages = ( + advantages - advantages.mean() + ) / (advantages.std() + 1e-8) + + states = torch.tensor( + states, + dtype=torch.float32, + device=self.device + ) + + actions = torch.tensor( + actions, + dtype=torch.long, + device=self.device + ) + + old_logprobs = torch.tensor( + old_logprobs, + dtype=torch.float32, + device=self.device + ) + + returns = torch.tensor( + returns, + dtype=torch.float32, + device=self.device + ) + + advantages = torch.tensor( + advantages, + dtype=torch.float32, + device=self.device + ) + + n = len(states) + + for _ in range(self.train_epochs): + + idx = torch.randperm(n, device=self.device) + + for start in range(0, n, self.batch_size): + + batch_idx = idx[start:start+self.batch_size] + + b_states = states[batch_idx] + b_actions = actions[batch_idx] + b_old_logprobs = old_logprobs[batch_idx] + b_returns = returns[batch_idx] + b_advantages = advantages[batch_idx] + + logits, values_pred = self.model(b_states) + + dist = Categorical(logits=logits) + + new_logprobs = dist.log_prob( + b_actions + ) + + entropy = dist.entropy().mean() + + ratio = torch.exp( + new_logprobs - b_old_logprobs + ) + + surr1 = ratio * b_advantages + + surr2 = torch.clamp( + ratio, + 1 - self.clip_ratio, + 1 + self.clip_ratio + ) * b_advantages + + policy_loss = -torch.min( + surr1, + surr2 + ).mean() + + value_loss = F.mse_loss( + values_pred, + b_returns + ) + + loss = ( + policy_loss + + self.value_coef * value_loss + - self.entropy_coef * entropy + ) + + self.optimizer.zero_grad() + loss.backward() + + torch.nn.utils.clip_grad_norm_( + self.model.parameters(), + 1.0 + ) + + self.optimizer.step() + + self.trajectory.clear() + + def savecheckpoint(self, symbol): + + os.makedirs( + "LSTM-PPO-saves", + exist_ok=True + ) + + filename = ( + f"LSTM-PPO-saves/" + f"{datetime.now().strftime('%Y-%m-%d')}-" + f"{symbol}.checkpoint.pt" + ) + + torch.save( + { + "model": self.model.state_dict(), + "optimizer": self.optimizer.state_dict() + }, + filename + ) + + def loadcheckpoint(self, symbol): + + if not os.path.exists("LSTM-PPO-saves"): + return + + files = [ + os.path.join("LSTM-PPO-saves", f) + for f in os.listdir("LSTM-PPO-saves") + if f.endswith(".checkpoint.pt") + and symbol in f + ] + + if not files: + return + + latest = max(files, key=os.path.getmtime) + + checkpoint = torch.load( + latest, + map_location=self.device + ) + + self.model.load_state_dict( + checkpoint["model"] + ) + + if "optimizer" in checkpoint: + self.optimizer.load_state_dict( + checkpoint["optimizer"] + ) + + print(f"Loaded checkpoint: {latest}") + +class WinRateKNN: + def __init__(self, symbol, k=10): + self.k = k + self.symbol = symbol + self.states = [] + self.labels = [] # 1 = win, 0 = loss + self.model = None + + def add(self, state, is_win): + try: + state = np.array(state, dtype=np.float32).flatten() # Force all elements to float + except Exception as e: + # print("❌ Could not convert state to float:", state, "| Error:", e) + return + + if not np.all(np.isfinite(state)): + # print("⚠️ Skipping state with NaN or Inf:", state) + return + + self.states.append(state) + self.labels.append(1 if is_win else 0) + + if len(self.states) >= 100: + self._remove_redundant_neighbor() + # self.states.pop(0) + # self.labels.pop(0) + + if len(self.states) >= self.k: + self._fit() + + def _remove_redundant_neighbor(self): + if len(self.states) < 2: + return # Nothing to remove + + X = np.array(self.states) + + # Compute pairwise similarity (cosine, or use euclidean if you prefer) + sim_matrix = cosine_similarity(X) + + # Zero out diagonal (self-similarity) + np.fill_diagonal(sim_matrix, 0) + + # Compute average similarity for each row (how redundant each entry is) + redundancy_scores = sim_matrix.mean(axis=1) + + # Remove the most redundant (highest avg similarity) + idx_to_remove = np.argmax(redundancy_scores) + + del self.states[idx_to_remove] + del self.labels[idx_to_remove] + def _fit(self): + """ + Fit the KNN model with stored data. + """ + if len(self.states) < 1: + # print("⚠️ Not enough data to fit KNN.") + return + + # Safety check + k_neighbors = max(1, min(self.k, len(self.states))) + + self.model = NearestNeighbors(n_neighbors=k_neighbors, algorithm="kd_tree") + self.model.fit(self.states) + + def predict_win_rate(self, state_seq, k_near=5, k_far=5): + """ + Return the win rate based on k nearest neighbors of the input state. + """ + # if not self.model or len(self.states) < self.k: + if len(self.states) < 1000: + # return True # Not enough data + return 1 # Not enough data + + # Find the 100 nearest neighbors + distances, indices = self.model.kneighbors(state.reshape(1, -1), n_neighbors=50) + distances = distances[0] + indices = indices[0] + + # Split into nearest and farthest groups + nearest_idx = indices[:k_near] + nearest_dist = distances[:k_near] + + farthest_idx = indices[-k_far:] + farthest_dist = distances[-k_far:] + + # Combine indices and distances + combined_idx = np.concatenate([nearest_idx, farthest_idx]) + combined_dist = np.concatenate([nearest_dist, farthest_dist]) + + # Get win/loss labels for selected neighbors + selected_labels = np.array([self.labels[i] for i in combined_idx]) + + # Calculate weights (closer gets higher weight) + weights = 1 / (combined_dist + 1e-6) # Add epsilon to avoid div-by-zero + + # Normalize weights + weights /= weights.sum() + + # Compute weighted win rate + win_rate = np.dot(selected_labels, weights) + + return win_rate + def save(self): + """ + Save the KNN model to disk. + """ + path = f"LSTM-PPO-saves/{datetime.now().strftime('%Y-%m-%d')}-{self.symbol}.win_rate_knn.pkl" + os.makedirs(os.path.dirname(path), exist_ok=True) + with open(path, "wb") as f: + pickle.dump({ + "states": self.states, + "labels": self.labels, + "model": self.model + }, f) + + def load(self): + """ + Load the KNN model from disk. + """ + # path = f"LSTM-PPO-saves/win_rate_knn-{self.symbol}.pkl" + files = sorted(os.listdir("LSTM-PPO-saves")) + files = [f for f in files if f.endswith(".win_rate_knn.pkl") and self.symbol in f] + if not files: + print(f"[!] No checkpoint found for {self.symbol}") + return + + latest = os.path.join("LSTM-PPO-saves", files[-1]) + + try: + with open(latest, "rb") as f: + data = pickle.load(f) + self.states = data["states"] + self.labels = data["labels"] + self.model = data["model"] + # print(f"✅ Loaded WinRateKNN from {latest}") + except FileNotFoundError: + print(f"⚠️ No saved KNN found at {path}. Starting fresh.") + +def sharpe_ratio(returns, risk_free_rate=0.0): + mean_ret = np.mean(returns) + std_ret = np.std(returns) + if std_ret == 0: + return 0 + return (mean_ret - risk_free_rate) / std_ret + +def sortino_ratio(returns, risk_free_rate=0.0): + mean_ret = np.mean(returns) + # Downside deviation: only consider returns below risk-free rate, and their square differences + downside_diff = [(r - risk_free_rate)**2 for r in returns if r < risk_free_rate] + + if len(downside_diff) == 0: + return 0 # Or float('inf') if you'd rather signal perfect performance + + downside_std = np.sqrt(np.mean(downside_diff)) + + if downside_std == 0: + return 0 + + return (mean_ret - risk_free_rate) / downside_std + +def max_drawdown(returns): + + if len(returns) == 0: + return 0 + + equity = np.cumsum(returns) + + peak = equity[0] + max_dd = 0 + + for value in equity: + + peak = max(peak, value) + + dd = peak - value + + max_dd = max(max_dd, dd) + + return max_dd + +def train_bot(symbol="XAUUSD"): + + df = load_last_mb_xauusd() + df = add_indicators(df) + + SEQ_LEN = 12 * 8 + + save_count = 1440 + + FEATURES = [ + "Open", + "High", + "Low", + "Close", + "k", + "k_smooth", + "adx", + "+di", + "-di", + "EMA7", + "EMA21", + "EMA_DIFF", + "indecision", + "bullish_ob", + "bearish_ob", + "bullish_fvg", + "bearish_fvg", + "eqh", + "eql", + "bearish_mb", + "bullish_mb", + "bullish_rb", + "bearish_rb", + "vwap", + "vwap_upper", + "vwap_lower", + "above_vwap", + "below_vwap", + "vwap_above_upper", + "vwap_below_lower", + "vwap_slope", + "volume_ma", + "asia_high", + "asia_low", + "pdh", + "pdl", + "sell_score", + "buy_score" + ] + + agent = LSTMPPOAgent( + state_size=len(FEATURES), + hidden_size=64, + action_size=3 + ) + + # knn = WinRateKNN(symbol) + + """ + try: + agent.loadcheckpoint(symbol) + # knn.load() + print(f"[{symbol}] Loaded checkpoint") + except: + print(f"[{symbol}] Starting fresh") + """ + + save_counter = 0 + + in_position = False + position_type = None + + entry_price = 0 + sl_price = 0 + tp_price = 0 + + entry_price = 0 + sl_price = 0 + + tp1_price = 0 + + position_size = 0.0 + realized_reward = 0.0 + + tp1_hit = False + + tp1_sl_moved = False + + trade_returns = [] + + # STANDARD_SL_PIPS = 100 + RR_RATIO = 0.5 + SPREAD_AND_COMMISSION = 0.6 + + SL_PIPS = 40 + + PIP_VALUE = 0.1 + + SPREAD_AND_COMMISSION = 0 + + state_buffer = deque(maxlen=SEQ_LEN) + + training_start_2 = time.time() + training_start_3 = time.time() + + # preload sequence + for i in range(SEQ_LEN): + row = df.iloc[i][FEATURES].values.astype(np.float32) + state_buffer.append(row) + + for i in range(SEQ_LEN, len(df)): + + current = df.iloc[i] + + current_price = current["Close"] + high = current["High"] + low = current["Low"] + TP1_PIPS = round(SL_PIPS * RR_RATIO, 0) + + state = current[FEATURES].values.astype(np.float32) + + state_buffer.append(state) + + if len(state_buffer) < SEQ_LEN: + continue + + state_seq = np.array(state_buffer) + + # === Select action ============================================ + result = agent.select_action(state_seq, in_position, training=True) + + if result is None: + continue + + action, logprob, value = result + + if df["adx"].iloc[i] < 20: + action = 0 + + pnl = 0.0 + reward = 0.0 + done = False + + if action == 0 and in_position: + if position_type == "long": + reward = round((current_price - entry_price) * 10, 0) + + # ============================================================== + # OPEN LONG + # ============================================================== + + # if action == 1 and not in_position and df["+di"].iloc[i] > df["-di"].iloc[i] and df["EMA_DIFF"].iloc[i] > 0 and df["k"].iloc[i] < 80: + if action == 1 and not in_position and df["EMA7"].iloc[i] > df["EMA21"].iloc[i] and df["k"].iloc[i] < 80: + # if action == 1: + in_position = True + position_type = "long" + + entry_price = current_price + + sl_price = entry_price - (SL_PIPS * 0.1) + tp_price = entry_price + ( + SL_PIPS * RR_RATIO * 0.1 + ) + + entry_price = current_price + + sl_price = entry_price - (SL_PIPS * PIP_VALUE) + + tp1_price = entry_price + (TP1_PIPS * PIP_VALUE) + + position_size = 1.0 + realized_reward = 0.0 + pnl = 0.0 + reward = 0.0 + + tp1_hit = False + + # ============================================================== + # OPEN SHORT + # ============================================================== + + # elif action == 2 and not in_position and df["-di"].iloc[i] > df["+di"].iloc[i] and df["EMA_DIFF"].iloc[i] < 0 and df["k"].iloc[i] > 20: + # elif action == 2 and not in_position and df["buy_score"].iloc[i] < df["sell_score"].iloc[i]: + elif action == 2 and not in_position and df["EMA7"].iloc[i] < df["EMA21"].iloc[i] and df["k"].iloc[i] > 20: + # elif action == 2: + in_position = True + position_type = "short" + + entry_price = current_price + + sl_price = entry_price + (SL_PIPS * 0.1) + tp_price = entry_price - ( + SL_PIPS * RR_RATIO * 0.1 + ) + + entry_price = current_price + + sl_price = entry_price + (SL_PIPS * PIP_VALUE) + + tp1_price = entry_price - (TP1_PIPS * PIP_VALUE) + + position_size = 1.0 + realized_reward = 0.0 + reward = 0.0 + pnl = 0.0 + + tp1_hit = False + + # ============================================================== + # MANAGE POSITION + # ============================================================== + if not in_position: + done = False + + if in_position: + + trade_closed = False + + # ================================================== + # LONG + # ================================================== + + if position_type == "long": + + if not tp1_hit and high >= tp1_price: + + # realized_reward += SL_PIPS - SPREAD_AND_COMMISSION + pnl += SL_PIPS * RR_RATIO - SPREAD_AND_COMMISSION + reward += SL_PIPS * RR_RATIO - SPREAD_AND_COMMISSION + position_size -= 0.25 + + tp1_hit = True + + sl_price = entry_price + trade_closed = True + + # print(f"tp1 hit, +{SL_PIPS - SPREAD_AND_COMMISSION:.0f}") + + if not trade_closed and low <= sl_price: + + remaining_pips = ( + (sl_price - entry_price) + / PIP_VALUE + ) + + # realized_reward += ( + reward += ( + # remaining_pips * (position_size / 0.25) - SPREAD_AND_COMMISSION * (position_size / 0.25) + remaining_pips - SPREAD_AND_COMMISSION + ) + pnl += ( + remaining_pips * (position_size) - SPREAD_AND_COMMISSION * (position_size) + # remaining_pips - SPREAD_AND_COMMISSION * (position_size / 0.25) + ) + + # print(f"sl hit, (+){remaining_pips * (position_size / 0.25) - SPREAD_AND_COMMISSION * (position_size / 0.25):.0f}") + # print(f"remaining pips: {remaining_pips:.0f}, positions: {position_size / 0.25:.0f}") + # reward = realized_reward + # if sl_price < entry_price: + # realized_reward = (SL_PIPS * 2) * -1 + trade_closed = True + + # ================================================== + # SHORT + # ================================================== + + elif position_type == "short": + + if not tp1_hit and low <= tp1_price: + + # realized_reward += SL_PIPS * 2 - SPREAD_AND_COMMISSION + reward += SL_PIPS * RR_RATIO - SPREAD_AND_COMMISSION + pnl += SL_PIPS * RR_RATIO - SPREAD_AND_COMMISSION + # position_size -= 0.25 + + tp1_hit = True + + sl_price = entry_price + trade_closed = True + # print(f"tp1 hit, +{SL_PIPS - SPREAD_AND_COMMISSION:.0f}") + + if not trade_closed and high >= sl_price: + + remaining_pips = ( + (entry_price - sl_price) + / PIP_VALUE + ) + + # realized_reward += ( + reward += ( + remaining_pips - SPREAD_AND_COMMISSION + ) + pnl += ( + remaining_pips * (position_size) - SPREAD_AND_COMMISSION * (position_size) + ) + + # print(f"sl hit, (+){remaining_pips * (position_size / 0.25) - SPREAD_AND_COMMISSION * (position_size / 0.25):.0f}") + # print(f"remaining pips: {remaining_pips:.0f}, positions: {position_size / 0.25:.0f}") + # reward = realized_reward + # if sl_price > entry_price: + # realized_reward = (SL_PIPS * 2) * -1 + trade_closed = True + + if trade_closed: + + in_position = False + done = True + trade_returns.append(pnl) + + # ============================================================== + # STORE PPO TRANSITION + # ============================================================== + + agent.store_transition( + state_seq, + action, + logprob, + value, + reward, + done + ) + + save_counter += 1 + + # ============================================================== + # WEEKLY TRAINING + # ============================================================== + + if save_counter % save_count == 0: + # if len(agent.trajectory) >= 512: + + # knn._fit() + # knn.save() + + # ========================================================== + # WEEKLY STATS + # ========================================================== + + if len(trade_returns) > 5: + + wins = [r for r in trade_returns if r > 0] + losses = [r for r in trade_returns if r < 0] + + weekly_pnl = np.sum(trade_returns) + + winrate = ( + len(wins) / len(trade_returns) + if len(trade_returns) > 0 else 0 + ) + + mean_win = ( + np.mean(wins) + if len(wins) > 0 else 0 + ) + + mean_loss = ( + np.mean(losses) + if len(losses) > 0 else 0 + ) + + sharpe = sharpe_ratio(trade_returns) + sortino = sortino_ratio(trade_returns) + + gross_profit = sum(wins) + gross_loss = abs(sum(losses)) + profit_factor = ( + gross_profit / gross_loss + if gross_loss > 0 + else float("inf") + ) + max_dd = max_drawdown(trade_returns) + R_pnl = weekly_pnl / (SL_PIPS) + + print() + print("================================================") + print(f"[{symbol}] WEEKLY PPO TRAINING") + print("================================================") + print(f"Trades: {len(trade_returns)}") + print(f"Weekly PnL: {weekly_pnl:.0f} pips") + print(f"Winrate: {winrate*100:.2f}%") + print(f"Mean Win: {mean_win:.0f} pips") + print(f"Mean Loss: {mean_loss:.0f} pips") + print(f"Max DD: {max_dd/(SL_PIPS):.2f}R") + print(f"PF: {profit_factor:.2f}") + print(f"Weekly R PnL: {R_pnl:.2f}R") + print(f"Sharpe: {sharpe:.2f}") + print(f"Sortino: {sortino:.2f}") + print("================================================") + print() + + print( + f"[{symbol}] " + f"[INFO] " + f" Trained on data (Elapsed: {timedelta(seconds=int(time.time() - training_start_3))})" + ) + + trade_returns = [] + + training_start = time.time() + + print(f"[{symbol}] [INFO] Training PPO") + + agent.train() + agent.savecheckpoint(symbol) + + print( + f"[{symbol}] " + f"[INFO] Finished training PPO " + f"(Elapsed: {timedelta(seconds=int(time.time() - training_start))})" + ) + + completed = int((save_counter / save_count)) + total = int(round(len(df) / 1440, 0)) + + elapsed = time.time() - training_start_2 + avg_time = elapsed / max(completed, 1) + + remaining = max(total - completed, 0) + eta = remaining * avg_time + + print( + f"[{symbol}] [INFO] " + f"{completed}/{total} " + f"({completed/total*100:.1f}%) | " + f"Elapsed: {timedelta(seconds=int(elapsed))} | " + f"ETA: {timedelta(seconds=int(eta))}" + ) + + # training_start_2 = time.time() + training_start_3 = time.time() + + agent.train() + agent.savecheckpoint(symbol) + + print( + f"[{symbol}] " + f"[INFO] Finished training PPO " + f"(Elapsed: {timedelta(seconds=int(time.time() - training_start))})" + ) + + # ============================================================== + # FINAL TRAINING + # ============================================================== + + agent.train() + + agent.savecheckpoint(symbol) + + # knn._fit() + + # knn.save() + + print(f"[{symbol}] Training complete.") + + return agent + +def open_long(symbol, lot_size): + + tick = mt5.symbol_info_tick(symbol) + + entry = tick.ask + + sl = entry - 4 + + tp1 = entry + 2 + # tp2 = entry + 10 + # tp3 = entry + 15 + # tp4 = entry + 20 + + tps = [tp1] + + for tp in tps: + + request = { + "action": mt5.TRADE_ACTION_DEAL, + "symbol": symbol, + "volume": lot_size, + "type": mt5.ORDER_TYPE_BUY, + "price": entry, + "sl": sl, + "tp": tp, + "deviation": 20, + "magic": 123456, + "comment": "bot trade", + "type_time": mt5.ORDER_TIME_GTC, + "type_filling": mt5.ORDER_FILLING_IOC + } + + result = mt5.order_send(request) + + # print(result) + +def open_short(symbol, lot_size): + + tick = mt5.symbol_info_tick(symbol) + + entry = tick.bid + + sl = entry + 4 + + tp1 = entry - 2 + # tp2 = entry - 10 + # tp3 = entry - 15 + # tp4 = entry - 20 + + tps = [tp1] + + for tp in tps: + + request = { + "action": mt5.TRADE_ACTION_DEAL, + "symbol": symbol, + "volume": lot_size, + "type": mt5.ORDER_TYPE_SELL, + "price": entry, + "sl": sl, + "tp": tp, + "deviation": 20, + "magic": 123456, + "comment": "bot trade", + "type_time": mt5.ORDER_TIME_GTC, + "type_filling": mt5.ORDER_FILLING_IOC + } + + result = mt5.order_send(request) + + # print(result) + +def open_positions(symbol): + positions = mt5.positions_get(symbol=symbol) + positions = [ + p + for p in positions + if p.magic == 123456 + ] + return len(positions) + +def test_bot(symbol="XAUUSD"): + SEQ_LEN = 12 * 8 + + mt5.initialize() + account = mt5.account_info() + # if account is None: + # print("Failed to get account info") + # print(mt5.last_error()) + # return + balance = account.balance + RISK = 0.005 + # risk_per_position = max(balance * RISK / 500 / 4, 0.01) + + # tick = mt5.symbol_info_tick(symbol) + # SL_PIPS = round(tick.bid * 0.00125 * 10, 0) + # risk_per_position = min( + # max((balance * RISK) / (SL_PIPS * 10) / 4, 0.01), + # 100.0 + # ) + # risk_per_position = round(risk_per_position, 2) + # print(f"volume: {risk_per_position}") + # print(f"sl pips: {SL_PIPS}") + + FEATURES = [ + "Open", + "High", + "Low", + "Close", + "k", + "k_smooth", + "adx", + "+di", + "-di", + "EMA7", + "EMA21", + "EMA_DIFF", + "indecision", + "bullish_ob", + "bearish_ob", + "bullish_fvg", + "bearish_fvg", + "eqh", + "eql", + "bearish_mb", + "bullish_mb", + "bullish_rb", + "bearish_rb", + "vwap", + "vwap_upper", + "vwap_lower", + "above_vwap", + "below_vwap", + "vwap_above_upper", + "vwap_below_lower", + "vwap_slope", + "volume_ma", + "asia_high", + "asia_low", + "pdh", + "pdl", + "sell_score", + "buy_score" + ] + + # last_m15 = None + last_m5 = None + + agent = LSTMPPOAgent( + state_size=len(FEATURES), + hidden_size=64, + action_size=3 + ) + + # agent.model.debug = True + try: + agent.loadcheckpoint("XAUUSD") + except: + print("No file for prior training, cancelling test.") + return + + # ========================================================== + # INITIAL LOAD + # ========================================================== + + rates_m5 = mt5.copy_rates_from_pos( + symbol, + mt5.TIMEFRAME_M5, + 0, + 600 + ) + + # rates_m1 = mt5.copy_rates_from_pos( + # symbol, + # mt5.TIMEFRAME_M15, + # 0, + # 500 + # ) + + df = pd.DataFrame(rates_m5) + + df.rename(columns={ + 'open': 'Open', + 'high': 'High', + 'low': 'Low', + 'close': 'Close', + 'time': 'Date' + }, inplace=True) + + df["Date"] = pd.to_datetime(df["Date"], unit="s") + df.set_index("Date", inplace=True) + + raw_df = df + + df = add_indicators(df) + + # last_m1 = rates_m1[-1]["time"] + last_m5 = df.index[-1] + + # last_m5 = None + # last_m1 = None + + # ========================================================== + # MAIN LOOP + # ========================================================== + + while True: + + now = datetime.now() + + seconds_until_next_5m = ( + (5 - now.minute % 5) * 60 + - now.second + - now.microsecond / 1_000_000 + ) + # print(f"sleeping {seconds_until_next_5m:.0f} seconds, current time: {datetime.now()}") + if seconds_until_next_5m <= 0: + seconds_until_next_5m += 300 + + time.sleep(seconds_until_next_5m) + # print(f"slept {seconds_until_next_5m:.0f} seconds, current time: {datetime.now()}") + + tick = mt5.symbol_info_tick(symbol) + # SL_PIPS = round(tick.bid * 0.00125 * 10, 0) + SL_PIPS = 40 + risk_per_position = min( + max((balance * RISK) / (SL_PIPS * 10), 0.01), + 100.0 + ) + risk_per_position = round(risk_per_position, 2) + + # ====================================================== + # MANAGE POSITIONS EVERY NEW M1 CANDLE + # ====================================================== + + # rates_m1 = mt5.copy_rates_from_pos( + # symbol, + # mt5.TIMEFRAME_M1, + # 0, + # 2 + # ) + + # current_m1 = rates_m1[-1]["time"] + + # if current_m1 != last_m1: + + # last_m1 = current_m1 + + # manage_positions(symbol, round(SL_PIPS / 10 / 7, 2)) + # print("exited manage_positions()") + + # ====================================================== + # CHECK FOR NEW M15 CANDLE + # ====================================================== + + new_m5 = mt5.copy_rates_from_pos( + symbol, + mt5.TIMEFRAME_M5, + 0, + 1 + ) + + current_m5 = new_m5[0]["time"] + + if current_m5 != last_m5: + + last_m5 = current_m5 + + # ================================================== + # APPEND NEW CANDLE + # ================================================== + + new_row = pd.DataFrame(new_m5) + + new_row.rename(columns={ + 'open': 'Open', + 'high': 'High', + 'low': 'Low', + 'close': 'Close', + 'time': 'Date' + }, inplace=True) + + new_row["Date"] = pd.to_datetime(new_row["Date"], unit="s") + new_row.set_index("Date", inplace=True) + + if new_row.index[-1] != df.index[-1]: + + # df = pd.concat( + # [df, new_row] + # # ignore_index=True + # ) + + # df.set_index("Date", inplace=True) + + # df = ( + # df.tail(200) + # # .reset_index(drop=True) + # ) + + raw_df = pd.concat( + [raw_df, new_row] + # ignore_index=True + ) + + # raw_df = raw_df.tail(200).reset_index(drop=True) + raw_df = raw_df.tail(600) + + # print("Before indicators:", len(df)) + # df = add_indicators(raw_df.copy()) + # print(type(raw_df.index)) + # print(raw_df.index[:5]) + df = add_indicators(raw_df.copy()) + # print("After indicators:", len(df)) + # print(df.tail()) + # print(df.shape) + + # ================================================== + # BUILD STATE SEQUENCE + # ================================================== + + state_seq = ( + df[FEATURES] + .tail(SEQ_LEN) + .values + .astype(np.float32) + ) + + # ================================================== + # POSITION CHECK + # ================================================== + + open_pos = open_positions(symbol) + + # ================================================== + # PPO DECISION + # ================================================== + + # print("df shape:", df.shape) + # print("state_seq shape:", state_seq.shape) + # print("len(df):", len(df)) + + # if len(df) < SEQ_LEN: + # print(f"Skipping: len(df)={len(df)}") + # continue + + state_seq = ( + df[FEATURES] + .tail(SEQ_LEN) + .values + .astype(np.float32) + ) + + if state_seq.shape[0] != SEQ_LEN: + print(f"Bad state shape: {state_seq.shape}") + continue + + action, _, _ = agent.select_action( + state_seq, + open_pos > 0, + training=False + ) + + if df["adx"].iloc[-1] < 20: + action = 0 + + print(f"Test action: {action}") + + # ================================================== + # OPEN NEW TRADE + # ================================================== + + if open_pos == 0: + + account = mt5.account_info() + + balance = account.balance + + # risk_per_position = max( + # balance * RISK / 500 / 4, + # 0.01 + # ) + + # if action == 1 and df["adx"].iloc[-1] > 20 and df["+di"].iloc[-1] > df["-di"].iloc[-1] and df["EMA_DIFF"].iloc[-1] > 0 and df["k"].iloc[-1] < 80: + # if action == 1 and df["buy_score"].iloc[-1] > df["sell_score"].iloc[-1]: + if action == 1 and df["EMA7"].iloc[-1] > df["EMA21"].iloc[-1] and df["k"].iloc[-1] < 80: + # if action == 1: + # print( + # f"[{symbol}] PPO BUY" + # ) + + open_long( + symbol, + risk_per_position + ) + + # elif action == 2 and df["adx"].iloc[-1] > 20 and df["-di"].iloc[-1] > df["+di"].iloc[-1] and df["EMA_DIFF"].iloc[-1] < 0 and df["k"].iloc[-1] > 20: + # elif action == 2 and df["buy_score"].iloc[-1] < df["sell_score"].iloc[-1]: + elif action == 2 and df["EMA7"].iloc[-1] < df["EMA21"].iloc[-1] and df["k"].iloc[-1] > 20: + # elif action == 2: + # print( + # f"[{symbol}] PPO SELL" + # ) + + open_short( + symbol, + risk_per_position + ) + + # else: + + # print( + # f"[{symbol}] PPO HOLD" + # ) + +def main(): + + parser = argparse.ArgumentParser() + + parser.add_argument("--train", action="store_true") + parser.add_argument("--test", action="store_true") + parser.add_argument("--symbol", default="XAUUSD-VIP") + + args = parser.parse_args() + + threads = [] + + if args.train: + t = threading.Thread( + target=train_bot, + # args=(args.symbol), + daemon=True + ) + t.start() + threads.append(t) + + if args.test: + t = threading.Thread( + target=test_bot, + args=(args.symbol,), + daemon=True + ) + t.start() + threads.append(t) + + for t in threads: + t.join() + +main()