434 lines
19 KiB
Python
434 lines
19 KiB
Python
import asyncio
|
|
import logging
|
|
import time
|
|
from collections import deque
|
|
from datetime import datetime, timedelta
|
|
|
|
import matplotlib.pyplot as plt
|
|
import numpy as np
|
|
import pandas as pd
|
|
from matplotlib.colors import LogNorm
|
|
|
|
from core.data_provider import DataProvider, MarketTick
|
|
from core.config import get_config
|
|
|
|
# Configure logging
|
|
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
class COBStabilityTester:
|
|
def __init__(self, symbol='ETHUSDT', duration_seconds=15):
|
|
self.symbol = symbol
|
|
self.duration = timedelta(seconds=duration_seconds)
|
|
self.ticks = deque()
|
|
|
|
# Set granularity (buckets) based on symbol
|
|
if 'ETH' in symbol.upper():
|
|
self.price_granularity = 1.0 # 1 USD for ETH
|
|
elif 'BTC' in symbol.upper():
|
|
self.price_granularity = 10.0 # 10 USD for BTC
|
|
else:
|
|
self.price_granularity = 1.0 # Default 1 USD
|
|
|
|
logger.info(f"Using price granularity: ${self.price_granularity} for {symbol}")
|
|
|
|
# Initialize DataProvider the same way as clean_dashboard
|
|
logger.info("Initializing DataProvider like in clean_dashboard...")
|
|
self.data_provider = DataProvider() # Use default constructor like clean_dashboard
|
|
|
|
# Initialize COB data collection like clean_dashboard does
|
|
self.cob_data_received = 0
|
|
self.latest_cob_data = {}
|
|
|
|
# Store all COB snapshots for heatmap generation
|
|
self.cob_snapshots = deque()
|
|
self.price_data = [] # For price line chart
|
|
|
|
self.start_time = None
|
|
self.subscriber_id = None
|
|
self.last_log_time = None
|
|
|
|
def _tick_callback(self, tick: MarketTick):
|
|
"""Callback function to receive ticks from the DataProvider."""
|
|
if self.start_time is None:
|
|
self.start_time = datetime.now()
|
|
logger.info(f"Started collecting ticks at {self.start_time}")
|
|
|
|
# Store all ticks
|
|
self.ticks.append(tick)
|
|
|
|
def _cob_data_callback(self, symbol: str, cob_data: dict):
|
|
"""Callback function to receive COB data from the DataProvider."""
|
|
# Debug: Log first few callbacks to see what symbols we're getting
|
|
if self.cob_data_received < 5:
|
|
logger.info(f"DEBUG: Received COB data for symbol '{symbol}' (target: '{self.symbol}')")
|
|
|
|
# Filter to only our requested symbol - handle both formats (ETH/USDT and ETHUSDT)
|
|
normalized_symbol = symbol.replace('/', '')
|
|
normalized_target = self.symbol.replace('/', '')
|
|
if normalized_symbol != normalized_target:
|
|
if self.cob_data_received < 5:
|
|
logger.info(f"DEBUG: Skipping symbol '{symbol}' (normalized: '{normalized_symbol}' vs target: '{normalized_target}')")
|
|
return
|
|
|
|
self.cob_data_received += 1
|
|
self.latest_cob_data[symbol] = cob_data
|
|
|
|
# Store the complete COB snapshot for heatmap generation
|
|
if 'bids' in cob_data and 'asks' in cob_data:
|
|
# Debug: Log structure of first few COB snapshots
|
|
if len(self.cob_snapshots) < 3:
|
|
logger.info(f"DEBUG: COB data structure - bids: {len(cob_data['bids'])} items, asks: {len(cob_data['asks'])} items")
|
|
if cob_data['bids']:
|
|
logger.info(f"DEBUG: First bid: {cob_data['bids'][0]}")
|
|
if cob_data['asks']:
|
|
logger.info(f"DEBUG: First ask: {cob_data['asks'][0]}")
|
|
|
|
snapshot = {
|
|
'timestamp': cob_data.get('timestamp', datetime.now()),
|
|
'bids': cob_data['bids'],
|
|
'asks': cob_data['asks'],
|
|
'stats': cob_data.get('stats', {})
|
|
}
|
|
self.cob_snapshots.append(snapshot)
|
|
|
|
# Log bucketed COB data every second
|
|
now = datetime.now()
|
|
if self.last_log_time is None or (now - self.last_log_time).total_seconds() >= 1.0:
|
|
self.last_log_time = now
|
|
self._log_bucketed_cob_data(cob_data)
|
|
|
|
# Convert COB data to tick-like format for analysis
|
|
if 'stats' in cob_data and 'mid_price' in cob_data['stats']:
|
|
mid_price = cob_data['stats']['mid_price']
|
|
if mid_price > 0:
|
|
# Store price data for line chart
|
|
self.price_data.append({
|
|
'timestamp': cob_data.get('timestamp', datetime.now()),
|
|
'price': mid_price
|
|
})
|
|
|
|
# Create a synthetic tick from COB data
|
|
synthetic_tick = MarketTick(
|
|
symbol=symbol,
|
|
timestamp=cob_data.get('timestamp', datetime.now()),
|
|
price=mid_price,
|
|
volume=cob_data.get('stats', {}).get('total_volume', 0),
|
|
quantity=0, # Not available in COB data
|
|
side='unknown', # COB data doesn't have side info
|
|
trade_id=f"cob_{self.cob_data_received}",
|
|
is_buyer_maker=False,
|
|
raw_data=cob_data
|
|
)
|
|
self.ticks.append(synthetic_tick)
|
|
|
|
if self.cob_data_received % 10 == 0: # Log every 10th update
|
|
logger.info(f"COB update #{self.cob_data_received}: {symbol} @ ${mid_price:.2f}")
|
|
|
|
def _log_bucketed_cob_data(self, cob_data: dict):
|
|
"""Log bucketed COB data every second"""
|
|
try:
|
|
if 'bids' not in cob_data or 'asks' not in cob_data:
|
|
logger.info("COB-1s: No order book data available")
|
|
return
|
|
|
|
if 'stats' not in cob_data or 'mid_price' not in cob_data['stats']:
|
|
logger.info("COB-1s: No mid price available")
|
|
return
|
|
|
|
mid_price = cob_data['stats']['mid_price']
|
|
if mid_price <= 0:
|
|
return
|
|
|
|
# Bucket the order book data
|
|
bid_buckets = {}
|
|
ask_buckets = {}
|
|
|
|
# Process bids (top 10)
|
|
for bid in cob_data['bids'][:10]:
|
|
try:
|
|
if isinstance(bid, dict):
|
|
price = float(bid['price'])
|
|
size = float(bid['size'])
|
|
elif isinstance(bid, (list, tuple)) and len(bid) >= 2:
|
|
price = float(bid[0])
|
|
size = float(bid[1])
|
|
else:
|
|
continue
|
|
|
|
bucketed_price = round(price / self.price_granularity) * self.price_granularity
|
|
bid_buckets[bucketed_price] = bid_buckets.get(bucketed_price, 0) + size
|
|
except (ValueError, TypeError, IndexError):
|
|
continue
|
|
|
|
# Process asks (top 10)
|
|
for ask in cob_data['asks'][:10]:
|
|
try:
|
|
if isinstance(ask, dict):
|
|
price = float(ask['price'])
|
|
size = float(ask['size'])
|
|
elif isinstance(ask, (list, tuple)) and len(ask) >= 2:
|
|
price = float(ask[0])
|
|
size = float(ask[1])
|
|
else:
|
|
continue
|
|
|
|
bucketed_price = round(price / self.price_granularity) * self.price_granularity
|
|
ask_buckets[bucketed_price] = ask_buckets.get(bucketed_price, 0) + size
|
|
except (ValueError, TypeError, IndexError):
|
|
continue
|
|
|
|
# Format for log output
|
|
bid_str = ", ".join([f"${p:.0f}:{s:.3f}" for p, s in sorted(bid_buckets.items(), reverse=True)])
|
|
ask_str = ", ".join([f"${p:.0f}:{s:.3f}" for p, s in sorted(ask_buckets.items())])
|
|
|
|
logger.info(f"COB-1s @ ${mid_price:.2f} | BIDS: {bid_str} | ASKS: {ask_str}")
|
|
|
|
except Exception as e:
|
|
logger.warning(f"Error logging bucketed COB data: {e}")
|
|
|
|
async def run_test(self):
|
|
"""Run the data collection and plotting test."""
|
|
logger.info(f"Starting COB stability test for {self.symbol} for {self.duration.total_seconds()} seconds...")
|
|
|
|
# Initialize COB collection like clean_dashboard does
|
|
try:
|
|
logger.info("Starting COB collection in data provider...")
|
|
self.data_provider.start_cob_collection()
|
|
logger.info("Started COB collection in data provider")
|
|
|
|
# Subscribe to COB updates
|
|
logger.info("Subscribing to COB data updates...")
|
|
self.data_provider.subscribe_to_cob(self._cob_data_callback)
|
|
logger.info("Subscribed to COB data updates from data provider")
|
|
except Exception as e:
|
|
logger.error(f"Failed to start COB collection or subscribe: {e}")
|
|
|
|
# Subscribe to ticks as fallback
|
|
try:
|
|
self.subscriber_id = self.data_provider.subscribe_to_ticks(self._tick_callback, symbols=[self.symbol])
|
|
logger.info("Subscribed to tick data as fallback")
|
|
except Exception as e:
|
|
logger.warning(f"Failed to subscribe to ticks: {e}")
|
|
|
|
# Start the data provider's real-time streaming
|
|
try:
|
|
await self.data_provider.start_real_time_streaming()
|
|
logger.info("Started real-time streaming")
|
|
except Exception as e:
|
|
logger.error(f"Failed to start real-time streaming: {e}")
|
|
|
|
# Collect data for the specified duration
|
|
self.start_time = datetime.now()
|
|
while datetime.now() - self.start_time < self.duration:
|
|
await asyncio.sleep(1)
|
|
logger.info(f"Collected {len(self.ticks)} ticks so far...")
|
|
|
|
# Stop streaming and unsubscribe
|
|
await self.data_provider.stop_real_time_streaming()
|
|
self.data_provider.unsubscribe_from_ticks(self.subscriber_id)
|
|
|
|
logger.info(f"Finished collecting data. Total ticks: {len(self.ticks)}")
|
|
|
|
# Plot the results
|
|
if self.price_data and self.cob_snapshots:
|
|
self.create_price_heatmap_chart()
|
|
elif self.ticks:
|
|
self._create_simple_price_chart()
|
|
else:
|
|
logger.warning("No data was collected. Cannot generate plot.")
|
|
|
|
def create_price_heatmap_chart(self):
|
|
"""Create a visualization with price chart and order book heatmap."""
|
|
if not self.price_data or not self.cob_snapshots:
|
|
logger.warning("Insufficient data to plot.")
|
|
return
|
|
|
|
logger.info(f"Creating price and order book heatmap chart...")
|
|
logger.info(f"Data summary: {len(self.price_data)} price points, {len(self.cob_snapshots)} COB snapshots")
|
|
|
|
# Prepare price data with consistent timestamp handling
|
|
price_df = pd.DataFrame(self.price_data)
|
|
price_df['timestamp'] = pd.to_datetime(price_df['timestamp'])
|
|
|
|
logger.info(f"Price data time range: {price_df['timestamp'].min()} to {price_df['timestamp'].max()}")
|
|
logger.info(f"Price range: ${price_df['price'].min():.2f} to ${price_df['price'].max():.2f}")
|
|
|
|
# Extract order book data for heatmap with consistent timestamp handling
|
|
heatmap_data = []
|
|
for snapshot in self.cob_snapshots:
|
|
timestamp = pd.to_datetime(snapshot['timestamp']) # Ensure datetime
|
|
|
|
for side in ['bids', 'asks']:
|
|
if side not in snapshot or not snapshot[side]:
|
|
continue
|
|
|
|
# Take top 50 levels for better visualization
|
|
orders = snapshot[side][:50]
|
|
for order in orders:
|
|
try:
|
|
# Handle both dict and list formats
|
|
if isinstance(order, dict):
|
|
price = float(order['price'])
|
|
size = float(order['size'])
|
|
elif isinstance(order, (list, tuple)) and len(order) >= 2:
|
|
price = float(order[0])
|
|
size = float(order[1])
|
|
else:
|
|
continue
|
|
|
|
# Apply granularity bucketing
|
|
bucketed_price = round(price / self.price_granularity) * self.price_granularity
|
|
|
|
heatmap_data.append({
|
|
'time': timestamp,
|
|
'price': bucketed_price,
|
|
'size': size,
|
|
'side': side
|
|
})
|
|
except (ValueError, TypeError, IndexError) as e:
|
|
continue
|
|
|
|
if not heatmap_data:
|
|
logger.warning("No valid heatmap data found, creating price chart only")
|
|
self._create_simple_price_chart()
|
|
return
|
|
|
|
heatmap_df = pd.DataFrame(heatmap_data)
|
|
logger.info(f"Heatmap data: {len(heatmap_df)} order book entries")
|
|
logger.info(f"Heatmap time range: {heatmap_df['time'].min()} to {heatmap_df['time'].max()}")
|
|
|
|
# Create plot with better time handling
|
|
fig, ax = plt.subplots(figsize=(16, 10))
|
|
|
|
# Determine overall time range
|
|
all_times = pd.concat([price_df['timestamp'], heatmap_df['time']])
|
|
time_min = all_times.min()
|
|
time_max = all_times.max()
|
|
|
|
# Create price range for heatmap
|
|
price_min = min(price_df['price'].min(), heatmap_df['price'].min()) - self.price_granularity * 2
|
|
price_max = max(price_df['price'].max(), heatmap_df['price'].max()) + self.price_granularity * 2
|
|
|
|
logger.info(f"Chart time range: {time_min} to {time_max}")
|
|
logger.info(f"Chart price range: ${price_min:.2f} to ${price_max:.2f}")
|
|
|
|
# Create heatmap first (background)
|
|
for side, cmap, alpha in zip(['bids', 'asks'], ['Greens', 'Reds'], [0.6, 0.6]):
|
|
side_df = heatmap_df[heatmap_df['side'] == side]
|
|
if not side_df.empty:
|
|
# Create more granular bins
|
|
time_bins = pd.date_range(time_min, time_max, periods=min(100, len(side_df) // 10 + 10))
|
|
price_bins = np.arange(price_min, price_max + self.price_granularity, self.price_granularity)
|
|
|
|
try:
|
|
# Convert to seconds for histogram
|
|
time_seconds = (side_df['time'] - time_min).dt.total_seconds()
|
|
time_range_seconds = (time_max - time_min).total_seconds()
|
|
|
|
if time_range_seconds > 0:
|
|
hist, xedges, yedges = np.histogram2d(
|
|
time_seconds,
|
|
side_df['price'],
|
|
bins=[np.linspace(0, time_range_seconds, len(time_bins)), price_bins],
|
|
weights=side_df['size']
|
|
)
|
|
|
|
# Convert back to datetime for plotting
|
|
time_edges = pd.to_datetime(xedges, unit='s', origin=time_min)
|
|
|
|
if hist.max() > 0: # Only plot if we have data
|
|
pcm = ax.pcolormesh(time_edges, yedges, hist.T,
|
|
cmap=cmap, alpha=alpha, shading='auto')
|
|
logger.info(f"Plotted {side} heatmap: max value = {hist.max():.2f}")
|
|
except Exception as e:
|
|
logger.warning(f"Error creating {side} heatmap: {e}")
|
|
|
|
# Plot price line on top
|
|
ax.plot(price_df['timestamp'], price_df['price'], 'yellow', linewidth=2,
|
|
label='Mid Price', alpha=0.9, zorder=10)
|
|
|
|
# Enhance plot appearance
|
|
ax.set_title(f'Price Chart with Order Book Heatmap - {self.symbol}\n'
|
|
f'Granularity: ${self.price_granularity} | Duration: {self.duration.total_seconds()}s\n'
|
|
f'Green=Bids, Red=Asks (darker = more volume)', fontsize=14)
|
|
ax.set_xlabel('Time')
|
|
ax.set_ylabel('Price (USDT)')
|
|
ax.legend(loc='upper left')
|
|
ax.grid(True, alpha=0.3)
|
|
|
|
# Format time axis
|
|
ax.set_xlim(time_min, time_max)
|
|
fig.autofmt_xdate()
|
|
|
|
plt.tight_layout()
|
|
plot_filename = f"price_heatmap_chart_{self.symbol.replace('/', '_')}_{datetime.now():%Y%m%d_%H%M%S}.png"
|
|
plt.savefig(plot_filename, dpi=150, bbox_inches='tight')
|
|
logger.info(f"Price and heatmap chart saved to {plot_filename}")
|
|
plt.show()
|
|
|
|
def _create_simple_price_chart(self):
|
|
"""Create a simple price chart as fallback"""
|
|
logger.info("Creating simple price chart as fallback...")
|
|
|
|
prices = []
|
|
times = []
|
|
|
|
for tick in self.ticks:
|
|
if tick.price > 0:
|
|
prices.append(tick.price)
|
|
times.append(tick.timestamp)
|
|
|
|
if not prices:
|
|
logger.warning("No price data to plot")
|
|
return
|
|
|
|
fig, ax = plt.subplots(figsize=(15, 8))
|
|
ax.plot(pd.to_datetime(times), prices, 'cyan', linewidth=1)
|
|
ax.set_title(f'Price Chart - {self.symbol}')
|
|
ax.set_xlabel('Time')
|
|
ax.set_ylabel('Price (USDT)')
|
|
fig.autofmt_xdate()
|
|
|
|
plot_filename = f"cob_price_chart_{self.symbol.replace('/', '_')}_{datetime.now():%Y%m%d_%H%M%S}.png"
|
|
plt.savefig(plot_filename)
|
|
logger.info(f"Price chart saved to {plot_filename}")
|
|
plt.show()
|
|
|
|
|
|
async def main(symbol='ETHUSDT', duration_seconds=15):
|
|
"""Main function to run the COB test with configurable parameters.
|
|
|
|
Args:
|
|
symbol: Trading symbol (default: ETHUSDT)
|
|
duration_seconds: Test duration in seconds (default: 15)
|
|
"""
|
|
logger.info(f"Starting COB test with symbol={symbol}, duration={duration_seconds}s")
|
|
tester = COBStabilityTester(symbol=symbol, duration_seconds=duration_seconds)
|
|
await tester.run_test()
|
|
|
|
|
|
if __name__ == "__main__":
|
|
import sys
|
|
|
|
# Parse command line arguments
|
|
symbol = 'ETHUSDT' # Default
|
|
duration = 15 # Default
|
|
|
|
if len(sys.argv) > 1:
|
|
symbol = sys.argv[1]
|
|
if len(sys.argv) > 2:
|
|
try:
|
|
duration = int(sys.argv[2])
|
|
except ValueError:
|
|
logger.warning(f"Invalid duration '{sys.argv[2]}', using default 15 seconds")
|
|
|
|
logger.info(f"Configuration: Symbol={symbol}, Duration={duration}s")
|
|
logger.info(f"Granularity: {'1 USD for ETH' if 'ETH' in symbol.upper() else '10 USD for BTC' if 'BTC' in symbol.upper() else '1 USD default'}")
|
|
|
|
try:
|
|
asyncio.run(main(symbol, duration))
|
|
except KeyboardInterrupt:
|
|
logger.info("Test interrupted by user.")
|