mm-toolbox 1.0.0b3

High-performance Python tools for market making systems

MM Toolbox

MM Toolbox is a Python library designed to provide high-performance tools for market making strategies.

Contents

mm-toolbox/
├── src/
│   └── mm_toolbox/
│       ├── candles/            # Tools for handling and aggregating candlestick data
│       ├── logging/            # Lightweight logger + Discord/Telegram support
│       │   ├── standard/       # Standard logger implementation
│       │   └── advanced/       # Distributed HFT logger (worker/master)
│       ├── misc/               # Filtering helpers
│       │   └── filter/         # Bounds-based change filter
│       ├── moving_average/     # Various moving averages (EMA/SMA/WMA/TEMA)
│       ├── orderbook/          # Multiple orderbook implementations & tools
│       │   ├── standard/       # Python-based orderbook
│       │   └── advanced/       # High-performance Cython orderbook
│       ├── rate_limiter/       # Token bucket rate limiter
│       ├── ringbuffer/         # Efficient fixed-size circular buffers
│       ├── rounding/           # Fast price/size rounding utilities
│       ├── time/               # Time utilities
│       ├── websocket/          # WebSocket clients + verification tools
│       └── weights/            # Weight generators (EMA/geometric/logarithmic)
├── tests/                      # Unit tests for all the modules
├── pyproject.toml              # Project configuration and dependencies
├── LICENSE                     # License information
├── README.md                   # Main documentation file
└── setup.py                    # Setup script for building Cython extensions

Installation

MM Toolbox is available on PyPI and can be installed using pip:

pip install mm_toolbox

To try the beta without replacing a stable install, use a separate virtual environment and install the pre-release:

python -m venv mm_toolbox_beta
source mm_toolbox_beta/bin/activate
pip install mm-toolbox==1.0.0b3

To always pull the latest pre-release:

pip install --pre mm-toolbox

To install directly from the source, clone the repository and install the dependencies:

git clone https://github.com/beatzxbt/mm-toolbox.git
cd mm-toolbox
# Install uv if you haven't already: curl -LsSf https://astral.sh/uv/install.sh | sh
uv sync --all-groups
make build  # Compile Cython extensions

Usage

After installation, you can start using MM Toolbox by importing the necessary modules:

from mm_toolbox.moving_average import ExponentialMovingAverage as EMA
from mm_toolbox.orderbook import Orderbook, OrderbookLevel
from mm_toolbox.logging.standard import Logger, LogLevel, LoggerConfig

# Example usage:
ema = EMA(window=10, is_fast=True)
tick_size = 0.01
lot_size = 0.001
orderbook = Orderbook(tick_size=tick_size, lot_size=lot_size, size=100)
orderbook.consume_bbo(
    ask=OrderbookLevel.from_values(100.01, 1.2, 1, tick_size, lot_size),
    bid=OrderbookLevel.from_values(100.00, 1.0, 1, tick_size, lot_size),
)
logger = Logger(
    name="Example",
    config=LoggerConfig(base_level=LogLevel.INFO, do_stdout=True),
)

Latest release notes (v1.0.0, feature complete)

Major Architecture Shift: Numba → Cython/C

MM Toolbox v1.0.0 represents a fundamental shift from Numba-accelerated code to Cython/C implementations. This transition brings significant benefits:

Performance Improvements: Core components now see speed improvements of 5–30x compared to previous Numba implementations, with some components achieving even greater gains.

Better Interoperability: Cython/C extensions integrate seamlessly with the Python ecosystem. Unlike Numba's JIT compilation, Cython extensions are pre-compiled, eliminating warm-up times and providing consistent performance from the first call. This makes MM Toolbox more suitable for production HFT systems where predictable latency is critical.

Type Safety & Tooling: Full type stub support (.pyi files) enables better IDE integration, static type checking with Pyright, and improved developer experience. Cython's explicit typing model also catches more errors at compile time.

Zero-Allocation Designs: Many components have been redesigned with zero-allocation patterns, reducing GC pressure and improving performance in tight loops.

The v1.0 feature set is complete. Each component ships with a focused README that covers API details, architecture notes, and usage examples.

Component Highlights

Candles (mm_toolbox.candles): High-performance candle aggregation with time, tick, volume, price, and multi-trigger buckets. Maintains a live latest_candle, stores completed candles in a ring buffer, and supports async iteration for stream processing.

Misc (mm_toolbox.misc): Utility helpers including DataBoundsFilter for bounds-based change detection.

Rate Limiter (mm_toolbox.rate_limiter): Token-bucket rate limiting with optional burst policies and per-second sub-buckets, plus explicit state tracking via RateLimitState.

Ringbuffer (mm_toolbox.ringbuffer): Efficient circular buffers with multiple implementations:

• NumericRingBuffer: Fast numeric data handling
• BytesRingBuffer: Optimized for byte arrays
• BytesRingBufferFast: Pre-allocated slots for predictable byte workloads
• GenericRingBuffer: Flexible support for any Python type
• IPCRingBuffer: PUSH/PULL transport for SPSC/MPSC/SPMC topologies
• SharedMemoryRingBuffer: SPSC shared-memory ring buffer (POSIX-only)

All ring buffers share consistent insert/consume semantics and overwrite oldest entries on overflow for bounded memory usage.

Moving Average (mm_toolbox.moving_average): Comprehensive moving average implementations including EMA, SMA, WMA, and TEMA (Triple Exponential Moving Average). All implementations support .next() for previewing future values without state mutation.

Orderbook (mm_toolbox.orderbook): Dual implementation approach with aligned APIs:

• standard: Pure Python implementation for flexibility
• advanced: Zero-allocation Cython implementation achieving >4x faster BBO updates and >5x faster per-level batch updates

Websocket (mm_toolbox.websocket): WebSocket connection management built on PicoWs with latency tracking, ring-buffered message ingestion, and pool routing to the fastest connection.

Logging (mm_toolbox.logging): Two-tier logging system:

• standard: Lightweight logger with Discord/Telegram support
• advanced: Distributed HFT logger with worker/master architecture, batching, and customizable handlers

Rounding (mm_toolbox.rounding): Fast, directional price/size rounding with scalar and vectorized paths.

Time (mm_toolbox.time): High-performance time utilities for timestamp operations.

Weights (mm_toolbox.weights): Weight generators for EMA, geometric, and logarithmic weighting schemes.

Breaking Changes

These notes compare this branch (v1.0b) against master.

• Top-level imports removed: mm_toolbox no longer re-exports classes/functions; import from submodules instead (e.g., mm_toolbox.orderbook, mm_toolbox.time, mm_toolbox.logging.standard).
• Numba stack removed: mm_toolbox.numba and all Numba-based implementations are gone (old orderbook, ringbuffers, rounding, and array helpers).
• Orderbook rewrite: the Numba Orderbook(size) (arrays + refresh/update_*/seq_id) is replaced by standard/advanced orderbooks that require tick_size + lot_size and ingest OrderbookLevel objects via consume_snapshot, consume_deltas, and consume_bbo(ask, bid).
• Candles redesign: candle aggregation now uses Trade/Candle objects, async iteration, and a generic ringbuffer; MultiTriggerCandles is renamed to MultiCandles with max_size, and PriceCandles was added.
• Logging restructure: mm_toolbox.logging.Logger and FileLogConfig/DiscordLogConfig/TelegramLogConfig were removed; use mm_toolbox.logging.standard or mm_toolbox.logging.advanced and pass handler objects directly.
• Ringbuffer API replaced: RingBufferSingleDim*, RingBufferTwoDim*, and RingBufferMultiDim were removed; use NumericRingBuffer, GenericRingBuffer, BytesRingBuffer, BytesRingBufferFast, and IPC/SHM variants.
• Rounding API replaced: Round was removed; use Rounder + RounderConfig (directional rounding is configurable).
• Websocket rewrite: SingleWsConnection, WsStandard, WsFast, WsPoolEvictionPolicy, and VerifyWsPayload were removed; use WsConnection, WsSingle, WsPool, and their config/state types.
• Moving averages/time changes: HullMovingAverage was removed; SimpleMovingAverage and TimeExponentialMovingAverage were added. Time helpers now return integers and time_iso8601() accepts an optional timestamp.

Migration Guide

Follow these steps when moving from master to v1.0b.

1. Install/build changes (source installs):

   • Poetry/requirements-based installs from master are replaced by uv + Cython builds.

   uv sync --all-groups
   make build
   

2. Update imports (top-level exports removed):

   # master
   from mm_toolbox import Orderbook, ExponentialMovingAverage, Round, time_s
   
   # v1.0b
   from mm_toolbox.orderbook import Orderbook
   from mm_toolbox.moving_average import ExponentialMovingAverage
   from mm_toolbox.rounding import Rounder, RounderConfig
   from mm_toolbox.time import time_s
   

3. Orderbook migration:

   • Old API used NumPy arrays + sequence IDs; new API uses OrderbookLevel objects and does not track seq_id.
   • refresh/update_bids/update_asks -> consume_snapshot/consume_deltas; update_bbo -> consume_bbo(ask, bid).

   # master
   ob = Orderbook(size=100)
   ob.refresh(asks_np, bids_np, new_seq_id=42)
   ob.update_bbo(bid_price, bid_size, ask_price, ask_size, new_seq_id=43)
   
   # v1.0b
   from mm_toolbox.orderbook import Orderbook, OrderbookLevel
   
   ob = Orderbook(tick_size=0.01, lot_size=0.001, size=100)
   asks = [
       OrderbookLevel.from_values(p, s, norders=0, tick_size=0.01, lot_size=0.001)
       for p, s in asks_np
   ]
   bids = [
       OrderbookLevel.from_values(p, s, norders=0, tick_size=0.01, lot_size=0.001)
       for p, s in bids_np
   ]
   ob.consume_snapshot(asks=asks, bids=bids)
   ob.consume_bbo(
       ask=OrderbookLevel.from_values(ask_price, ask_size, 0, 0.01, 0.001),
       bid=OrderbookLevel.from_values(bid_price, bid_size, 0, 0.01, 0.001),
   )
   

   • If you need the Cython implementation, import AdvancedOrderbook from mm_toolbox.orderbook.advanced.

4. Candles migration:

   • Trades are now passed as Trade objects and candles are stored as Candle objects.
   • MultiTriggerCandles -> MultiCandles (max_volume -> max_size, max_ticks is now int).

   from mm_toolbox.candles import TimeCandles, MultiCandles
   from mm_toolbox.candles.base import Trade
   
   candles = TimeCandles(secs_per_bucket=1.0, num_candles=1000)
   candles.process_trade(Trade(time_ms=1700000000000, is_buy=True, price=100.0, size=0.5))
   

5. Logging migration:

   • Standard logger lives in mm_toolbox.logging.standard, advanced logger in mm_toolbox.logging.advanced.

   from mm_toolbox.logging.standard import Logger, LoggerConfig, LogLevel
   from mm_toolbox.logging.standard.handlers import FileLogHandler
   
   logger = Logger(
       name="example",
       config=LoggerConfig(base_level=LogLevel.INFO, do_stdout=True),
       handlers=[FileLogHandler("logs.txt", create=True)],
   )
   

6. Ringbuffer migration:

   • RingBufferSingleDimFloat/Int -> NumericRingBuffer(max_capacity=..., dtype=...)
   • RingBufferTwoDim*/RingBufferMultiDim -> GenericRingBuffer (store arrays/objects)
   • BytesRingBufferFast now rejects inserts larger than its slot size.

7. Rounding migration:

   from mm_toolbox.rounding import Rounder, RounderConfig
   
   rounder = Rounder(RounderConfig.default(tick_size=0.01, lot_size=0.001))
   price = rounder.bid(100.1234)
   

8. Websocket migration:

   from mm_toolbox.websocket import WsConnectionConfig, WsSingle
   
   config = WsConnectionConfig.default("wss://example", on_connect=[b"SUBSCRIBE ..."])
   ws = WsSingle(config)
   await ws.start()
   

9. Moving averages + time:

   • HullMovingAverage was removed; use SimpleMovingAverage or TimeExponentialMovingAverage.
   • time_s/time_ms/... return integers now; time_iso8601() optionally formats a provided timestamp.

Roadmap

v1.1.0

• Websocket: Move WsPool and WsSingle into Cython classes to eliminate call_soon_threadsafe overhead in hot paths.
• Logging: Move more advanced logger components into C to unlock similar performance gains.
• Orderbook: Add Cython helpers to build/consume levels from string pair lists (e.g., [[price, size], ...]) to avoid Python loops in depth snapshots/deltas.

v1.2.0

Parsers: Introduction of high-performance parsing utilities including JSON parsers and crypto exchange-specific parsers (e.g., Binance top-of-book parser).

License

MM Toolbox is licensed under the MIT License. See the LICENSE file for more information.

Contributing

Contributions are welcome! Please read the CONTRIBUTING.md for guidelines on how to contribute to this project.

Contact

For questions or support, please open an issue. I can also be reached on Twitter and Discord :D