pylivetrader 0.0.7

simple live trading framework

pylivetrader

pylivetrader is a simple python live trading framework with zipline interface. The main purpose is to run algorithms developed in the Quantopian platform in live trading via broker API. In order to convert your algorithm for pylivetrader, please read the migration document.

Simple Usage

Here is the example dual moving average algorithm (by quantopian/zipline). We provide mostly the same API interfaces with zipline.

from pylivetrader.api import order_target, symbol

def initialize(context):
    context.i = 0
    context.asset = symbol('AAPL')

def handle_data(context, data):
    # Compute averages
    # data.history() has to be called with the same params
    # from above and returns a pandas dataframe.
    short_mavg = data.history(context.asset, 'price', bar_count=100, frequency="1m").mean()
    long_mavg = data.history(context.asset, 'price', bar_count=300, frequency="1m").mean()

    # Trading logic
    if short_mavg > long_mavg:
        # order_target orders as many shares as needed to
        # achieve the desired number of shares.
        order_target(context.asset, 100)
    elif short_mavg < long_mavg:
        order_target(context.asset, 0)

You can run your algorithm from the CLI tool named pylivetrader, simply like below. Then your algorithm starts running with broker API. You don't need the data bundle file in advance unlike zipline does.

$ pylivetrader run -f algo.py --backend-config config.yaml

Config file is just simple yaml or json format.

$ cat config.yaml
key_id: BROKER_API_KEY
secret: BROKER_SECRET

Installation

Install with pip.

$ pip install pylivetrader

Additionally, pylivetrader works well with pipeline-live.

Supported Broker

Alpaca

Configuration by environment variables.

$ export APCA_API_KEY_ID={your api key id}
$ export APCA_API_SECRET_KEY={your api secret key}
$ pylivetrader run -f algo.py

Configuration by config file. Either yaml or json.

$ cat config.yaml
key_id: {your api key id}
secret: {your api secret key}
$ pylivetrader run -f algo.py --backend-config config.yaml

Docker

If you are already familiar with Docker, it is a good idea to try our docker image alpacamarkets/pylivetrader. This has installed pylivetrader so you can start right away without worrying about your python environment. See more details in the dockerfiles directory.

If your algorithm file is called algo.py, this could be it.

docker run -v $PWD:/work -w /work alpacamarkets/pylivetrader pylivetrader run -f algo.py

Make sure you set up environment variables for the backend (use -e KEY=VAL for docker command).

Smoke Test

pylivetrader provides a facility for smoke testing. This helps catch issues such as typos, program errors and simple oversights. The following is an example of smoke testing.

import algo

from pylivetrader.testing.smoke import harness


def before_run(context, backend):
    '''This hook is called before algorithm starts.'''

    # Populate existing position
    backend.set_position(
        'A', 10, 200,
    )

    # modify some fields of context after `initialize(context)` is called
    _init = context._initialize
    def wrapper(ctx):
        _init(ctx)
        ctx.age[ctx.symbol('A')] = 3
        ctx.age[ctx.symbol('B')] = 2

    context._initialize = wrapper

def test_algo():
    pipeline = harness.DefaultPipelineHooker()

    # run the algorithm under the simulation environment
    harness.run_smoke(algo,
        before_run_hook=before_run,
        pipeline_hook=pipeline,
    )


if __name__ == '__main__':
    import logging
    logging.basicConfig(level=logging.DEBUG)
    test_algo()

This exercises the algorithm code by harnessing synthesic backend and price data. The pylivetrader.testing.smoke package provides the backend and simulator clock classes so that it simulates a market day from open to close.

By default, the backend creates a universe with 50 stocks ('A' .. 'AX'). For each symbol, you can query synthesic historical price, and orders are managed within this simulator without having to set up a real remote backend API. Additionally, you can hook up a couple of code injection points such as before_run_hook and pipeline_hook. In this example, the setup code creates a pre-populated position in the backend so you can test the algorithm code path that accepts existing positions.

A DefaultPipelineHooker instance can return a synthesic pipeline result with the same column names/types, inferred from the pipeline object given in the attach_pipeline API.

Again, the purpose of this smoke testing is to actually exercise various code path to make sure there is no easy mistakes. This code works well with standard test framework such as pytest and you can easily report line coverage using those frameworks too.