Python live trade execution library with zipline interface.
Branch: master
Clone or download
Pull request Compare This branch is 105 commits behind alpacahq:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


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 --backend-config config.yaml

Config file is just simple yaml or json format.

$ cat config.yaml


Install with pip. pylivetrader currently supports Python 3.5, 3.6 and 3.7+

$ pip install pylivetrader

Additionally, pylivetrader works well with pipeline-live.

Command Reference


pylivetrader run starts live trading using your algorithm script. It starts by calling the initialize() function if any, and waits until the market opens. It calls the before_trading_start function if it is 8:45 ET (45 minutes before the session starts) or if it starts after that. Once the session starts, it calls the handle_data() function every minute until the session ends, or any functions that are registered by schedule_function API.

The options are as follows.

  • -f or --file: the file path to the algorithm source
  • -b or --backend: the name of backend to use
  • --backend-config: the yaml file for backend parameters
  • -s or --statefile: the file path to the persisted state file (look for the State Management section below)
  • -r or --retry: the algorithm runner continues execution in the event a general exception is raised


pylivetrader shell goes into the IPython interactive shell mode as if you are in the algorithm script namespace. It means, you can call Algorithm API such as symbol() and data.history() so you can check the behavior of each operation. At the start of shell, nothing has been called, so you may want to initialize the context by initialize(context) which would execute your initialize() function.

$ pylivetrader shell

The options are the same as run.

State Management

One of the things you need to understand in live trading is that things can happen and you may need to restart the script or the program dies in the middle of process due to some external errors. There are couple of things to know in advance.

First, pylivetrader saves the property fields to the disk that you add to the context object. It is stored in the pickle format and will be restored on the next startup.

Second, because the context properties are restored, you may need to take care of the extra steps. Often an algorithm is written under the assumption that initialize() is called only once and before_start_trading() is called once every morning. If you are to restart the program in the middle of day, these functions are called again, with the restored context object. Therefore, you might need to check if the fields are from the other session or in the same session to make sure you don't override the indermediate states in the day.

Supported Broker


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

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 --backend-config config.yaml


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, this could be it.

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

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
        'A', 10, 200,

    # modify some fields of context after `initialize(context)` is called
    _init = context._initialize
    def wrapper(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

if __name__ == '__main__':
    import logging

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.