pylivetrader

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.

example code

check out the examples folder. you will find there the following examples:

• simple MACD "momentum" trading
• using pipeline-live to screen top stocks every day
• a potfolio optimizer (used to optimize an existing porfolio. not to buy new stocks)

each sample code contains a readme file and a smoke runner (read further to understand what smoke is)

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
base_url: https://paper-api.alpaca.markets
feed: iex  # <== change to pro if you have a paid account

Usage with redis

If you are running pylivetrader in an environment with an ephemeral file store and need your context to persist across restarts, you can use the redis storage engine. This is useful if you launch in a place like heroku.

To use this, you must install the redis-py library.

$ pip install redis

After that everything is the same as above, except the run command looks like the following:

$ pylivetrader run -f algo.py --backend-config config.yaml --storage-engine redis

Assuming you have redis running, this will now serialize your context object to and from redis.

Installation

Install with pip. **pylivetrader currently supports only Python 3.6.

$ python3.6 -m venv venv
$ source venv/bin/activate
(venv)$ pip install pylivetrader

Additionally, pylivetrader works well with pipeline-live.

Command Reference

run

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
• --storage-engine: the storage engine to use for persisting the context. ('file' or 'redis')
• -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
• -l or --log-level: the minimum level of log which will be written ('DEBUG', 'INFO', 'WARNING', 'ERROR', or 'CRITICAL')

shell

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.

$ pylivetrader shell algo.py

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

things you could do with the shell

• get asset price data. e.g: data.history(symbol("AAPL"), "close", 10, "1d")
• check if you can trade a certain asset. e.g: data.can_trade(symbol("AAPL"))
• get your account information: context.account
• get your portfolio information: context.account.portfolio
• get all opened orders: context.get_open_orders()
• get all orders: context.get_all_orders()
• get a list of all available assets. eg: context._backend._api.list_assets(asset_class='us_equity')

migrate

pylivetrader migrate allows you to easily migrate your quantopian/zipline code to pylivetrader compatible code.
how to run:

pylivetrader migrate -i zipline_code.py -o pylivetrader_compatible.py

now you could execute it with the run command
note: we do not support the optimize api by quantopian since it is not a part of zipline

Working with Pipline-live

You can see an example usage under the examples folder.
To work with pipeline-live you need to do the following steps:

• Create the pipeline (usually you will create it in a method, convention name is make_pipeline(). Inside you will define your universe and add factors and filters.
• DO NOT store the pipe in the context object. We cannot store it to the statefile and you don't need to do it. we store it for you.
• attach the pipeline to the Algortihm instance. do it like this: context.attach_pipeline(pipe, "my_pipe"). this should be done in the intialize() or before_trading_start() methods.
• Now, to get the output of the pipeline, you do this: context.pipeline_output('my_pipe'). you should call it in handle_data() or any other method you use the scheduler for.

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_trading_start() 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 Brokers

Alpaca

Configuration by environment variables.

$ export APCA_API_KEY_ID={your api key id}
$ export APCA_API_SECRET_KEY={your api secret key}
$ export APCA_API_BASE_URL={https://api.alpaca.markets/ or https://paper-api.alpaca.markets}
$ 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}
base_url: {https://api.alpaca.markets/ or https://paper-api.alpaca.markets}
$ 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 all you need to run 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).

you could also build the docker image from source like this:
docker build -t alpaca/pylivetrader-dev -f dockerfiles/Dockerfile-dev .
it gives you the power to run it locally and edit or debug the code if you desire.

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 synthetic 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 synthetic 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 synthetic 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 paths to make sure there are no easy mistakes. This code works well with standard test frameworks such as pytest and you can easily report line coverage using those frameworks too.

Running Multiple Strategies

There's a way to execute more than one algorithm at once.
The websocket connection is limited to 1 connection per account.
For that exact purpose this was created
The steps to execute this are:

• Run the Alpaca Proxy Agent as described in the project's README
• Define this env variable: DATA_PROXY_WS to be the address of the proxy agent. (e.g: DATA_PROXY_WS=ws://127.0.0.1:8765)
• execute your algorithm. it will connect to the servers through the proxy agent allowing you to execute multiple strategies