Files
catalyst/catalyst/exchange/live_graph_clock.py
T
Frederic Fortier 9c33ee123c BLD: Housekeeping
2018-01-06 19:51:41 -05:00

75 lines
2.6 KiB
Python

import pandas as pd
from catalyst.constants import LOG_LEVEL
from catalyst.exchange.utils.stats_utils import prepare_stats
from catalyst.gens.sim_engine import (
BAR,
SESSION_START
)
from logbook import Logger
log = Logger('LiveGraphClock', level=LOG_LEVEL)
class LiveGraphClock(object):
"""Realtime clock for live trading.
This class is a drop-in replacement for
:class:`zipline.gens.sim_engine.MinuteSimulationClock`.
This mixes the clock with a live graph.
Notes
-----
This seemingly awkward approach allows us to run the program using a single
thread. This is important because Matplotlib does not play nice with
multi-threaded environments. Zipline probably does not either.
Matplotlib has a pause() method which is a wrapper around time.sleep()
used in the SimpleClock. The key difference is that users
can still interact with the chart during the pause cycles. This is
what enables us to keep a single thread. This is also why we are not using
the 'animate' callback of Matplotlib. We need to direct access to the
__iter__ method in order to yield events to Zipline.
The :param:`time_skew` parameter represents the time difference between
the exchange and the live trading machine's clock. It's not used currently.
"""
def __init__(self, sessions, context, callback=None,
time_skew=pd.Timedelta('0s')):
self.sessions = sessions
self.time_skew = time_skew
self._last_emit = None
self._before_trading_start_bar_yielded = True
self.context = context
self.callback = callback
def __iter__(self):
from matplotlib import pyplot as plt
yield pd.Timestamp.utcnow(), SESSION_START
while True:
current_time = pd.Timestamp.utcnow()
current_minute = current_time.floor('1T')
if self._last_emit is None or current_minute > self._last_emit:
log.debug('emitting minutely bar: {}'.format(current_minute))
self._last_emit = current_minute
yield current_minute, BAR
recorded_cols = list(self.context.recorded_vars.keys())
df, _ = prepare_stats(
self.context.frame_stats, recorded_cols=recorded_cols
)
self.callback(self.context, df)
else:
# I can't use the "animate" reactive approach here because
# I need to yield from the main loop.
# Workaround: https://stackoverflow.com/a/33050617/814633
plt.pause(1)