def get_generator(self):
if self.trading_client is not None:
return self.trading_client.transform()
perf = None
if self.perf_tracker is None:
tracker = self.perf_tracker = PerformanceTracker(
sim_params=self.sim_params,
trading_calendar=self.trading_calendar,
env=self.trading_environment,
)
# Set the dt initially to the period start by forcing it to change.
self.on_dt_changed(self.sim_params.start_session)
# Unpacking the perf_tracker and positions if available
perf = get_algo_object(
algo_name=self.algo_namespace,
key='cumulative_performance',
)
if not self.initialized:
self.initialize(*self.initialize_args, **self.initialize_kwargs)
self.initialized = True
# Call the simulation trading algorithm for side-effects:
# it creates the perf tracker
# TradingAlgorithm._create_generator(self, self.sim_params)
if perf is not None:
tracker.cumulative_performance = perf
period = self.perf_tracker.todays_performance
period.starting_cash = perf.ending_cash
period.starting_exposure = perf.ending_exposure
period.starting_value = perf.ending_value
period.position_tracker = perf.position_tracker
self.trading_client = ExchangeAlgorithmExecutor(
algo=self,
sim_params=self.sim_params,
data_portal=self.data_portal,
clock=self.clock,
benchmark_source=self._create_benchmark_source(),
restrictions=self.restrictions,
universe_func=self._calculate_universe,
)
return self.trading_client.transform()
def _init_trading_client(self):
"""
This replaces Ziplines `_create_generator` method. The main difference
is that we are restoring performance tracker objects if available.
This allows us to stop/start algos without loosing their state.
"""
self.state = get_algo_object(
algo_name=self.algo_namespace,
key='context.state_{}'.format(self.mode_name),
)
if self.state is None:
self.state = {}
if self.perf_tracker is None:
# Note from the Zipline dev:
# HACK: When running with the `run` method, we set perf_tracker to
# None so that it will be overwritten here.
tracker = self.perf_tracker = PerformanceTracker(
sim_params=self.sim_params,
trading_calendar=self.trading_calendar,
env=self.trading_environment,
)
# Set the dt initially to the period start by forcing it to change.
self.on_dt_changed(self.sim_params.start_session)
new_position_tracker = tracker.position_tracker
tracker.position_tracker = None
# Unpacking the perf_tracker and positions if available
cum_perf = get_algo_object(
algo_name=self.algo_namespace,
key='cumulative_performance_{}'.format(self.mode_name),
)
if cum_perf is not None:
tracker.cumulative_performance = cum_perf
# Ensure single common position tracker
tracker.position_tracker = cum_perf.position_tracker
today = pd.Timestamp.utcnow().floor('1D')
todays_perf = get_algo_object(
algo_name=self.algo_namespace,
key=today.strftime('%Y-%m-%d'),
rel_path='daily_performance_{}'.format(self.mode_name),
)
if todays_perf is not None:
# Ensure single common position tracker
if tracker.position_tracker is not None:
todays_perf.position_tracker = tracker.position_tracker
else:
tracker.position_tracker = todays_perf.position_tracker
tracker.todays_performance = todays_perf
if tracker.position_tracker is None:
# Use a new position_tracker if not is found in the state
tracker.position_tracker = new_position_tracker
if not self.initialized:
# Calls the initialize function of the algorithm
self.initialize(*self.initialize_args, **self.initialize_kwargs)
self.initialized = True
self.trading_client = ExchangeAlgorithmExecutor(
algo=self,
sim_params=self.sim_params,
data_portal=self.data_portal,
clock=self.clock,
benchmark_source=self._create_benchmark_source(),
restrictions=self.restrictions,
universe_func=self._calculate_universe,
)
