mirror of
https://github.com/wassname/catalyst.git
synced 2026-07-26 13:18:31 +08:00
Eddie Hebert
7a1a6ddb37
Instead of using the pandas.Series datetime index for every single vector, get the index at the beginning of the update loop based on the dt and then use that index to set the values. Also, since the dt lookup is no longer needed, store the values as numpy arrays, which are more lightweight. Locally, this patch cuts out about 60% of the time spent in the update method.
540 lines
21 KiB
Python
540 lines
21 KiB
Python
#
|
|
# Copyright 2015 Quantopian, Inc.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
"""
|
|
|
|
Performance Tracking
|
|
====================
|
|
|
|
+-----------------+----------------------------------------------------+
|
|
| key | value |
|
|
+=================+====================================================+
|
|
| period_start | The beginning of the period to be tracked. datetime|
|
|
| | in pytz.utc timezone. Will always be 0:00 on the |
|
|
| | date in UTC. The fact that the time may be on the |
|
|
| | prior day in the exchange's local time is ignored |
|
|
+-----------------+----------------------------------------------------+
|
|
| period_end | The end of the period to be tracked. datetime |
|
|
| | in pytz.utc timezone. Will always be 23:59 on the |
|
|
| | date in UTC. The fact that the time may be on the |
|
|
| | next day in the exchange's local time is ignored |
|
|
+-----------------+----------------------------------------------------+
|
|
| progress | percentage of test completed |
|
|
+-----------------+----------------------------------------------------+
|
|
| capital_base | The initial capital assumed for this tracker. |
|
|
+-----------------+----------------------------------------------------+
|
|
| cumulative_perf | A dictionary representing the cumulative |
|
|
| | performance through all the events delivered to |
|
|
| | this tracker. For details see the comments on |
|
|
| | :py:meth:`PerformancePeriod.to_dict` |
|
|
+-----------------+----------------------------------------------------+
|
|
| todays_perf | A dictionary representing the cumulative |
|
|
| | performance through all the events delivered to |
|
|
| | this tracker with datetime stamps between last_open|
|
|
| | and last_close. For details see the comments on |
|
|
| | :py:meth:`PerformancePeriod.to_dict` |
|
|
| | TODO: adding this because we calculate it. May be |
|
|
| | overkill. |
|
|
+-----------------+----------------------------------------------------+
|
|
| cumulative_risk | A dictionary representing the risk metrics |
|
|
| _metrics | calculated based on the positions aggregated |
|
|
| | through all the events delivered to this tracker. |
|
|
| | For details look at the comments for |
|
|
| | :py:meth:`zipline.finance.risk.RiskMetrics.to_dict`|
|
|
+-----------------+----------------------------------------------------+
|
|
|
|
"""
|
|
|
|
from __future__ import division
|
|
import logbook
|
|
import pickle
|
|
from six import iteritems
|
|
from datetime import datetime
|
|
|
|
import numpy as np
|
|
import pandas as pd
|
|
from pandas.tseries.tools import normalize_date
|
|
|
|
import zipline.finance.risk as risk
|
|
from zipline.finance.trading import TradingEnvironment
|
|
from . period import PerformancePeriod
|
|
|
|
from zipline.utils.serialization_utils import (
|
|
VERSION_LABEL
|
|
)
|
|
from . position_tracker import PositionTracker
|
|
|
|
log = logbook.Logger('Performance')
|
|
|
|
|
|
class PerformanceTracker(object):
|
|
"""
|
|
Tracks the performance of the algorithm.
|
|
"""
|
|
def __init__(self, sim_params):
|
|
|
|
self.sim_params = sim_params
|
|
env = TradingEnvironment.instance()
|
|
|
|
self.period_start = self.sim_params.period_start
|
|
self.period_end = self.sim_params.period_end
|
|
self.last_close = self.sim_params.last_close
|
|
first_open = self.sim_params.first_open.tz_convert(env.exchange_tz)
|
|
self.day = pd.Timestamp(datetime(first_open.year, first_open.month,
|
|
first_open.day), tz='UTC')
|
|
self.market_open, self.market_close = env.get_open_and_close(self.day)
|
|
self.total_days = self.sim_params.days_in_period
|
|
self.capital_base = self.sim_params.capital_base
|
|
self.emission_rate = sim_params.emission_rate
|
|
|
|
all_trading_days = env.trading_days
|
|
mask = ((all_trading_days >= normalize_date(self.period_start)) &
|
|
(all_trading_days <= normalize_date(self.period_end)))
|
|
|
|
self.trading_days = all_trading_days[mask]
|
|
|
|
self.dividend_frame = pd.DataFrame()
|
|
self._dividend_count = 0
|
|
|
|
self.position_tracker = PositionTracker()
|
|
|
|
self.perf_periods = []
|
|
|
|
if self.emission_rate == 'daily':
|
|
self.all_benchmark_returns = pd.Series(
|
|
index=self.trading_days)
|
|
self.cumulative_risk_metrics = \
|
|
risk.RiskMetricsCumulative(self.sim_params)
|
|
|
|
elif self.emission_rate == 'minute':
|
|
self.all_benchmark_returns = pd.Series(index=pd.date_range(
|
|
self.sim_params.first_open, self.sim_params.last_close,
|
|
freq='Min'))
|
|
|
|
self.cumulative_risk_metrics = \
|
|
risk.RiskMetricsCumulative(self.sim_params,
|
|
returns_frequency='daily',
|
|
create_first_day_stats=True)
|
|
|
|
self.minute_performance = PerformancePeriod(
|
|
# initial cash is your capital base.
|
|
self.capital_base,
|
|
# the cumulative period will be calculated over the
|
|
# entire test.
|
|
self.period_start,
|
|
self.period_end,
|
|
# don't save the transactions for the cumulative
|
|
# period
|
|
keep_transactions=False,
|
|
keep_orders=False,
|
|
# don't serialize positions for cumualtive period
|
|
serialize_positions=False
|
|
)
|
|
self.minute_performance.position_tracker = self.position_tracker
|
|
self.perf_periods.append(self.minute_performance)
|
|
|
|
# this performance period will span the entire simulation from
|
|
# inception.
|
|
self.cumulative_performance = PerformancePeriod(
|
|
# initial cash is your capital base.
|
|
self.capital_base,
|
|
# the cumulative period will be calculated over the entire test.
|
|
self.period_start,
|
|
self.period_end,
|
|
# don't save the transactions for the cumulative
|
|
# period
|
|
keep_transactions=False,
|
|
keep_orders=False,
|
|
# don't serialize positions for cumualtive period
|
|
serialize_positions=False,
|
|
)
|
|
self.cumulative_performance.position_tracker = self.position_tracker
|
|
self.perf_periods.append(self.cumulative_performance)
|
|
|
|
# this performance period will span just the current market day
|
|
self.todays_performance = PerformancePeriod(
|
|
# initial cash is your capital base.
|
|
self.capital_base,
|
|
# the daily period will be calculated for the market day
|
|
self.market_open,
|
|
self.market_close,
|
|
keep_transactions=True,
|
|
keep_orders=True,
|
|
serialize_positions=True,
|
|
)
|
|
self.todays_performance.position_tracker = self.position_tracker
|
|
|
|
self.perf_periods.append(self.todays_performance)
|
|
|
|
self.saved_dt = self.period_start
|
|
# one indexed so that we reach 100%
|
|
self.day_count = 0.0
|
|
self.txn_count = 0
|
|
|
|
self.account_needs_update = True
|
|
self._account = None
|
|
|
|
def __repr__(self):
|
|
return "%s(%r)" % (
|
|
self.__class__.__name__,
|
|
{'simulation parameters': self.sim_params})
|
|
|
|
@property
|
|
def progress(self):
|
|
if self.emission_rate == 'minute':
|
|
# Fake a value
|
|
return 1.0
|
|
elif self.emission_rate == 'daily':
|
|
return self.day_count / self.total_days
|
|
|
|
def set_date(self, date):
|
|
if self.emission_rate == 'minute':
|
|
self.saved_dt = date
|
|
self.todays_performance.period_close = self.saved_dt
|
|
|
|
def update_dividends(self, new_dividends):
|
|
"""
|
|
Update our dividend frame with new dividends. @new_dividends should be
|
|
a DataFrame with columns containing at least the entries in
|
|
zipline.protocol.DIVIDEND_FIELDS.
|
|
"""
|
|
|
|
# Mark each new dividend with a unique integer id. This ensures that
|
|
# we can differentiate dividends whose date/sid fields are otherwise
|
|
# identical.
|
|
new_dividends['id'] = np.arange(
|
|
self._dividend_count,
|
|
self._dividend_count + len(new_dividends),
|
|
)
|
|
self._dividend_count += len(new_dividends)
|
|
|
|
self.dividend_frame = pd.concat(
|
|
[self.dividend_frame, new_dividends]
|
|
).sort(['pay_date', 'ex_date']).set_index('id', drop=False)
|
|
|
|
def initialize_dividends_from_other(self, other):
|
|
"""
|
|
Helper for copying dividends to a new PerformanceTracker while
|
|
preserving dividend count. Useful if a simulation needs to create a
|
|
new PerformanceTracker mid-stream and wants to preserve stored dividend
|
|
info.
|
|
|
|
Note that this does not copy unpaid dividends.
|
|
"""
|
|
self.dividend_frame = other.dividend_frame
|
|
self._dividend_count = other._dividend_count
|
|
|
|
def update_performance(self):
|
|
# calculate performance as of last trade
|
|
for perf_period in self.perf_periods:
|
|
perf_period.calculate_performance()
|
|
|
|
def get_portfolio(self, performance_needs_update):
|
|
if performance_needs_update:
|
|
self.update_performance()
|
|
self.account_needs_update = True
|
|
return self.cumulative_performance.as_portfolio()
|
|
|
|
def get_account(self, performance_needs_update):
|
|
if performance_needs_update:
|
|
self.update_performance()
|
|
self.account_needs_update = True
|
|
if self.account_needs_update:
|
|
self._update_account()
|
|
return self._account
|
|
|
|
def _update_account(self):
|
|
self._account = self.cumulative_performance.as_account()
|
|
self.account_needs_update = False
|
|
|
|
def to_dict(self, emission_type=None):
|
|
"""
|
|
Creates a dictionary representing the state of this tracker.
|
|
Returns a dict object of the form described in header comments.
|
|
"""
|
|
if not emission_type:
|
|
emission_type = self.emission_rate
|
|
_dict = {
|
|
'period_start': self.period_start,
|
|
'period_end': self.period_end,
|
|
'capital_base': self.capital_base,
|
|
'cumulative_perf': self.cumulative_performance.to_dict(),
|
|
'progress': self.progress,
|
|
'cumulative_risk_metrics': self.cumulative_risk_metrics.to_dict()
|
|
}
|
|
if emission_type == 'daily':
|
|
_dict['daily_perf'] = self.todays_performance.to_dict()
|
|
elif emission_type == 'minute':
|
|
_dict['minute_perf'] = self.todays_performance.to_dict(
|
|
self.saved_dt)
|
|
|
|
return _dict
|
|
|
|
def process_trade(self, event):
|
|
# update last sale, and pay out a cash adjustment
|
|
cash_adjustment = self.position_tracker.update_last_sale(event)
|
|
if cash_adjustment != 0:
|
|
for perf_period in self.perf_periods:
|
|
perf_period.handle_cash_payment(cash_adjustment)
|
|
|
|
def process_transaction(self, event):
|
|
self.txn_count += 1
|
|
self.position_tracker.execute_transaction(event)
|
|
for perf_period in self.perf_periods:
|
|
perf_period.handle_execution(event)
|
|
|
|
def process_dividend(self, dividend):
|
|
|
|
log.info("Ignoring DIVIDEND event.")
|
|
|
|
def process_split(self, event):
|
|
leftover_cash = self.position_tracker.handle_split(event)
|
|
if leftover_cash > 0:
|
|
for perf_period in self.perf_periods:
|
|
perf_period.handle_cash_payment(leftover_cash)
|
|
|
|
def process_order(self, event):
|
|
for perf_period in self.perf_periods:
|
|
perf_period.record_order(event)
|
|
|
|
def process_commission(self, event):
|
|
|
|
self.position_tracker.handle_commission(event)
|
|
for perf_period in self.perf_periods:
|
|
perf_period.handle_commission(event)
|
|
|
|
def process_benchmark(self, event):
|
|
if self.sim_params.data_frequency == 'minute' and \
|
|
self.sim_params.emission_rate == 'daily':
|
|
# Minute data benchmarks should have a timestamp of market
|
|
# close, so that calculations are triggered at the right time.
|
|
# However, risk module uses midnight as the 'day'
|
|
# marker for returns, so adjust back to midnight.
|
|
midnight = pd.tseries.tools.normalize_date(event.dt)
|
|
else:
|
|
midnight = event.dt
|
|
|
|
if midnight not in self.all_benchmark_returns.index:
|
|
raise AssertionError(
|
|
("Date %s not allocated in all_benchmark_returns. "
|
|
"Calendar seems to mismatch with benchmark. "
|
|
"Benchmark container is=%s" %
|
|
(midnight,
|
|
self.all_benchmark_returns.index)))
|
|
|
|
self.all_benchmark_returns[midnight] = event.returns
|
|
|
|
def process_close_position(self, event):
|
|
|
|
# CLOSE_POSITION events contain prices that must be handled as a final
|
|
# trade event
|
|
self.process_trade(event)
|
|
|
|
txn = self.position_tracker.create_close_position_transaction(event)
|
|
if txn:
|
|
self.process_transaction(txn)
|
|
|
|
def check_upcoming_dividends(self, completed_date):
|
|
"""
|
|
Check if we currently own any stocks with dividends whose ex_date is
|
|
the next trading day. Track how much we should be payed on those
|
|
dividends' pay dates.
|
|
|
|
Then check if we are owed cash/stock for any dividends whose pay date
|
|
is the next trading day. Apply all such benefits, then recalculate
|
|
performance.
|
|
"""
|
|
if len(self.dividend_frame) == 0:
|
|
# We don't currently know about any dividends for this simulation
|
|
# period, so bail.
|
|
return
|
|
|
|
# Get the next trading day and, if it is outside the bounds of the
|
|
# simulation, bail.
|
|
next_trading_day = TradingEnvironment.instance().\
|
|
next_trading_day(completed_date)
|
|
if (next_trading_day is None) or (next_trading_day >= self.last_close):
|
|
return
|
|
|
|
# Dividends whose ex_date is the next trading day. We need to check if
|
|
# we own any of these stocks so we know to pay them out when the pay
|
|
# date comes.
|
|
ex_date_mask = (self.dividend_frame['ex_date'] == next_trading_day)
|
|
dividends_earnable = self.dividend_frame[ex_date_mask]
|
|
|
|
# Dividends whose pay date is the next trading day. If we held any of
|
|
# these stocks on midnight before the ex_date, we need to pay these out
|
|
# now.
|
|
pay_date_mask = (self.dividend_frame['pay_date'] == next_trading_day)
|
|
dividends_payable = self.dividend_frame[pay_date_mask]
|
|
|
|
position_tracker = self.position_tracker
|
|
if len(dividends_earnable):
|
|
position_tracker.earn_dividends(dividends_earnable)
|
|
|
|
if not len(dividends_payable):
|
|
return
|
|
|
|
net_cash_payment = position_tracker.pay_dividends(dividends_payable)
|
|
|
|
for period in self.perf_periods:
|
|
# notify periods to update their stats
|
|
period.handle_dividends_paid(net_cash_payment)
|
|
|
|
def handle_minute_close(self, dt):
|
|
self.update_performance()
|
|
todays_date = normalize_date(dt)
|
|
account = self.get_account(False)
|
|
|
|
self.minute_performance.rollover()
|
|
|
|
bench_returns = self.all_benchmark_returns.loc[todays_date:dt]
|
|
# cumulative returns
|
|
bench_since_open = (1. + bench_returns).prod() - 1
|
|
|
|
self.cumulative_risk_metrics.update(todays_date,
|
|
self.todays_performance.returns,
|
|
bench_since_open,
|
|
account)
|
|
|
|
# if this is the close, update dividends for the next day.
|
|
if dt == self.market_close:
|
|
self.check_upcoming_dividends(todays_date)
|
|
|
|
def handle_intraday_market_close(self, new_mkt_open, new_mkt_close):
|
|
"""
|
|
Function called at market close only when emitting at minutely
|
|
frequency.
|
|
"""
|
|
# increment the day counter before we move markers forward.
|
|
self.day_count += 1.0
|
|
self.market_open = new_mkt_open
|
|
self.market_close = new_mkt_close
|
|
|
|
def handle_market_close_daily(self):
|
|
"""
|
|
Function called after handle_data when running with daily emission
|
|
rate.
|
|
"""
|
|
self.update_performance()
|
|
completed_date = self.day
|
|
account = self.get_account(False)
|
|
|
|
# update risk metrics for cumulative performance
|
|
self.cumulative_risk_metrics.update(
|
|
completed_date,
|
|
self.todays_performance.returns,
|
|
self.all_benchmark_returns[completed_date],
|
|
account)
|
|
|
|
# increment the day counter before we move markers forward.
|
|
self.day_count += 1.0
|
|
|
|
# Take a snapshot of our current performance to return to the
|
|
# browser.
|
|
daily_update = self.to_dict()
|
|
|
|
# On the last day of the test, don't create tomorrow's performance
|
|
# period. We may not be able to find the next trading day if we're at
|
|
# the end of our historical data
|
|
if self.market_close >= self.last_close:
|
|
return daily_update
|
|
|
|
# move the market day markers forward
|
|
env = TradingEnvironment.instance()
|
|
self.market_open, self.market_close = \
|
|
env.next_open_and_close(self.day)
|
|
self.day = env.next_trading_day(self.day)
|
|
|
|
# Roll over positions to current day.
|
|
self.todays_performance.rollover()
|
|
self.todays_performance.period_open = self.market_open
|
|
self.todays_performance.period_close = self.market_close
|
|
|
|
# Check for any dividends
|
|
self.check_upcoming_dividends(completed_date)
|
|
|
|
return daily_update
|
|
|
|
def handle_simulation_end(self):
|
|
"""
|
|
When the simulation is complete, run the full period risk report
|
|
and send it out on the results socket.
|
|
"""
|
|
|
|
log_msg = "Simulated {n} trading days out of {m}."
|
|
log.info(log_msg.format(n=int(self.day_count), m=self.total_days))
|
|
log.info("first open: {d}".format(
|
|
d=self.sim_params.first_open))
|
|
log.info("last close: {d}".format(
|
|
d=self.sim_params.last_close))
|
|
|
|
bms = pd.Series(
|
|
index=self.cumulative_risk_metrics.cont_index,
|
|
data=self.cumulative_risk_metrics.benchmark_returns_cont)
|
|
ars = pd.Series(
|
|
index=self.cumulative_risk_metrics.cont_index,
|
|
data=self.cumulative_risk_metrics.algorithm_returns_cont)
|
|
acl = self.cumulative_risk_metrics.algorithm_cumulative_leverages
|
|
self.risk_report = risk.RiskReport(
|
|
ars,
|
|
self.sim_params,
|
|
benchmark_returns=bms,
|
|
algorithm_leverages=acl)
|
|
|
|
risk_dict = self.risk_report.to_dict()
|
|
return risk_dict
|
|
|
|
def __getstate__(self):
|
|
state_dict = \
|
|
{k: v for k, v in iteritems(self.__dict__)
|
|
if not k.startswith('_')}
|
|
|
|
state_dict['dividend_frame'] = pickle.dumps(self.dividend_frame)
|
|
|
|
state_dict['_dividend_count'] = self._dividend_count
|
|
|
|
# we already store perf periods as attributes
|
|
del state_dict['perf_periods']
|
|
|
|
STATE_VERSION = 3
|
|
state_dict[VERSION_LABEL] = STATE_VERSION
|
|
|
|
return state_dict
|
|
|
|
def __setstate__(self, state):
|
|
|
|
OLDEST_SUPPORTED_STATE = 3
|
|
version = state.pop(VERSION_LABEL)
|
|
|
|
if version < OLDEST_SUPPORTED_STATE:
|
|
raise BaseException("PerformanceTracker saved state is too old.")
|
|
|
|
self.__dict__.update(state)
|
|
|
|
# Handle the dividend frame specially
|
|
self.dividend_frame = pickle.loads(state['dividend_frame'])
|
|
|
|
# properly setup the perf periods
|
|
self.perf_periods = []
|
|
p_types = ['cumulative', 'todays', 'minute']
|
|
for p_type in p_types:
|
|
name = p_type + '_performance'
|
|
period = getattr(self, name, None)
|
|
if period is None:
|
|
continue
|
|
period._position_tracker = self.position_tracker
|
|
self.perf_periods.append(period)
|