def rerun_existing_simulations():
total_count = MySimulationCondition.objects.all().count()
for counter, condition in enumerate(MySimulationCondition.objects.filter(strategy=2)):
logger.debug('%s: %d/%d'%(inspect.stack()[1][3], counter, total_count))
# simulation run
backtesting_simulation_consumer.delay(cPickle.dumps(condition), is_update=True)
def form_valid(self, form):
# persist simulation conditions
condition, created = MySimulationCondition.objects.get_or_create(
**form.cleaned_data)
# run simulation for 1st time
if not MyPosition.objects.filter(simulation=condition):
# MySimulationCondition is not JSON serializable
# so we use cPickle to pass object to background queue task
backtesting_simulation_consumer.delay(cPickle.dumps(condition))
return HttpResponseRedirect(reverse_lazy('simulation_result_list'))
else:
return HttpResponseRedirect(reverse_lazy('condition_detail', kwargs={'pk': condition.id}))
def batch_simulation_daily_return(date_range, strategies = [1,2], capital = 10000, per_trade = 1000):
sources = [1, 2, 3, 5]
strategy_values = [1]
# get 8211 related sectors
stock_8211 = MyStock.objects.filter(symbol__startswith="8821")
sectors = [None] + reduce(lambda x, y: x + y,
[list(s.mysector_set.all()) for s in stock_8211])
# simulation run
conditions = []
for (source, strategy, strategy_value, sector) in itertools.product(sources, strategies, strategy_values, sectors):
for (start,end) in date_range:
# we only try strategy 2 with source 1
if strategy == 2 and source != 1:
continue
# we only try strategy 5 with sector
if sector and source != 5:
continue
# simulation parameters
logger.debug(source, strategy, strategy_value, start, end, sector)
# cutoffs have different meanings based on strategy
if strategy == 1:
step = 25
buy_cutoff = range(0, 100, step)
sell_cutoff = [b + step for b in buy_cutoff]
cutoffs = zip(buy_cutoff, sell_cutoff)
elif strategy == 2:
buy_cutoff = range(1,6,1)
sell_cutoff = range(1,6,1)
cutoffs = itertools.product(buy_cutoff, sell_cutoff)
# cutoffs = [(1,1)]
# Set up simulation condition objects based on
# combination of specified parameters. Note that
# the count of this matrix increases dramatically
# if we expand this parameter list.
for (buy_cutoff, sell_cutoff) in cutoffs:
condition, created = MySimulationCondition.objects.get_or_create(
data_source=source,
sector=sector,
data_sort=1, # descending
strategy=strategy,
strategy_value=strategy_value,
start=start,
end=end,
capital=capital,
per_trade=per_trade,
buy_cutoff=buy_cutoff,
sell_cutoff=sell_cutoff
)
conditions.append(condition)
# simulation run!
for condition in conditions:
# if condition.data_source == 1 and strategy == 2:
# MySimulationCondition is not json-able,
# using python pickle instead. The downside of this is that
# we are relying on a python-specif data format.
# But it is safe in this context.
if strategy == 2:
# buy low sell high
# Set is_update=True will remove all existing results first
# and then rerun all simulations. This is necessary
# because SP500 is gettting new data each day.
backtesting_simulation_consumer.delay(
cPickle.dumps(condition), is_update=True)
else:
# alpha
# Because computing alpha index values are very time consuming,
# so we are to skip existing results to save time. Ideally
# we should set is_update=True to recompute from a clean sheet.
backtesting_simulation_consumer.delay(
cPickle.dumps(condition), is_update=False)
