TODO

Parallel along date axis
========================

Please see file 'orca/alpha/base' and find the '#TODO' label.

The idea is to run a **single** alpha in parallel by splitting simulation along date axis.

For example, paste the following code into a .py file:

.. code-block:: python

   from orca.mongo.quote import QuoteFetcher

   startdate, enddate = '20140101', '20141017'
   close = QuoteFetcher(datetime_index=True, reindex=True).fetch('close', startdate=startdate, enddate=enddate)

   from orca.alpha.base import BacktestingAlpha

   class MyAlpha(BacktestingAlpha):

       def generate(self, date):
           self.alphas[date] = close.ix[date]

   if __name__ == '__main__':
       alpha = MyAlpha()
       alpha.run(startdate, enddate, parallel=True)

       from orca.utils.testing import frames_equal
       assert frames_equal(alpha.get_alphas(), close)


Parallel in parameter space
===========================

See `PyAlgoTrade <http://gbeced.github.io/pyalgotrade/docs/v0.15/html/tutorial.html#optimizing>`_ for reference.

The picture I have in mind is pretty much alike:

.. code-block:: python

   from orca.mongo.quote import QuoteFetcher

   startdate, enddate = '20140101', '20141017'
   close = QuoteFetcher(datetime_index=True, reindex=True).fetch('close', startdate=startdate, enddate=enddate)

   from orca.alpha.base import BacktestingAlpha
   import time

   class MyAlpha(BacktestingAlpha):

       def __init__(self, n):
           self.n = n

       def generate(self, date):
           time.sleep(self.n % 100)
           self.alphas[date] = close.ix[date] * self.n

   def gen_params():
       for i in xrange(10000):
           yield i

   if __name__ == '__main__':
       from orca.utils.parallel import local
       import multiprocessing
       alphas = local.run(MyAlpha, 
                          gen_params, 
                          batch=multiprocessing.cpu_count(),
                          universe = XXX,
                          lambda x: x.get_original().get_ir() > 0.1)