It's important that we keep the factor files consistent with respect to
the date that they were generated. This enables us to run the regression
algorithms in the cloud and get the same results by using the factor files
from the correct date.
A mechanical refactoring was performed to make algorithms currently used in
regression algorithms to implement IRegressionAlgorithmDefinition, which allows
algorithms to define their own expected statistics and what languages should be
run as part of regression. The type name of the C# type is used to determine the
file/model name for python. This was for simplicity, but if needed, could later be
refactored to expose more information, but for now the convention of keeping names
the same makes sense and just works easily.
The engine defines securities for each universe to properly track them within
the data feed. These securities are not tradable and have no price data associated
with them, and as such, we should not be sending history requests for these symbols.
This change removes all universe symbols from history requests.
NOTE: Requests made directly to the history provider are not filtered out, as the
filtering happens within the QCAlgorithm implementation.
On June 30, 2017, GOOG and YAHOO databases have been deleted from the Quandl site, so now History requests for these symbols are returning empty lists and regression test is failing.
The HistoryAlgorithm has been updated to use the CHRIS/CME_SP1 ticker instead of YAHOO/INDEX_SPY.
Non-generic history methods returning IEnumerable<TradeBar> have been marked as obsolete because they return an empty result when called with Forex and CFD security types.
The recommended replacements are their generic equivalents, using TradeBar or QuoteBar data type arguments, depending on the security type.
Fixes bug in backtest history provider
Requests for bar counts were always incorrect because the history provider
was emitting bars whose end time is equal to the start time.
Implemented a FilterEnumerator<T> privately so he can filter those out
when using non-tick data
Michael Handschuh