QuantConnect Community

I also tried declaring the dictionary before the calling of the Universe. But same error as above.
 

    def Initialize(self):

        self.SetStartDate(2021, 1, 12)

        self.SetEndDate(2021, 1, 15)

        self.SetCash(100000) 

        self.histories = {}

        self.AddUniverseSelection(UniverseModule())

        self.UniverseSettings.Resolution = Resolution.Daily

        # self.SetAlpha(RGAlphaModel())

Hi Red Ghost ,

QuantConnect/Lean is a C# engine that supports algorithms written in Python thanks to pythonnet. It entails some limitations. For example, the QCAlgorithm parameter in AlphaModel.Update is not exactly the same object that we have defined on main as it doesn't include the user-defined attributes such as self.histories. In other words, we can only count with attributes of the base class (QCAlgorithm).

We have two alternatives:
1. Use a Global variable;
2. Pass the reference to the object with the constructor:

self.SetAlpha(RGAlphaModel(self.histories))

Let me note that both alternatives violate the separation of concerns principle. 

Hi Alexandre Catarino ,

Thanks for your answer; this demystifies the error. 

I just have a few follow-ups to your answer, if you don't mind.

1. Is there a document I can read about the limitations? Because if there are some dealbreakers, I can simply write my algorithms in C#; I just prefer Python.

2. Does what I'm doing (saving the histories) make a significant difference, or I might as well call algorithm.History() in my Alpha?  It's getting the history about 100 bars back.

Anyway I created a module SharedState.py that holds shared state. It's namespaced, so it achieves the same thing I wanted anyway; it's just attached to SharedState instead of QCAlgorithm. 

_histories = {}



def updateHistories(key, value):

   _histories[str(key)] = value

   

def getHistories():

   return _histories

   

def getHistory(key):

   return _histories[str(key)]

Hi Red Ghost,
 

Is there a document I can read about the limitations? Because if there are some dealbreakers, I can simply write my algorithms in C#; I just prefer Python.

Currently, there is no document on the limitation of Python on QC. There is as such no major limitation except mentioned by Alex in the same thread, and C# provides faster execution of the strategy. We will add the limitation section in the new docs.

Does what I'm doing (saving the histories) make a significant difference, or I might as well call algorithm.History() in my Alpha? It's getting the history about 100 bars back.

If we have a small to medium Universe, saving histories is always better than calling the history method multiple times. Multiple History calls can reduce the speed of the algorithm considerably.
Best,
Varad Kabade