You can make a variable context.win_length = 10 in the function def initialize(context):

Then in your other functions you can access the context.win_length variable. Adding the "context." to the beginning and putting within the initialize function makes it a global variable.

def initialize(context):  
    context.WIN_LENGTH = 10  
def make_pipeline():

  mean_close_XX = SimpleMovingAverage(inputs=[USEquityPricing.close], window_length=context.WIN_LENGTH)  
  log.info('Your window length is %d' % context.WIN_LENGTH)

  return Pipeline(  
    columns={  
      'XX_day_mean_close': mean_close_XX  
    }  
  )  
def before_trading_start(context, data):  
    context.WIN_LENGTH  
    context.WIN_LENGTH = 100  
    po = pipeline_output('my_pipeline')  

Hope that solves your issue.
Best

Why thank you for your response! I gave it a worthy try but as expected I got an error:

NameError: global name 'context' is not defined.. in make_pipeline

I think the issue is that normally, I'd be able to use context but make_pipeline() does not take that session object as an argument and so it does not recognize it within that scope. Of course, not so with the other two functions initialize(context) and before_trading_start(context, data).

Maybe there is another easy solution.

Hi Tariq,

make_pipeline is not a pre-defined function. It's defined in the algorithm, so you can simply specify that you want context to be a parameter in the definition, and pass it as an argument when you call make_pipeline.

For example, you could do something like this:

def initialize(context):  
    context.my_var = 10  
    ...  
    my_pipe = make_pipeline(context)

def make_pipeline(context):  
    print context.my_var  
    ...  
    return Pipeline( ... )  

Does this help?

Hi Jamie, interesting. I knew that make_pipeline was a custom function but was not clear on how to pass additional arguments when calling attach_pipeline. Thanks for the suggestion. Being able to use the context variable in a pipeline is a powerful feature.

However, I gave the suggestion a go and then created a plain vanilla momentum strategy that uses a shorter and shorter window as time passes by, decrementing the window by 1 day each day until it hits 10 days. The attached example, logs only the initial context.DYN_WINDOW, indicating that make_pipeline gets called only once, which is surprising to me. I thought it was calculated every day. Isn't it?

Hi Tariq,

Ah, I see what you are trying to do now. Yes, pipeline is run for every day in the simulation, but it is defined in initialize (in make_pipeline). It is defined only once (as initialize is run only once) so you won't be able to make state-dependent changes.

As a workaround, you could instead create a CustomFactor that changes its window length based on the value of today in the compute function.

If you haven't been through it already, I recommend going through the Pipeline Tutorial. Lesson 10 specifically covers CustomFactors which should help.