Hello Franklin,

This is a solution with a queue. I would like to see some others.

P.

This uses a batch transform.

P.

Hello Franklin (and Peter),

Perhaps not the most elegant solution, but you can create your own data accumulator by creating an empty context variable. Then, just append the data to it, as in the attached backtest. Here's the output, illustrating the accumulation of the price:

2008-01-04PRINT[179.93]  
2008-01-07PRINT[179.93, 177.58]  
2008-01-08PRINT[179.93, 177.58, 171.23]  
2008-01-09PRINT[179.93, 177.58, 171.23, 179.5]  
2008-01-10PRINT[179.93, 177.58, 171.23, 179.5, 178.02]  
2008-01-11PRINT[179.93, 177.58, 171.23, 179.5, 178.02, 172.4]  

Grant

Hello Grant,

As elegant as any other so far!

P.

I think using deque is the best way currently. If you supply the maxlen argument you don't have to handle removing events from the window, i.e.

# in initialize  
context.window = deque(maxlen=30)  
...
# append to window  
context.window.append(data)  
# access data from 30 days ago  
context.window[0]  

Having said that, we recently came up with a new design that will make this trivial on Quantopian. Stay tuned!

Thomas,

I haven't tried deques yet. Would you expect a deque to be faster than the batch transform?

Grant

In this particular instance -- yes. deque does not need a continuous area of memory and in this case you just need to hold on to that last value.

batch_transform will be faster in the case where you do need a continuous array.

An easy way to remember this is: if you at any point you turn data into a np.array or np.matrix or pandas.DataFrame, use a batch_transform which is optimized for this case and gives you a panel directly (which you can slice to a dataframe easily and without copying memory), while a list or deque will have to be copied internally which is quite costly. If you don't need an array, deque (or list) will be faster.