I think only Scott can really answer if this is possible, provided you don't do the whole thing with matrix numpy operations and then just compare your [-1] signal vs your [-2] (which would be my personal recommendation anyway).

Thanks!! I will try the matrix approach to solve this particular issue. I would be interested in a solution to the general issue of having a state associated with each member of the asset universe, and then being able to use that state and current data to modify the state and make decisions based on that state.
Again, thanks for your reply.
Jim

Well you could always store the results in context and compare each time in before_trading_start...

I played with that for a little while ... just sort of enamored with the pipeline functionality. Again, thanks,
Jim

Hi Jim,

Can you elaborate a bit more on what you want your PNF factor to do? I couldn't find a definition for that term after a bit of googling.

Regarding your original question, there isn't a well-supported way to maintain/mutate state on a CustomFactor right now. If it's computationally-feasible to do so, the most straightforward way to do what you're asking within the Pipeline API would be to just recompute the previous day's value each day.

Depending on the particular transformations you're performing on the underlying data, I'd be a bit nervous about comparing Pipeline results from day N to results from day N - 1, because the underlying price/volume values are adjusted for splits/dividends/mergers from day to day, which means comparing today's computed pipeline values to yesterday's isn't always an apples-to-apples comparison.

I'd be curious to hear more about your specific use-case. I've thought a bit about how we might add a sane state-maintenance API to CustomFactors, but hearing specific use-cases helps a lot to ensure that design actually helps solve people's real problems.

Scott:
Hello. I must have invented my own acronym! When I refer to PNF, I am referring to Point and Figure Charting.
It looks at a stocks price and determines whether the stock is in a BUY or SELL mode. The way it does it is by comparing the price to a moving/ratcheting comparison value. If the stock is in a BUY mode the ratcheting value is increasing based upon the most recent high. Simplistically, I might use a value that is 97% of the most recent high. If my stock price is greater than my most recent high, I change the ratchet/comparison value to 97% of that value. If my stock price is not greater than the most recent high, but is greater than the ratchet/comparison value, I do nothing.
If my stock price is less than the ratchet/comparison value, I :
1. change the stock state to SELL
2. Set a local minimum value to the stock price
3. set the ratchet/comparison value to 1.03 times the current minimum value
Then while in the SELL state, I do the following with each new price:
1. Is it less than the current minimum? If yes calculate a new ratchet/comparison value
2. Is it greater than the current ratchet/comparison value? If yes, the change state to BUY and initialize that state
3. if both 1 and 2 are false, do nothing.
To solve this problem from the pipeline functionality point view, I need to remember three things for each stock:(1) The state (2) The most recent high or low and (3) The state transition function
In thinking about it, it would be kind of neat to have a stack functionality where I could push a set of values to a stack and then pop them off when I needed them
The reason I was looking so hard at the pipeline functionality is that it takes care of resolving that the data is there for all elements in the universe.
I banged this out pretty quickly, so I hope it makes some sense.