I would suggest encapsulating most (or ideally all) of your stock selection logic into a pipeline definition. It sounds like you have done this. The typical workflow would be this:

• Create a pipeline factor which you intend to trade on. This could be a simple 1 or 0 for hold or not hold, or a number such as RSI.
  
• Use Alphalens to analyze the results of this factor. Alphalens runs much faster than a backtest and, moreover, it can analyze a number of groups of factor values at one time. For example, one could find the optimum RSI values at which to buy or sell. Instead of setting up a filter for RSI > max_rsi and then varying max_rsi, one could simply output the RSI value in the factor. In Alphalens, set buckets for 10, 20, ... 80, 90, 100. Aphalens will then show the n-day return for each bucket. Optimize your max_rsi value based on returns for each bucket.
  
• Copy the factor to an algo in the IDE. Add simple logic to trade on the signal. Trade every day at open to get results similar to Alphalens.
  
• After the backtest has run, use the get_backtest method in a notebook to get the backtest results. All the backtest results are then available to analyze in the notebook environment.

Using the get_backtest method and then the associated AlgorithmResult attributes, one can get the desired backtest metrics. There is info on all the data one can fetch in the docs https://www.quantopian.com/docs/api-reference/research-api-reference#quick-reference

@DanWhitnable

Thanks for the quick response! So I can use Alphalens for backtesting in the research notebooks? The issue I'm having (in part at least) is most of my logic can be kept inside the pipeline, but certain methods can't - eg. only sell if price is above/below X, or can that be done in pipeline? for now I've hard coded that logic using if statements. Maybe there's a way around this?

I'm not too concerned with using get_backtest in the notebook. For the first pass at looking at optimising values I was purely looking at returns % since there is a minimum returns I'm aiming for (which should be easy to get in the notebook backtest) - after that I'll fine tune to handle other attributes!

I'll look into using Alphalens, thanks!