absolute values for strategy market optimization

  • 1
  • Idea
  • Updated 8 years ago
  • Not Planned
I think the scoreboard on this page http://eterna.cmu.edu/htmls/strategy_...

should be ordered differently

stratgies with a high 'absolute' value in the ordering column should be at the top, strategies with a low absolute value should be at the bottom

thats based on the idea that a strategy worth -0.500 should make a bigger contribution to the final build (as a thing to avoid) than a strategy which has a score of 0.499 (which is a thing not to avoid) .

Anyway that then means coders optimising strategies to get your best placing on the board - ie I think we want the highest 'absolute' value to the 'ordering' result of your strategy (not the highest).

If I have a strategy that can be optimised all the way up to 0.100 or a different tweak takes it down to -0.300 I'd ratehr have the -0.300 next to my name :)
Photo of Edward Lane

Edward Lane

  • 139 Posts
  • 8 Reply Likes

Posted 8 years ago

  • 1
Photo of Jeehyung Lee

Jeehyung Lee, Alum

  • 708 Posts
  • 94 Reply Likes
Edward,

Yes - that's very good point. A strategy with highly negative value is equally important as a strategy with highly positive value.

In fact, when we create the ensemble strategy, this is automatically done - it will reverse the strategy by assigning a negative way to the strategy and use it.

We decided to display negative values in the bottom for the simplicity. Although negative strategies can be useful, players still should aim for the positive strategy (since a player can easily turn negative strategy into a positive strategy) because that is the simple goal of the strategy market - get the right ordering.
Photo of Edward Lane

Edward Lane

  • 139 Posts
  • 8 Reply Likes
I notice two of Eli's designs have a zero rating, does that mean they have zero impact on whether something works/not ?

does that mean you can freely change things in the manner Eli's strategy describes without affecting how well the RNA folds

is that significant ?
Photo of Jeehyung Lee

Jeehyung Lee, Alum

  • 708 Posts
  • 94 Reply Likes
Edward, it means those strategies cannot distinguish good designs from bad designs by themselves, based on the synthesis data we have.

It could be possible that those strategy may play more important roles when work together with other strategies.

I should point out that there is no single number we can use the define how good/bad a strategy is. A strategy should be added to the ensemble, then the ensemble must be tested on the actual synthesis data with various controls to actually figure out how good a strategy is. Numbers in the market page are only there as references, and should not be 100% relied on to determine the quality of strategies.