How do you manage huge optimizations?

• 08/07/2020 at 6:00 PM #141105
  Francesco

  Hi guys, i have a system with 13 variables and i’m stuck on how to optimize them without influencing each other through the process.

  If i would like to optimize them all togheter i’ll need multiple days and multiple optimizations of decreasing intervals; if i would like to optimize one of two at the same time, the next ones will be based on the previous optimizations.
  I will never have “an impartial result” without leaving the computer running for days working on one system?

  How do you manage high variables systems?

  

  Francesco

  Participant

  Topics: 32

  Replies: 439

  Been thanked: 200 times
  08/07/2020 at 7:36 PM #141114
  GraHal

  Coarse / wide steps then successively tighter / smaller steps so 25 100 25 then if the result you want to use is for example 50 then use 15 75 15 then if the result you want to use is for example 30 then use 15 30 5 … or some similar reduction of the number of combinations.

  

  GraHal

  Participant

  Topics: 203

  Replies: 8375

  Been thanked: 2024 times
  08/07/2020 at 8:00 PM #141115
  Francesco

  The problem is that also with wide steps its going to take days in order to complete a single system optimization…. it will be better to start now so 🙂

  

  Francesco

  Participant

  Topics: 32

  Replies: 439

  Been thanked: 200 times
  08/07/2020 at 10:34 PM #141129
  AlgoAlex

  with 13 variables you can’t avoid overfitting…

  Could you minimize the number of variables? If not, why?

  

  AlgoAlex

  Participant

  Topics: 33

  Replies: 64

  Been thanked: 25 times
  08/08/2020 at 10:46 AM #141151
  Francesco

  AlexF wrote:

  with 13 variables you can’t avoid overfitting…

  Yet with this complex system I obtained very high robustness scores like never before, nothing is certain 🙂

  

  Francesco

  Participant

  Topics: 32

  Replies: 439

  Been thanked: 200 times
  08/12/2020 at 4:24 PM #141485
  zilliq

  Agree with Alex,

  The more variables you have the more overfitt you do and the more difference you will have between backtest and the future

  2-3 variables is the maximum

  Zilliq

  

  zilliq

  Participant

  Topics: 58

  Replies: 683

  Been thanked: 152 times
  08/12/2020 at 6:04 PM #141489
  Francesco

  As I told to alex nothing is certain in this world, you may be surprised by the potential of more complex systems as well as simple ones.

  The real mistake would be excluding one of the two ways 🙂

  

  Francesco

  Participant

  Topics: 32

  Replies: 439

  Been thanked: 200 times
  08/13/2020 at 8:10 PM #141536
  jebus89

  You could try to optimize big filters first, then the signals. This depends on how the big the big filters are etc

  1 user thanked author for this post.

  Francesco

  

  jebus89

  Participant

  Topics: 118

  Replies: 586

  Been thanked: 238 times
  08/21/2020 at 3:21 PM #142222
  Dow Jones

  When there is a lot of variables, I optimize them a bit like multitask, e.g. the Moving Average type from 0 to 6, I define it as a variable then run 6 back tests in parallel each for 1 type of moving average. Wait for few hours for all the 6 results, then pick the one I’m interested, fixed that variable to continue.

  In case of variables from different time frame, I try to treat them separately. I would comment out some filters from certain time frame temporary and optimize others first, then add them back piece by piece incremental after some variables are optimized with OOS and satisfied to define with a static value.

  2 users thanked author for this post.

  Kovit, Francesco

  

  Dow Jones

  Participant

  Topics: 10

  Replies: 135

  Been thanked: 75 times
• Author
  Posts