I am trying to build a QSAR model with 47 compounds. I have more or less 600 descriptors. I am building the QSAR model with QSARiN. The genetic algorithm application is asking for mutation rate, population size and iteration number. If i choose high numbers (i.e ; iteration:3000, pop. size: 500, mutation rate:80)the calculation needs too much time to complete. How can i decide optimum values for a reliable genetic algorithm calculation?
Hi Sedat,
There is no rules to choose optimum values for a reliable genetic algorithm. In fact, to reduce your iteration number, you can follow the diversity of the population or the cost function during the genetic algorithm process. When these quantities are stable (do not change during long time), it is not necessary to continu your optimization or identification and you get the maximum iteration number. But 3000 iterations seems to be too large. You can increase your population size (ex: 1000). Moreover, the mutation rate has to be between 0.01 and 0.1 and the crossover rate can be between 0.8 and 0.9. Nevertheless, in some specific cases, the mutation rate can reach 0.3 or 0.4. It is used when the diversity of the population is poor.
Another advice, try to use different genetic algorithm such as NSGA II. It is a multi-objective genetic algorithm.
I hope these information help you.
Bye
- Badges
- Science topic
- Similar topics
- Chemistry