I have 10 earthquakes, with two input variables time and acceleration(pga), a total of 50,000 instances. How to decide which instances are to be considered for training and testing?
ex. Earthquake 1 has T vs PGA starting from 0 sec to 99 sec,total of 5000 instances, Earthquake 2 has T vs PGA from 0 to 65 sec, total of 3800 instances.. and so on..so total of 50,000 instances? so how i gonna choose data instances for training? Shall i use sub.clustering?? Please help
Thank you
Learn cross validation and apply using machine learning classifiers.
Learn the concept of cross validation technique which is the effective methodology.
Ayyanathan
First make a fitness function (output function) according to your input data. Then apply any optimization technique to select a best one. It may be a Biogeography Based optimization, Ant colony optimization or any other.
.
could you describe your learning task ? is it something like time series prediction ?
.
Thank you sir N.Ayyanathan, the cross validation technique you mean V fold technique?? or any other..kindly reply
Thank you
Fabrice Clerot, sir actually I have earthquake time series and output is sloshing amplitude...the 10th earthquake combines to form 57000 instances i.e rows..each earthquake has time steps 0.02 sec.hence earthquake 1 has 0 to 99 sec makes 5000 instances, earthquake 2, 0 to 65 sec makes 3800 instances,so combining all 57000 instances are there.
Thank you sir @ S.Anbazhagan, but if m taking 43000 data instances for training, with epocs 100, its taking 4 hour..so what shall i do??am i doing right?
K-fold cross validation would be a good choice. Vary the value of k (3, 5, 7 etc.). Or you can take some percentage of data for training and rest of parts are for testing (also some people keep another set as validation set).
- Badges
- Science topic