1) What is the effect of randomSeed
parameter on dimensionality reduction
by random projection
in weka
?
2) Secondly it is said that dimensionality reduction
does not loss information, But I have observed that if we set the numberOfAttributes
smaller, it improves accuracy, Whereas if we set numberOfAttributes
close to current(actual) or a large value, whether it reduces accuracy?
Google for "random number seed" to understand the random seed parameter.
Dimensionality reduction loses information, but the lower dimensionality can make things easier to optimize. The data lost can also be distracting noise, so don't be surprised to see the lossy approach improve performance sometimes. Just don't rely on it.