Conclusion

Thoughts


The non-linear SVMs seemed to improve performance, but without cross-validation there is really no way of knowing.

A better strategy for me to take would have been to tune the baseline parameters to a local maximum and then proceed
with the fancy stuff. It seems likely that the basis expansion would have improved the score more with better underlying features.

Regardless, this was a lot of fun.



4 / N
Prev | Home | Next