My suggestion would be to do some preliminary QC on the sequence data first, which may help dictate which assemblers you may want to look into. Run a k-mer analysis to determine the level of actual coverage and complexity of the data (you could use Jellyfish, khmer, and a whole slew of tools to generate this data). Also, we run preQC to give a more complete assessment.
This, plus what library types you have, normally helps dictate which assemblers may work best. If you have overlapping shotgun libraries and a genome with low heterozygosity, ALLPATHS-LG or DISCOVAR are great (with the latter you would need to scaffold with a separate tool). Which one depends on the length of the sequence data you have.
If the het. rate is pretty high you could give Platanus a go; we've had fairly reasonable luck with it on a few troublesome genomes. You can also use SOAPdenovo, though I believe it's now deprecated in favor of MEGAHIT (we haven't tried this one yet).
"Best assembler" is in the eye of the beholder. What are your requirements? Longest NG50? Most comprehensive gene coverage? Accurate resolution of heterozygosity? Best long range connectivity? Most reads remapping to your assembly?
There is no single best assembler, or single best metric for determining the best assembly. I recommend the Assemblathon 2 paper for its discussion of assembly evaluation, as well as challenges posed by heterozygosity, repetitive sequences, etc.
Here is a recent paper discussing using DISCOVAR for insect assembly: http://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-016-2531-7
Might be helpful