search

Neuroglycerin / neukrill-net-work

NDSB competition repository for scripting, note taking and writing submissions.
MIT License
0 stars 0 forks source link

Submission with simple bag of visual words #26

Closed gngdb closed 9 years ago

gngdb commented 9 years ago

We'd like to find out what kind of score you can get using just simple things like the size of the image, black/white ratio, shape of the image or anything else you can think of that is really obvious and easy to compute.

Will require closing this issue to calculate sizes and shapes. Make more issues for other bits of processing as required.

Close this once we've made a submission using each of these in turn on their own, and combined in whatever way and we've got a report in the wiki on what happened.

gngdb commented 9 years ago

Should also include part in the report on how well we can do with just class priors derived from the training data, compared to uniform predictions.

gngdb commented 9 years ago

@scottclowe didn't you already do this?

scottclowe commented 9 years ago

I did some submissions, but I did not optimise the many possible permutations of simple properties to include. This will require running with lots of settings and doing a comparison. Would be useful to optimise further (independent of the rest of the code) so we know which are the most useful simple features, and which are not worth including in further predictions.