This summer I will be working on implementing distance metric learning technique for Large Margin Nearest Neighbor (LMNN) Classification using LR-SDP optimizer along with an adaptation of LMNN using a Boosting algorithm.
Everything went well during this period. During last few days, I built prototype structure for both LMNN and BoostMetric. In addition to the structure, Constraints class for generating target neighbors and impostors constraints was implemented. It will be shared by both the methods. After completing all the preliminary things, I tried building up the SDP for LMNN, which didn't result in success though. Things got clearer after a long hour of discussion earlier today on IRC. Hopefully, a primal-dual SDP version of LMNN will be ready by the end of the week.
Looking for a great summer ahead with all these cool people here at mlpack. :)
Generated by 1.8.13