Dongxu Zhang 14-12-01

From cslt Wiki
Jump to: navigation, search

Accomplished this week

  • Accomplish a simple program to train max(0, 1+dist(head,tail)-dist(head,wrong)) on both dot product and euclidean. with a low predict accuracy ,just like Unstructured model which mentioned in previous papers. Up to 0.1 and then down to ~0.5 due to overfitting maybe( I'll also check my code again). And also tried transE, have not finished yet.
  • Adding a penalty on weight parameter of rwthlm leads to a similar result :(
  • Analysis the wiki html.

Planned for next week

  • Finish the vocabulary as soon as possible.( Maybe Wed.)
  • Analysis the wiki category infomation into json.
  • check the code of TransE and unstructure.