| don't show "true" loss by default in 1-slack ssvm |
|
More...
|
almost 13 years ago
|
| ENH use sparse matrix in 1-slack svm |
|
More...
|
almost 13 years ago
|
| ENH: add option to show true loss in one-slack ssvm. |
|
More...
|
almost 13 years ago
|
| ENH use sparse matrices for constraints in n-slack svm. ~5-10x speedup. Oh, the humanity! |
|
More...
|
almost 13 years ago
|
| Added repr to StructuredProblem, unstructured svms |
|
More...
|
almost 13 years ago
|
| ENH: added decaying learning rate to SubgradientStructuredSVM |
|
More...
|
almost 13 years ago
|
| ENH repr for SSVM objects |
|
More...
|
almost 13 years ago
|
| ENH add another convenience method for pickling... |
|
More...
|
about 13 years ago
|
| ENH don't do input validation in psi of CrammerSingerSVM: way to expensive. |
|
More...
|
about 13 years ago
|
| ENH some speed improvements in one_slack_svm |
|
More...
|
about 13 years ago
|
| intermediate checkin for benchmarking... |
|
More...
|
about 13 years ago
|
| EXAMPLE Add multi-class svm example |
|
More...
|
about 13 years ago
|
| EXAMPLE binary svm remove unnecessary tol argument |
|
More...
|
about 13 years ago
|
| FIX now that tol is -10 by default, we can uncomment in 1-slack ssvm |
|
More...
|
about 13 years ago
|
| FIX set cvxopt verbosity in the right place. otherwise it might be messed up since creating the model. |
|
More...
|
about 13 years ago
|
| Implement ssvm loss as macro average |
|
More...
|
about 13 years ago
|
| Implement batch_size=-1 as batch learning for n-slack version |
|
More...
|
about 13 years ago
|
| ENH set default tol for both qp based ssvms to -10, i.e. don't stop on convergence of the dual, wait for no more constraints. |
|
More...
|
about 13 years ago
|
| EXAMPLE: in binary_svm, all learners should get the same C. |
|
More...
|
about 13 years ago
|
| ENH update qp only after adding constraints, if passed some constraints, solve qp once in the beginning. |
|
More...
|
about 13 years ago
|
| TST multi-class crammer singer tests. fails for previous bug. |
|
More...
|
about 13 years ago
|
| FIX horrible horrible bug in batch version of cutting plane algo |
|
More...
|
about 13 years ago
|
| optional random seed in easy blocks dataset |
|
More...
|
about 13 years ago
|
| test n-test latens ssvm with 1-slack ssvm |
|
More...
|
about 13 years ago
|
| TST add test for batch training in n-slack ssvm |
|
More...
|
about 13 years ago
|
| FIX latent SSVM: update constraints only for n-slack formulation |
|
More...
|
about 13 years ago
|
| FIX and test for using other ssvm learners with latent ssvm |
|
More...
|
about 13 years ago
|
| fix visualiztion in latent svm example |
|
More...
|
about 13 years ago
|
| FIX example for new predict method of latent SSVM |
|
More...
|
about 13 years ago
|
| ENH allow different number of hidden states per label :) |
|
More...
|
about 13 years ago
|
