| EXAMPLE: compare svm solvers on digits |
|
More...
|
about 13 years ago
|
| ENH minor fixes in cutting plance svm (only run qp once for standard svm) |
|
More...
|
about 13 years ago
|
| FIX one-slack-svm test for already active constraints. |
|
More...
|
about 13 years ago
|
| ENH added OneSlackSVM learner, simple test working. |
|
More...
|
about 13 years ago
|
| ENH refactor common functions out of SSVMs |
|
More...
|
about 13 years ago
|
| ENH refactor exhaustive inference helpers for tests |
|
More...
|
about 13 years ago
|
| ENH add constraints to cutting plane method in batches |
|
More...
|
about 13 years ago
|
| ENH get rid of some debug code, some refactoring in cutting plane svm |
|
More...
|
about 13 years ago
|
| ENH refactor constraint check out of main loop |
|
More...
|
about 13 years ago
|
| FIX some fixes in kmeans_init after introduction of LatentGraphCRF |
|
More...
|
about 13 years ago
|
| remove trace from subgradient svm for tests (that bug was fixed) |
|
More...
|
about 13 years ago
|
| FIX in binary svm tests after removing bias |
|
More...
|
about 13 years ago
|
| ENH speed optimization for online learning |
|
More...
|
about 13 years ago
|
| ENH refactor show_loss into cutting plane svm |
|
More...
|
about 13 years ago
|
| ENH add parameter so subgradient descent to show the true loss |
|
More...
|
about 13 years ago
|
| FIX D'OH online subgradient descent fix |
|
More...
|
about 13 years ago
|
| ENH don't add bias to unstructured problems in inner loop. it is way inefficient :-/ |
|
More...
|
about 13 years ago
|
| ENH subgradient svm: don't remember stuff we will never need again. |
|
More...
|
about 13 years ago
|
| DOC subgradient ssvm: docstring tell defaults |
|
More...
|
about 13 years ago
|
| example: some updates for grid_crf |
|
More...
|
about 13 years ago
|
| FIX for DirectionalGridCRF psi |
|
More...
|
about 13 years ago
|
| ENH possibly return energy in ad3 |
|
More...
|
about 13 years ago
|
| more / better tests for energy. now fails :-/ |
|
More...
|
about 13 years ago
|
| add test for online learning in subgradient ssvm |
|
More...
|
about 13 years ago
|
| Fix test for learning_rate rename |
|
More...
|
about 13 years ago
|
| COSMIT init w with zeros. no hack needed any more. |
|
More...
|
about 13 years ago
|
| Rename learningrate to learning_rate, add online learning to subgradient structured svm |
|
More...
|
about 13 years ago
|
| score is 1 - loss |
|
More...
|
about 13 years ago
|
| EXAMPLE fix for new interface |
|
More...
|
about 13 years ago
|
| FIX only add if loss nonzero. Important for latent SVM, where psi(x,h) != psi(x, h_hat) but y == y_hat. |
|
More...
|
about 13 years ago
|
