if we continue fitting loss goes up

dswah / sgcrfpy

SGCRFpy: Sparse Gaussian Conditional Random Fields in Python

MIT License

29 stars 5 forks source link

if we continue fitting loss goes up #14

Closed dswah closed 8 years ago

dswah commented 8 years ago

might be solved by checking the line search equations

dswah commented 8 years ago

first 200 iterations

dswah commented 8 years ago

next 200

dswah commented 8 years ago

getting some very small fluctuations on this problem (random Lambda with clusters, 50x50). perhaps this is due to the slack parameter, which allows loss to increase slightly... If we make it smaller perhaps we'll see much smaller increases?

dswah commented 8 years ago

this is fixed by the newer check_descent() methods.