Age | Commit message (Collapse) | Author | |
---|---|---|---|
2017-11-28 | Merge pull request #165 from ProGamerGov/patch-1HEADmaster | Soumith Chintala | |
Fixed the link to the Adam research paper | |||
2017-11-28 | Fixed the link to the Adam research paper | ProGamerGov | |
Fixed the link to the, "Adam: A Method for Stochastic Optimization" research paper. This link no longer works: http://arxiv.org/pdf/1412.6980.pdf I and many others involved with machine learning, find it's better to link to the research paper's arXiv page itself, and not directly to the PDF file. This is because it's not easy to get to the research paper's arXiv page, directly from the PDF, but it is easy to get to the PDF from the arXiv page. | |||
2017-02-08 | Merge pull request #150 from Amir-Arsalan/patch-1 | Soumith Chintala | |
Update algos.md | |||
2017-02-08 | Update algos.md | Amir Arsalan Soltani | |
2016-10-16 | Merge pull request #142 from ibmua/patch-1 | Soumith Chintala | |
Fixed misspelling | |||
2016-10-16 | Update algos.md | Menshykov | |
2016-10-09 | Merge pull request #138 from DmitryUlyanov/master | Soumith Chintala | |
Fix polyinterp to let lbfgs wtih lswolfe work on GPU | |||
2016-10-09 | clean up comments | Dmitry Ulyanov | |
2016-10-09 | fix polyinterp, so lswolfe can be used with CUDA | Dmitry Ulyanov | |
2016-09-30 | Merge pull request #137 from Atcold/patch-1 | Soumith Chintala | |
Update intro.md | |||
2016-09-30 | Fix formatting and add Cuda training info | Alfredo Canziani | |
2016-09-30 | Update intro.md | Alfredo Canziani | |
Refactored text for consistency with the rest of the doc. The goal of training a nn is to perform well on the validation set, not the training set. Removed `local` from snippet, so they are runnable in the interpreter. | |||
2016-09-29 | Merge pull request #136 from wydwww/master | Soumith Chintala | |
Fix typos | |||
2016-09-29 | Fix typos | Yiding Wang | |
2016-09-27 | Merge pull request #135 from Atcold/local-doc | Soumith Chintala | |
Enable local doc for inline help | |||
2016-09-27 | Enable local doc for iline help | Alfredo Canziani | |
2016-09-20 | Merge pull request #134 from hughperkins/migrate-example-from-nn | Soumith Chintala | |
move optim doc from nn | |||
2016-09-20 | move optim doc from nn | Hugh Perkins | |
2016-09-15 | Merge pull request #132 from codeAC29/master | Soumith Chintala | |
Prevent displaying of plots and documentation for it | |||
2016-09-15 | Added documentation for display and logscale | Abhishek Chaurasia | |
2016-09-15 | Added option to set/reset displaying of plot | Abhishek Chaurasia | |
2016-09-13 | Merge pull request #131 from korymath/patch-1 | Soumith Chintala | |
Spelling mistake. | |||
2016-09-13 | Spelling mistake. | Kory | |
2016-09-06 | make initialMean to be configurable | Soumith Chintala | |
2016-09-06 | Merge pull request #130 from iassael/master | Soumith Chintala | |
Reverted to zero mean squared values init | |||
2016-09-06 | reverted to zero mean squared values init | Yannis Assael | |
2016-08-25 | Merge pull request #127 from gcinbis/patch-2 | Soumith Chintala | |
Copy C1 value, in case it is a Tensor reference | |||
2016-08-25 | Keep objective values, in case they are references | R. Gokberk Cinbis | |
When opfunc() simply returns the output state variable of a nn model (ie. when opfunc() simply returns my_net:forward()'s output), the second opfunc() call within the for loop updates not only C2, but also C1. In this case, dC_est is wrongly 0. Avoid this behaviour by blindly copying C1 contents when it is a Tensor/CudaTensor. The overhead should be bearable as C1 is a scalar. | |||
2016-08-25 | Merge pull request #126 from gcinbis/patch-1 | Soumith Chintala | |
Reduce numerical errors. | |||
2016-08-25 | Reduce numerical errors. | R. Gokberk Cinbis | |
x[i]+eps-2*eps may not result in exactly the same x[i], which may increase approximation error in the gradient estimate. | |||
2016-08-08 | Merge pull request #124 from Atcold/patch-1 | Soumith Chintala | |
One-line Logger initialisation | |||
2016-08-08 | One-line Logger initialisation | Alfredo Canziani | |
A `Logger` can be created and setup in one line log = optim.logger('foo'):setNames{'a', 'b'}:style{'-', '-'} | |||
2016-08-07 | Merge pull request #123 from torch/de | Soumith Chintala | |
Add Differential Evolution | |||
2016-08-07 | Add Differential Evolution | Li Zhijian | |
2016-07-30 | fixing to be tensor type agnostic | Soumith Chintala | |
2016-07-21 | Merge pull request #122 from Cadene/master | Soumith Chintala | |
Add LearningRateDecay to Adam | |||
2016-07-21 | Add Adam learningRateDecay to doc | Cadene | |
2016-07-21 | Add learningRateDecay to Adam | Cadene | |
2016-06-30 | Merge pull request #121 from Atcold/doc-fix | Soumith Chintala | |
Documentation and code refactoring | |||
2016-06-30 | Add optim.Logger() documentation | Alfredo Canziani | |
2016-06-30 | Fix bad alignment, trailing spaces and tabs | Alfredo Canziani | |
2016-06-30 | Fix state/config improper documentation | Alfredo Canziani | |
2016-06-27 | Refactoring documentation | Alfredo Canziani | |
2016-06-15 | Merge pull request #119 from chenb67/master | Soumith Chintala | |
add weight decay support to adamax | |||
2016-06-15 | add weight decay support to adamax | Chen Buskilla | |
2016-06-10 | Merge pull request #118 from gcheron/adam-wdec | Soumith Chintala | |
add weight decay support to adam | |||
2016-06-10 | add weight decay support to adam | gcheron | |
2016-06-09 | Merge pull request #117 from andreaskoepf/rmsprop_warmup | Soumith Chintala | |
Init rmsprop mean square state 'm' with 1 instead 0 | |||
2016-06-09 | Init rmsprop mean square state 'm' with 1 instead 0 | Andreas Köpf | |
With alpha near 1 (e.g. the default value 0.99) the gradient was likely scaled up by a division by a number <1 during the first few iterations. With the original impl the learning rate had to be set to a much smaller value when using rmsprop compared to plain-vanilla sgd in order not to diverge. | |||
2016-06-03 | Merge pull request #115 from torch/revert-113-sgd-lrs-fix | Soumith Chintala | |
Revert "Fix bug with sgd individual learning rates" |