Skip to content
This repository was archived by the owner on Nov 17, 2023. It is now read-only.

Add several optimizer in R#2822

Merged
terrytangyuan merged 3 commits intoapache:masterfrom
ziyeqinghan:optimizer
Jul 26, 2016
Merged

Add several optimizer in R#2822
terrytangyuan merged 3 commits intoapache:masterfrom
ziyeqinghan:optimizer

Conversation

@ziyeqinghan
Copy link
Contributor

Add Adam, AdaGrad, AdaDelta optimizer in R and test them in mnist classification. Part of the test results are below.

  • Adam optimizer with parameters learning.rate=0.001, optimizer="adam"
Batch [50] Train-accuracy=0.414375
Batch [100] Train-accuracy=0.571875
...
Batch [400] Train-accuracy=0.77541015625
Batch [450] Train-accuracy=0.788229166666667
[1] Train-accuracy=0.791672243040685
[1] Validation-accuracy=0.900841346153846
Batch [50] Train-accuracy=0.9034375
...
[5] Train-accuracy=0.965745192307692
[5] Validation-accuracy=0.96474358974359
Batch [50] Train-accuracy=0.971875
...
Batch [450] Train-accuracy=0.986423611111111
[10] Train-accuracy=0.986378205128205
[10] Validation-accuracy=0.972956730769231
  • AdaGrad optimizer with parameters learning.rate=0.05, optimizer="adagrad"
Batch [50] Train-accuracy=0.576875
Batch [100] Train-accuracy=0.716484375
...
Batch [450] Train-accuracy=0.877534722222222
[1] Train-accuracy=0.879901632762313
[1] Validation-accuracy=0.947015224358974
Batch [50] Train-accuracy=0.94875
...
Batch [450] Train-accuracy=0.974982638888889
[5] Train-accuracy=0.975060096153846
[5] Validation-accuracy=0.969350961538462
Batch [50] Train-accuracy=0.9771875
...
Batch [450] Train-accuracy=0.984826388888889
[10] Train-accuracy=0.984909188034188
[10] Validation-accuracy=0.973056891025641
  • AdaDelta optimizer with parameters optimizer="adam"
Batch [50] Train-accuracy=0.116875
Batch [100] Train-accuracy=0.24375
Batch [150] Train-accuracy=0.395208333333333
...
Batch [400] Train-accuracy=0.708984375
Batch [450] Train-accuracy=0.733871527777778
[1] Train-accuracy=0.741652168094218
[1] Validation-accuracy=0.950120192307692
Batch [50] Train-accuracy=0.9453125
...
[5] Train-accuracy=0.984124599358974
[5] Validation-accuracy=0.973657852564103
Batch [50] Train-accuracy=0.989375
...
Batch [450] Train-accuracy=0.994201388888889
[10] Train-accuracy=0.994190705128205
[10] Validation-accuracy=0.975460737179487

@terrytangyuan
Copy link
Member

Looks great! Please resolve conflicts and we'll merge.

@ziyeqinghan
Copy link
Contributor Author

The branch is rebased.

@terrytangyuan
Copy link
Member

I cannot merge it if it doesn't pass tests :-(

@ziyeqinghan
Copy link
Contributor Author

I don't know why cpp_test failed because I didn't change the c/c++ code. Could you gave me some suggestions? thank you so much!

@terrytangyuan
Copy link
Member

@tqchen Could you take a look? I don't have permission to merge this and cannot rerun Travis for some reason

@terrytangyuan
Copy link
Member

I rerun the tests but then it's outdated again...

@ziyeqinghan
Copy link
Contributor Author

The branch is rebased again.

@terrytangyuan terrytangyuan merged commit 1767f0b into apache:master Jul 26, 2016
@terrytangyuan
Copy link
Member

Thanks!

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants