Generalize gradient check for multiple weight matrices #818
Conversation
|
@prlz77, thanks for your PR! By analyzing the history of the files in this pull request, we identified @Randl, @nyanp and @edgarriba to be potential reviewers. |
tiny_dnn/network.h
Outdated
|
|
||
| if (w.empty()) continue; | ||
| if ((*weights[0]).empty()) continue; |
There was a problem hiding this comment.
if (weights.empty() || (*weights[0]).empty()) continue; is preferred in case of current->weights() returns empty vector.
|
You need to run clang-format |
tiny_dnn/network.h
Outdated
|
|
||
| if (w.empty()) continue; | ||
| if (weights.empty() || (*weights[0]).empty()) continue; |
There was a problem hiding this comment.
@nyanp what would make that weights is not empty but yes first indexed element ?
There was a problem hiding this comment.
@edgarriba
Ah probably if (weights.empty()) is enough in this case. Checking emptiness of each elements can be performed in calc_delta.
There was a problem hiding this comment.
Should I change it then? I can do it in just a moment but I think it is correct both ways.
|
@prlz77 use clang-format so that we can merge |
|
Problems with training? Looks strange |
|
@Randl It seems a precission problem, I think this can be solved just by training for more iterations. |
1 similar comment
|
@edgarriba this is complementary to #846, so it is still necessary. |
* Generalize gradient check for multiple weight matrices * Add nyanp fix * clang-format
No description provided.