More tests for concat and convolution #33
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
update pom.xml
Multiple strategies to adjust the learning rate
add Tensor and TensorMath comments
Because of some unknown reasons, the back propagation method in Concat is not `updateGradInput`, but `backward` instead, which should not override in the class inherited from module. So the testcases in concat should adopt to the situation.
1. Add some testcases for layers which use mkl dnn api. There are some testcases in WebScaleML. Alghough it has been passed all of testcases of WebScaleML, for some big input, like Convolution from GoogLeNet, AlexNet, the result may be wrong. Based on current testcases, we found that we must do more test for float and big input. 2. Fix the bug about wrong result of gradInput of Pooling (Max and Avg), because MKL-DNN will not erase the data existing in gradInput. 3. Fix the bug about wrong result when some layers in concat layer are not MKL-DNN layer. 4. Note, because the different implementation of layers between MKL-DNN and Spark-DL, the result is not always same for convolution, lrn and batch norm. So the output and gradInput of AlexNet, GoogLeNet v1 and GoogLeNet v2 are not completely same with SparkDL w/ MKL-Blas. Currently, the error we set may be 1e-4~1e-5. We need some convergence test for the implementation of MKL-DNN.
1. memset optmized with openmp 2. omit double conversion 3. fix backward filter and bias of convolution, which will get wrong answer at first layer in alexnet, googlenet and so on.
wzhongyuan
pushed a commit
to wzhongyuan/BigDL
that referenced
this pull request
Jan 16, 2019
* Add opencv support * Add opencv support
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
fix #23 . Some problems exist in spatial convolution and concat .