search

TruX-DTF / debug-method-name

Learning to Spot and Refactor Inconsistent Method Names
16 stars 17 forks source link

What is the connection between the score at iteration and the CNN model trainde? #2

Closed Kelin-hao closed 3 years ago

Kelin-hao commented 4 years ago

I notice that the score will be lower when training. Whether does it indicate lower score means the CNN model is more accurarte?

Kui-Liu commented 4 years ago

Hello Wang,

Thank you for your interests in our work.

I am sorry. It has been a long time for the experiments, so I do not remember the detail scores in the experiment any more. So I cannot answer your question as well.

I remember that, I did not try many experiments to assess the relationship between score and accuracy. It might be an interesting task to assess the relationship.

Best regards,

Kui

On Fri, 10 Jul 2020 at 13:59, Wang-JLU notifications@github.com wrote:

I notice that the score will be lower when training. Whether does it indicate lower score means the CNN model is more accurarte?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/SerVal-DTF/debug-method-name/issues/2, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEE6IKSRVD3XRBQTYLOHKGDR22U43ANCNFSM4OWIPYHQ .

Kelin-hao commented 4 years ago

Another question, if the deep learning is unsupervuised,the number of outcomes is the size of input vector, how does the model calculate MAE since there is no target ouput?

Kui-Liu commented 4 years ago

Yes, it is unsupervised learning. The mean absolute error and other parameters are tuned through a visualizing network training UI (https://deeplearning4j.org/visualization) provided by DeepLearning4J. It seems that the training UI link is not available anymore.

On Sun, 12 Jul 2020 at 14:55, Wang-JLU notifications@github.com wrote:

Another question, if the deep learning is unsupervuised,the number of outcomes is the size of input vector, how does the model calculate MAE since there is no target ouput?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/SerVal-DTF/debug-method-name/issues/2#issuecomment-657183399, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEE6IKSCFZ5U2XQJBAVUMULR3FM7LANCNFSM4OWIPYHQ .

Kelin-hao commented 4 years ago

Yes, it is unsupervised learning. The mean absolute error and other parameters are tuned through a visualizing network training UI (https://deeplearning4j.org/visualization) provided by DeepLearning4J. It seems that the training UI link is not available anymore. … On Sun, 12 Jul 2020 at 14:55, Wang-JLU @.***> wrote: Another question, if the deep learning is unsupervuised,the number of outcomes is the size of input vector, how does the model calculate MAE since there is no target ouput? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEE6IKSCFZ5U2XQJBAVUMULR3FM7LANCNFSM4OWIPYHQ .

Thakns for your replay, actually I want to know whether the MAE is between the outcome of the outputlayer and the input vector?

Kui-Liu commented 4 years ago

https://github.com/SerVal-DTF/debug-method-name/blob/master/LearningModel/src/main/java/edu/lu/uni/deeplearning/extractor/CNNFeatureExtractor.java#L158

The output of the dense layer is considered as the extracted features for method body code.

On Sun, 12 Jul 2020 at 16:19, Wang-JLU notifications@github.com wrote:

Yes, it is unsupervised learning. The mean absolute error and other parameters are tuned through a visualizing network training UI ( https://deeplearning4j.org/visualization) provided by DeepLearning4J. It seems that the training UI link is not available anymore. … <#m-6995311358388405859> On Sun, 12 Jul 2020 at 14:55, Wang-JLU @.***> wrote: Another question, if the deep learning is unsupervuised,the number of outcomes is the size of input vector, how does the model calculate MAE since there is no target ouput? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment) https://github.com/SerVal-DTF/debug-method-name/issues/2#issuecomment-657183399>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEE6IKSCFZ5U2XQJBAVUMULR3FM7LANCNFSM4OWIPYHQ .

Thakns for your replay, actually I want to know whether the MAE is between the outcome of the outputlayer and the input vector?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/SerVal-DTF/debug-method-name/issues/2#issuecomment-657190260, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEE6IKXB462QM7BNLKWC3WLR3FWZDANCNFSM4OWIPYHQ .