Page History

...

The labels below were chosen for prediction initially by the model. Only the issues which are specific to these labels are what is being tested on, in other words either the specific label being tested on was predicted by the model or the specific label was the actual label on the issue. The accuracy shown below denotes where the model predicted a label and that was one of the actual labels in the repo.

Target Labels: [Performance, Test, Clojure, Java, Python, C++, Scala, Question, Doc, Installation, Example, Bug, Build, ONNX, Gluon, Flaky, Feature,  CI, Cuda] 

*** The accuracy metric was collected using sklearn's accuracy_score method ***

...

Programming languages were trained on large amounts of data pulled from a wide array of repositories we are able to deliver these high metrics especially with regards to programming languages by making use of MXNet for deep learning to learn similarities among these languages we consider (which are the programming languages that are present in the repo). Specifically this was trained on data snippets of files pulled from the data files present here: https://github.com/aliostad/deep-learning-lang-detection/tree/master/data. Thus, we can believe that this accuracy measurement can be maintained on prediction of new issues which have code snippets presented within them. Training was done with a 6 layer deep model in Keras-MXNet using the 2000 files present (and creating snippets out of them) for the languages we are interested in from the repository data above. For inference, we are using pure MXNet with the model being served from the work of this repository: https://github.com/awslabs/mxnet-model-server. 

Motivations/Conclusion:

We do notice that there is a case that may be present of overfitting here, especially with the case of the Performance label. However in looking further into the issues labeled as Performance, we notice that similar words and phrases are included across issues labeled as Performance (i.e. in most cases the word itself, and words like speed..). The training data for the word embeddings that our model has trained on is able to give these kinds of results due because of word2vec which provides us with a high cosine similarity - we can speculate that these common words were grouped together and hence the model was able to predict these labels with a high accuracy. Given this data, we are able to see which labels the model can predict accurately for. Given a certain accuracy threshold, the bot has the potential to label an issue given that surpasses this value. As a result, we would be able to accurately provide some labels to new issues. . Overall, the mxnet-label-bot will be able to provide an improved experience for its developers.

Two models combined for specific groups of labels allow for us to be able to deliver this capability:

...