You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 23 Next »


Completed/ Working

  • NDArray API
  • Symbol API
  • Module API
  • Most examples ported over from Scala package
  • Most tests ported over from Scala package

Needs General Help

  • Test Coverage plugin integration 
  • Benchmarks - See python package examples/image-classification/benchmark.py for a guideline
  • Good API Documentation
  • Examples turned into integration tests
  • More tests, increased test coverage - In Progress
  • Add GPU tests
  • Integration with CI
  • Documentation on the Website
  • Install Documentation with OSX - Ubuntu, Arch, Deep Learning AMI
  • Use logging vs println in source code 
  • Improving the Debug experience (mxnet-core, scala, clojure)

Needs Help Bugs

  • GAN example has a memory leak (this could be due to NDArray iterator)
  • Style transfer example - (can't get the content style transfer to work at the higher relu levels like 4-2 from noise, only at the first layer). Also is cpu only right now, needs to be converted to GPU.
  • Visualization example - was working with conv layers and then stopped after switching to scala jars. Bug is parsing attr information from the nodes. Not sure if it is on Scala side or Clojure. Need to investigate.
  • CNN text example - needs testing with Word2Vec instead of Glove - also compare to python results

Needs Help Features

  • Porting the Scala Inference to Clojure
  • Custom Operators 
  • KV Store Testing - There is interop and general support, but needs testing (I was unable to do so sufficiently working locally)
  • Gluon API - Needs discussion and coordination with Scala package
  • Feed Forward API - Needs discussion and coordination with Scala package
  • Arch Linux/ Ubuntu support - Needs discussion and coordination with Scala package
  • OSX gpu support - Needs discussion and coordination with Scala package
  • Investigate autogen of ndarray and symbol as a automatic compile time step rather than manual
    • Look into the same process that the Scala package uses to generate docs with file signatures


Current Test Coverage (6/22/18)

|-----------------------------------------+---------+---------|
|                               Namespace | % Forms | % Lines |
|-----------------------------------------+---------+---------|
|                           dev.generator |   93.07 |   94.33 |
|           org.apache.clojure-mxnet.base |  100.00 |  100.00 |
|       org.apache.clojure-mxnet.callback |  100.00 |  100.00 |
|        org.apache.clojure-mxnet.context |   76.00 |   85.71 |
|          org.apache.clojure-mxnet.dtype |  100.00 |  100.00 |
|    org.apache.clojure-mxnet.eval-metric |   95.00 |  100.00 |
|       org.apache.clojure-mxnet.executor |   89.66 |   95.00 |
|    org.apache.clojure-mxnet.initializer |   25.00 |   43.75 |
|             org.apache.clojure-mxnet.io |   71.71 |   80.00 |
|        org.apache.clojure-mxnet.kvstore |   82.91 |   76.00 |
| org.apache.clojure-mxnet.kvstore-server |   58.73 |   50.00 |
|   org.apache.clojure-mxnet.lr-scheduler |   22.22 |   66.67 |
|         org.apache.clojure-mxnet.module |   74.99 |   88.62 |
|        org.apache.clojure-mxnet.monitor |  100.00 |  100.00 |
|        org.apache.clojure-mxnet.ndarray |   90.86 |   91.38 |
|      org.apache.clojure-mxnet.optimizer |   91.21 |   88.24 |
|       org.apache.clojure-mxnet.profiler |   20.00 |   40.00 |
|         org.apache.clojure-mxnet.random |  100.00 |  100.00 |
|          org.apache.clojure-mxnet.shape |   94.44 |   90.00 |
|         org.apache.clojure-mxnet.symbol |   48.87 |   54.62 |
|           org.apache.clojure-mxnet.util |   98.86 |   99.16 |
|  org.apache.clojure-mxnet.visualization |    6.82 |   20.00 |
|-----------------------------------------+---------+---------|
|                               ALL FILES |   80.01 |   84.14 |
|-----------------------------------------+---------+---------|

  • No labels