[GitHub] [incubator-mxnet] braindotai edited a comment on issue #9686: [Discussion] MXNet 2.0 Roadmap (was: APIs that might be a good idea to break in 2.0)

[GitBox <gi...@apache.org>]

braindotai edited a comment on issue #9686: [Discussion] MXNet 2.0 Roadmap (was: APIs that might be a good idea to break in 2.0)
URL: https://github.com/apache/incubator-mxnet/issues/9686#issuecomment-511816440
 
 
   I have some suggestions down here:-
   - MXNet only has 4 prebuild datasets, I think we should have more...
       Here are some important missing datasets-
         [KMNIST](https://github.com/rois-codh/kmnist), [EMNIST](https://www.westernsydney.edu.au/bens/home/reproducible_research/emnist), [LSUN](https://www.yf.io/p/lsun), [STL10](https://cs.stanford.edu/~acoates/stl10/), [SVHN](http://ufldl.stanford.edu/housenumbers/), [PhotoTour](http://phototour.cs.washington.edu/patches/default.htm), [SBU](http://www.cs.virginia.edu/~vicente/sbucaptions/), [Flickr8k](http://nlp.cs.illinois.edu/HockenmaierGroup/8k-pictures.html), [Flickr30k](http://web.engr.illinois.edu/~bplumme2/Flickr30kEntities/), [Cityscapes](http://www.cityscapes-dataset.com/), [SBD](http://home.bharathh.info/pubs/codes/SBD/download.html) and FakeData(A fake data generator that generates fake images, useful for GANS)
   
   - As we got the [Estimator](https://github.com/apache/incubator-mxnet/pull/15009/commits) for gluon, (which will be supposedly released with MXNet 1.5), we can have many predefined estimators(very similar to Scikit Learn, with awesome gpu support), for example:-
   ```python
   from mxnet.gluon import estimator
   model = estimator.linearregression
   model = estimator.logisticregression
   model = estimator.ridgeregression
   model = estimator.lassoregression
   model = estimator.knearestneighbors
   model = estimator.kmeansclusttering
   model = estimator.svm
   ....etc
   ```
   These classical ML algorithms work better than DL for some specific tasks and many users want such ML algorithms with gpu support, so that'd be quite awesome.
   
   - We need to have a good and updated website design, for instance in the bottom of [homepage](https://mxnet.apache.org/), even the "copyright(2017 - 2018)" statement is not updated, which sounds like MXNet is no longer sponsored by the Apache Incubator(BTW is this True?). 
       - Also, you can't access this [useful link](http://mxnet.incubator.apache.org/versions/master/tutorials/python/) at all, because the link is unnecessarily and massively hidden somewhere inside the website.(fortunately I got it from an email from my friend, and I don't know how he got it!!)
       - We should update the benchmarks available [here](https://mxnet.incubator.apache.org/versions/master/faq/perf.html)
       - We should put these [FAQ tutorials](https://mxnet.incubator.apache.org/versions/master/faq/index.html) available directly from our default homepage because these are really good tutorials(but hidden under docs) for a beginner in MXNet.
       - If you click on any link for documentation available [here](https://mxnet.apache.org/api/python/index.html) under __some section__, let's say __"Gluon API"__ then afterwards any link for other docs(on the left side) available outside of __"Gluon API" section__ wouldn't work. And this behaviour is common for all __sections__.
       - We should have more information about MXNet displayed right on the homepage(instead of provided links), for example, its key features, ecosystem, community, a link to [holy book of MXNet](http://d2l.ai/), [GluonCV](https://gluon-cv.mxnet.io/), [GluonNLP](https://gluon-nlp.mxnet.io/)... etc.
   
   The reason why I am so worried about the website is because "IT IS IMPORTANT", more we show to the user directly, better the understanding a user can have.(For instance ME!, when I opened the website first time, it was very difficult for me to find good tutorials and examples, instead, I had to rely on GitHub and had to ask in the forum separately about that.)
   
   - KVStore API should note be public, because I haven't really seen any use of it in any tutorial or implementation, secondly `gluon.Trainer` already handles the functionality of KVStore internally, thirdly at the top of [KVStore Tutorial](https://mxnet.apache.org/api/python/kvstore/kvstore.html) it's clearly written that 
   > .. note:: Direct interactions with KVStore are dangerous and not recommended.
   
   so why we are telling users how to use it if it's so dangerous and not recommended?
   
   I love MXNet, that's why I am so picky about small improvements and updates. 
   Sorry if I've written something wrong above.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services