These are chat archives for dmlc/mxnet

12th
Jun 2016
杨博 (Yang Bo)
@Atry
Jun 12 2016 08:01
Hi, @javelinjs, I found that ml.dmlc.mxnet.spark.MXNet did not implement org.apache.spark.ml.Estimator. Instead, it provides its own API. Why?
Yizhi Liu
@javelinjs
Jun 12 2016 09:29
Hi @Atry , I don't think there's any technical reason which prevents it from implementing spark.ml.Estimator. I did so just because I was rushing to release the first version. It would be really appreciated if you could help to improve the APIs.
杨博 (Yang Bo)
@Atry
Jun 12 2016 10:17

Usually we have DataFrame or Dataset imported from outside world, then we create a org.apache.spark.ml.Pipeline to perform tasks on the data.

MXNet would be hard to integrate into the pipeline if it did not implement org.apache.spark.ml.PipelineStage.

Yizhi Liu
@javelinjs
Jun 12 2016 10:20
Exactly. Feel free to open a PR for this.
杨博 (Yang Bo)
@Atry
Jun 12 2016 10:21
😃