Rolf Jagerman
Rolf Jagerman
I've been unexpectedly busy these last summer months, so I unfortunately did not find much time to work on Glint. There has been some interesting discussion from users in issues...
In order to facilitate an easier setup for localhost testing it would be nice to spawn a glint subsystem from within an application that uses the framework. E.g.: ``` val...
By default the parameter server aggregates new values by addition. This is ideal for some algorithms, but not for others. I've made some changes to make aggregation more flexible in...
To make it easier to use, we should publish the package to a centralized repository such as Sonatype.
We need to add instructions on how to run the system on a cluster.
We should separate out the performance benchmark tests and run them automatically CI. Additionally, more performance benchmark tests should be written to properly test the performance of all the different...
It would be highly interesting to compare the performance of GlintLDA to [DistML](https://github.com/intel-machine-learning/DistML). Especially considering DistML also runs in Spark.
The API for computing the LDA model and obtaining the result is different from the standard MLLib implementations (e.g. the online or EM LDA implementations). This can make it difficult...