Yeah, but WEKA is a piece of Java bloatware. Waffles looks awesome, especially if they keep it lean and mean and truly try to adhere to the unix philosophy like they say. I'm excited.
Java hate? I never said all Java programs were bloatware, only that WEKA was and that was a distinct difference between Waffles, which is what the OP asked.
I found the phrase 'Java bloatware' to be indicative of a certain sentiment towards the language (or JVM environment) in general (as compared to just 'bloatware'). And, given common use cases for machine learning toolkits, the overhead incurred by the JVM doesn't warrant (IMHO) ruling out Java-based options. I must have misread your comment, thanks for clarifying!
One of the cool things they've taken the trouble to do is to make sure that almost all functionality is accessible from the command line.
As such WEKA is easily scriptable.
This is great for integration into bigger projects.
It can easily be chained like standard unix tools; I've used this functionality in projects with good success.
I do a lot of machine learning, and I'm not sure how useful this is. Machine learning technology is not currently automatic enough that it can be effective at the command-line level. By moving one level down the stack, to the API and library bindings level, it becomes a lot more effective.
What I'm saying is that to use command-line machine learning, I would have to couch it in a bunch of scripting. At that point in time, I might as well use library bindings. So command-line is not a selling point for me.
14 comments
[ 2.3 ms ] story [ 58.5 ms ] threadI'll have a look at the underlying C++ classes API (http://waffles.sourceforge.net/apidoc/html/index.html).
One of the cool things they've taken the trouble to do is to make sure that almost all functionality is accessible from the command line.
As such WEKA is easily scriptable. This is great for integration into bigger projects. It can easily be chained like standard unix tools; I've used this functionality in projects with good success.
What I'm saying is that to use command-line machine learning, I would have to couch it in a bunch of scripting. At that point in time, I might as well use library bindings. So command-line is not a selling point for me.