Test linearSVM with random data and plotting

[kordjamshidi]

See this example and the breeze plotting. let me know if you agree with merging this and then I clean it a bit. (ignore intializeSparseNetwork test, it leaked on this branch)

[danyaljj]

@kordjamshidi thanks for the example plot. I'm afraid I don't like the usage. In my opinion it should be much simpler interface. For example, for plotting each property we should be able to just do:

ClassifierUtils.plotFeature(property, instances)

or variations of this (like where to save, how to filter the instances, etc).

What's your thoughts?

[kordjamshidi]

Actually this is the original breeze usage, I did not do anything for plotting. I think if we agree on using breeze then we can modify the way we use it. This PR is 1) for having a training test for this specific toy data which is linearly separable 2) adding breeze as a dependency.

[kordjamshidi]

@danyaljj

[danyaljj]

Sorry for my delay. I'm trying to better understand the goal of this PR. Here are two things I'm very confused about:

• What are we trying to get here?
• More importantly, what is the long-term vision for this?

[kordjamshidi]

have you seen my above comment? This is not necessarily a deep PR. 1) Do we want a good test case for learning algorithms, with a clean linearly separable data? this is a good test case for testing the learning algorithms, and we lack this in ur examples and test cases. 2) Do we want to visualize data? I guess, this is very important for us and here is using breeze until we have our own possible plotting in place.

[danyaljj]

How about we discuss this (and other related issues) on Wednesday after meeting with Dan? Would that work?

[kordjamshidi]

Sure, agree.

[kordjamshidi]

@danyaljj this is up to date

[bhargav]

Looks good to me. In the future, we can add unit tests based on the RandomDataModel.

[kordjamshidi]

The first test I was doing was the SparsePerceptron that does not work and needs someone's attention :-) .