Additional preprocessors

[matejklemen]

Description of Changes

Addresses #7. Implements 3 new preprocessors:

1. Binarizer: convert numerical columns into binary columns based on a threshold. Works with either a Double scalar or a DenseVector[Double]. In the second case, each value of vector represents a threshold for a separate column, e.g. 1st value is used as threshold for 1st numeric column.
   Motivated by sklearn's Binarizer.

2. Normalizer: converts rows to unit norm. Currently, there are 3 norms supported (L1, L2, max). These should probably be exposed in some kind of distance module for further code reuse, I am open to ideas on where to put these norm functions.
   Motivated by sklearn's Normalizer.

3. RangeScaler: converts numerical columns to a certain range, e.g. from 0 to 1.
   Motivated by sklearn's MinMaxScaler (RangeScaler or IntervalScaler seems like a more intuitive name to me, though I can change it if min-max scaling is a "standardized" name for this).

Includes

• Code changes
• Tests
• Documentation

[inejc]

Thanks @matejklemen 👍, will look at the PR as soon as I find time for it.

[matejklemen]

Alright, no rush.

I've fixed the compilation error that was previously present and tried to address Codacy's issues with the PR, though it seems that I don't understand what the issue "Found unnamed result or parameter" actually means 😅

[inejc]

Thanks again for the PR @matejklemen 🙂. This is definitely in the right direction but requires some small changes. Let me know if my comments aren't clear or you have any further questions.

[matejklemen]

Alright, I think I got all of them. Thanks for the helpful comments 🙂

I was a bit confused about sealed traits, so I did it slightly differently than what you said. It seemed redundant to me to define available Norms twice - once in Normalizer and again in Norms so I just did it in Norms.
Please let me know if that's ok.

[inejc]

I've added some comments but we are at the doorstep of merging this 😄. Do you want to add docs for the new preprocessors in this PR or in #69 (I can do it too once this one is merged if you don't want to)?

[matejklemen]

New batch of changes:

• added tests for Norms (my bad, completely forgot before 😅 )
• made minor style fixes
• added docs for the three preprocessors (I only put the imports that I think are non-obvious in the examples, as I'm doing in Update documentation #69)
• added a test with different range for RangeScaler
• removed unnecessary copying that is done by toDenseMatrix

[inejc]

@matejklemen everything LGTM except the comment I wrote regarding compilation (a single and simple change). I will merge once this one is pushed 😄.