Use BYU Fork of Dsbox Encoder

[epeters3]

Using a fork of d3m.primitives.data_preprocessing.encoder.DSBOX (unit tests included in this PR) to give us these benefits:

• One hot encoding categorical and ordinal columns of all structural types
• When encoding, if a column has more than 12 categories, we only create one hot columns for the 12 most common categories, and have a 13th column for all the less common ones (to prevent feature explosion)
• Unseen categories present in the test set will not cause the pipeline to fail
• Datasets with no categorical or ordinal columns will not cause the one hot encoding process to fail

[codecov-io]

Codecov Report

Merging #75 into develop will increase coverage by 2.29%. The diff coverage is 82.35%.

@@             Coverage Diff             @@
##           develop      #75      +/-   ##
===========================================
+ Coverage    65.56%   67.85%   +2.29%     
===========================================
  Files           28       31       +3     
  Lines         1661     1910     +249     
===========================================
+ Hits          1089     1296     +207     
- Misses         572      614      +42

Impacted Files	Coverage Δ
test/test_pipeline_builder.py	97.97% <ø> (ø)	:arrow_up:
experimenter/pipeline_builder.py	93.87% <ø> (-0.28%)	:arrow_down:
test/config.py	100% <ø> (ø)	:arrow_up:
test/test_pipeline_generator.py	100% <100%> (ø)	:arrow_up:
experimenter/constants.py	100% <100%> (ø)	:arrow_up:
experimenter/config.py	100% <100%> (ø)	:arrow_up:
experimenter/primitives/dsbox_encoder.py	78.44% <78.44%> (ø)
test/test_dsbox_encoder.py	90.32% <90.32%> (ø)
... and 1 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 73cd314...895a5c2. Read the comment docs.

[epeters3]

Hey @orionw, would you be willing to review this PR? If looks good, go ahead and merge it, and try including the code in your efforts to run the experiments. This PR should get rid of those errors we were getting yesterday where sometimes the encoders would run into unseen values in the test data sets, and the errors that were happening when a dataset had no categorical attributes.