Forecast Evaluator - fixes SMAPE, adds MASE and Seasonal Error metrics

[wsuchy]

Fixed SMAPE and added seasonal error and MASE.
Computation is done locally due to the way the seasonal error is being calculated.

[codecov]

Codecov Report

Merging #342 into master will decrease coverage by 7.83%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #342      +/-   ##
==========================================
- Coverage   86.65%   78.82%   -7.84%     
==========================================
  Files         335      335              
  Lines       10763    10781      +18     
  Branches      344      335       -9     
==========================================
- Hits         9327     8498     -829     
- Misses       1436     2283     +847

Impacted Files	Coverage Δ
...m/salesforce/op/evaluators/EvaluationMetrics.scala	86.66% <ø> (ø)	⬆️
...salesforce/op/evaluators/OpForecastEvaluator.scala	96.42% <100%> (+6.42%)	⬆️
...stages/impl/feature/TimePeriodMapTransformer.scala	0% <0%> (-100%)	⬇️
...stages/impl/feature/GeolocationMapVectorizer.scala	0% <0%> (-100%)	⬇️
...scala/com/salesforce/op/utils/numeric/Number.scala	0% <0%> (-100%)	⬇️
...alesforce/op/aggregators/TimeBasedAggregator.scala	0% <0%> (-100%)	⬇️
...p/stages/impl/feature/OpScalarStandardScaler.scala	0% <0%> (-100%)	⬇️
.../impl/feature/UrlMapToPickListMapTransformer.scala	0% <0%> (-100%)	⬇️
...e/op/stages/impl/insights/RecordInsightsCorr.scala	0% <0%> (-98.25%)	⬇️
...e/op/stages/impl/selector/RandomParamBuilder.scala	0% <0%> (-94.45%)	⬇️
... and 57 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 afc910c...6d217d2. Read the comment docs.

[tovbinm]

if SeasonalAbsDiff is not associative, then how are you planning to reduce on it? Spark does not guarantee an order.

[wsuchy]

if SeasonalAbsDiff is not associative, then how are you planning to reduce on it? Spark does not guarantee an order.

@tovbinm Therefore I am using fold (with coealesce(1)) instead of reduce which makes me wonder if doing that math on spark makes still sense...

[tovbinm]

val (y, yHat) = rows(i)

[tovbinm]

lgtm!

[salesforce-cla]

Thanks for the contribution! Before we can merge this, we need @wsuchy to sign the Salesforce.com Contributor License Agreement.