Add implementation for Isolation Forest as an Anomaly Detection Algorithm

[JPDSousa]

Summary: As the first Anomaly Detection algorithm for the H2O provider, this commit required additional changes not only to guarantee the support for anomaly detection algorithms, but also but maintain the support for the already supported supervised algorithms.

Regarding the Isolation Forest implementation, there are some nuances:

• There is a bug in the Isolation Forest serialization/deserialization where the _min_path_length parameter is deserialized as an integer, where its type is actually long (thus leading to NumberFormatException when the value is > Integer.MAX_VALUE)
• Since this in an unsupervised algorithm, there is no need for target field, which means that:
  • getClassDistribution will return an array [NOT FRAUD, FRAUD], where FRAUD = 1 - NOT_FRAUD
  • classify checks the anomaly score and returns 0 if not fraud and 1 otherwise.
  • tests were adapted to consider the algorithm requirement where the dataset must be bigger than the sample_size parameter, otherwise no out-of-bag instances will be considered, which forbids the model from classifying instances as anomalous.

Finally, a bump to OpenML API 1.0.1 was included in this commit, which required additional changes.

[codecov[bot]]

Codecov Report

Merging #9 into master will increase coverage by 1.6%. The diff coverage is 87.75%.

@@             Coverage Diff             @@
##             master       #9     +/-   ##
===========================================
+ Coverage     75.43%   77.03%   +1.6%     
- Complexity      220      251     +31     
===========================================
  Files            22       27      +5     
  Lines           753      823     +70     
  Branches         70       75      +5     
===========================================
+ Hits            568      634     +66     
  Misses          147      147             
- Partials         38       42      +4

Impacted Files	Coverage Δ	Complexity Δ
...m/feedzai/openml/datarobot/DataRobotAlgorithm.java	100% <ø> (ø)	3 <0> (ø)	:arrow_down:
...main/java/com/feedzai/openml/h2o/H2OAlgorithm.java	100% <100%> (ø)	3 <0> (ø)	:arrow_down:
...va/com/feedzai/openml/h2o/algos/H2OBayesUtils.java	100% <100%> (ø)	6 <1> (ø)	:arrow_down:
...java/com/feedzai/openml/h2o/algos/H2ODrfUtils.java	100% <100%> (ø)	6 <1> (ø)	:arrow_down:
...nml/h2o/algos/AbstractSupervisedH2OParamUtils.java	100% <100%> (ø)	2 <2> (?)
...l/h2o/algos/AbstractUnsupervisedH2OParamUtils.java	100% <100%> (ø)	3 <3> (?)
...enml/h2o/algos/H2OGeneralizedLinearModelUtils.java	86.2% <100%> (+2.87%)	13 <1> (ø)	:arrow_down:
...zai/openml/h2o/AbstractClassificationH2OModel.java	70% <100%> (ø)	6 <2> (?)
.../com/feedzai/openml/h2o/algos/H2OXgboostUtils.java	100% <100%> (ø)	8 <1> (ø)	:arrow_down:
...dzai/openml/h2o/algos/H2OIsolationForestUtils.java	100% <100%> (ø)	6 <6> (?)
... and 14 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 f543267...3f5057c. Read the comment docs.