Anomaly detection

[JedrzejFurmann]

Anomaly detection by autoencoder model has a serious problem. The MOJO model cannot be imported (h2o.import_mojo) to use with h2o.anomaly to get reconstruction MSE values for new data. Some solution is needed to by-pass this shortcoming. This might be difficulkt though, as the h2o.predict function return the reconstruction error per feature, with categorical encoding, which cannot be generated for new data

[JedrzejFurmann]

Tasks

• go through H2O documentation
• write logic for training anomaly detection models
• write logic for storing the models
• write logic for applying the models