Preserve data on non-active items for PopModel

[chEbba]

Model:

• 100 items (1-100) with properties
• 100 main events.
• Events during last days (backfillField.duration) only for top 10 Items (91-100)
• Engine with "recsModel": "all"

Training: https://github.com/PredictionIO/template-scala-parallel-universal-recommendation/blob/v0.2.3/src/main/scala/URAlgorithm.scala#L152-L168

• As we have "recsModel": "all" (popular=true) we calculate PopModel
• PopModel using events only for backfillField.duration, so we got pop results only for items 91-100
• Then we combine data form allFields with popModel

      val pm = if (pms._1.nonEmpty && pms._2.nonEmpty) {
          val newPM = pms._1.head.fields + (backfillFieldName -> JDouble(pms._2.head))
          PropertyMap(newPM, pms._1.head.firstUpdated, DateTime.now())
        } else if (pms._2.nonEmpty) PropertyMap(Map(backfillFieldName -> JDouble(pms._2.head)), DateTime.now(), DateTime.now())
        else PropertyMap( Map.empty[String, JValue], DateTime.now, DateTime.now) // some error????

• As we don't have popularity for items 1-90 we get empty properties for all that items

Query:

• If we use date filters we will select only from items 91-100 (as all other items missed date properties)
• Boost/filter by fields works only on items 91-100

Solution:

• Preserve data on non-active items