google-places-json-to-dataframe-spark.scala

val conf = new SparkConf().setAppName("SparkTest").setMaster("local[1]")
val sparkContext = new SparkContext(conf)

val sqlContext = new SQLContext(sparkContext)
import sqlContext.implicits.{getClass => _, _}


    val fileNameGoogle = "/google-places-data.json" // 1 json places per line
    val locationsGoogle = sqlContext.read.json(getClass.getResource(fileNameGoogle).getPath).
      toDF().
      withColumn("address_components", explode($"address_components")).
      select(
        $"address_components.long_name",
        $"address_components.short_name",
        $"address_components.types".getItem(0).alias("type"),
        $"geometry.location.lat",
        $"geometry.location.lng",
        $"place_id".alias("id")
      ).
      groupBy("id","lat","lng").
      pivot("type").
      agg(first("long_name"), first("short_name")).
      select(
        $"id".alias("googleId"),
        $"street_number_first(long_name)()".alias("houseNumber"),
        $"route_first(long_name)()".alias("street"),
        $"sublocality_level_1_first(long_name)()".alias("area"),
        $"sublocality_level_2_first(long_name)()".alias("municipality"),
        $"locality_first(long_name)()".alias("city"),
        $"administrative_area_level_2_first(long_name)()".alias("nuts3"),
        $"neighborhood_first(long_name)()".alias("neighbourhood"),
        $"administrative_area_level_1_first(long_name)()".alias("nuts2"),
        $"administrative_area_level_1_first(short_name)()".alias("nuts2Code"),
        $"country_first(short_name)()".alias("countryCode"),
        $"country_first(long_name)()".alias("country"),
        $"postal_code_first(long_name)()".alias("postalCode"),
        $"postal_code_prefix_first(long_name)()".alias("postalCodePrefix"),
        $"lat".alias("latitude"),
        $"lng".alias("longitude")
      ).
      show()

Probably not the most efficient way, but semi-readable, cotains many learnings about sparkSQL and it actually works in spark 1.6.1