* streaming handler is still useful for spark, though there is flink as alternative * RDD is also useful for transform especially for non-structure data * there are many SQL products in market like Drill/Impala, but spark is more powerful for distributed deployment as far as I know * we never used spark for AI training, but use keras/pytorch which are pretty easy for development a model.
Perhaps you should try other systems in the market first, that will give an unbiased view of databricks and SPARK being just over glamourised tool. The hope of extending SPARK with a separate easy to use query engine for deep learning and other AI systems is gone now with Ray, SPARK community now just defends the lack of support, and direction in this matter largely, which is a joke.
--------------------------------------------------------------------- To unsubscribe e-mail: user-unsubscr...@spark.apache.org