Can you tell me the difference between narrow and wide transformations?

In Spark, narrow transformations are those where each input partition contributes to at most one output partition, so no data shuffling is required. Examples include map, filter, and sample. Wide transformations, by contrast, require data to be shuffled across the cluster because each output partition may receive data from many input partitions. Examples are reduceByKey, groupByKey, and join. Narrow transformations are more efficient because they can be pipelined and executed without network I/O, whereas wide transformations incur the cost of shuffling data between executors.