使用scala的spark數據幀中一行中的值計數

Question

我有一個數據框。 它包含跨不同銷售網點的不同項目的銷售額。 下面顯示的數據框僅顯示了幾個銷售網點中的少數項目。 每件商品每天銷售 100 件商品的基准。 對於每件售出超過 100 件的商品，將其標記為“是”，將低於 100 件的商品標記為“否”

val df1 = Seq(
("Mumbai", 90,  109, , 101, 78, ............., "No", "Yes", "Yes", "No", .....),
("Singapore", 149,  129, , 201, 107, ............., "Yes", "Yes", "Yes", "Yes", .....),
("Hawaii", 127,  101, , 98, 109, ............., "Yes", "Yes", "No", "Yes", .....),
("New York", 146,  130, , 173, 117, ............., "Yes", "Yes", "Yes", "Yes", .....),
("Los Angeles", 94,  99, , 95, 113, ............., "No", "No", "No", "Yes", .....),
("Dubai", 201,  229, , 265, 317, ............., "Yes", "Yes", "Yes", "Yes", .....),
("Bangalore", 56,  89, , 61, 77, ............., "No", "No", "No", "No", .....))
.toDF("Outlet","Boys_Toys","Girls_Toys","Men_Shoes","Ladies_shoes", ............., "BT>100", "GT>100", "MS>100", "LS>100", .....)

現在，我想添加一列“Count_of_Yes”，其中對於每個銷售網點（每行），“Count_of_Yes”列的值將是該行中“是”的總數。 如何遍歷每一行以獲得 Yes 的計數？

我預期的數據框應該是

val output_df = Seq(
("Mumbai", 90,  109, , 101, 78, ............., "No", "Yes", "Yes", "No", ....., 2),
("Singapore", 149,  129, , 201, 107, ............., "Yes", "Yes", "Yes", "Yes", ....., 4),
("Hawaii", 127,  101, , 98, 109, ............., "Yes", "Yes", "No", "Yes", ....., 3),
("New York", 146,  130, , 173, 117, ............., "Yes", "Yes", "Yes", "Yes", ....., 4),
("Los Angeles", 94,  99, , 95, 113, ............., "No", "No", "No", "Yes", ....., 1),
("Dubai", 201,  229, , 265, 317, ............., "Yes", "Yes", "Yes", "Yes", ....., 4),
("Bangalore", 56,  89, , 61, 77, ............., "No", "No", "No", "No", ....., 0))
.toDF("Outlet","Boys_Toys","Girls_Toys","Men_Shoes","Ladies_shoes", ............., "BT>100", "GT>100", "MS>100", "LS>100", ....., "Count_of_Yes")

Answer 1

可以列的選擇的列表轉換成Array的1 s（對於“是”）和0 s（對於“否”），並總結與陣列元件aggregate在SQL表達式中使用selectExpr ，如下圖所示：

val df = Seq(
  (1, 120, 80, 150, "Y", "N", "Y"),
  (2, 50, 90, 110, "N", "N", "Y"),
  (3, 70, 160, 90, "N", "Y", "N")
).toDF("id", "qty_a", "qty_b", "qty_c", "over100_a", "over100_b", "over100_c")

val cols = df.columns.filter(_.startsWith("over100_"))

df.
  withColumn("arr", array(cols.map(c => when(col(c) === "Y", 1).otherwise(0)): _*)).
  selectExpr("*", "aggregate(arr, 0, (acc, x) -> acc + x) as yes_count").
  show
// +---+-----+-----+-----+---------+---------+---------+---------+---------+
// | id|qty_a|qty_b|qty_c|over100_a|over100_b|over100_c|      arr|yes_count|
// +---+-----+-----+-----+---------+---------+---------+---------+---------+
// |  1|  120|   80|  150|        Y|        N|        Y|[1, 0, 1]|        2|
// |  2|   50|   90|  110|        N|        N|        Y|[0, 0, 1]|        1|
// |  3|   70|  160|   90|        N|        Y|        N|[0, 1, 0]|        1|
// +---+-----+-----+-----+---------+---------+---------+---------+---------+

---

可替代地，使用explode和groupBy/agg到求和Array元素：

df.
  withColumn("arr", array(cols.map(c => when(col(c) === "Y", 1).otherwise(0)): _*)).
  withColumn("flattened", explode($"arr")).
  groupBy("id").agg(sum($"flattened").as("yes_count"))

Answer 2

如何遍歷每一行以獲得 Yes 的計數？ 您可以使用映射轉換來轉換每個記錄。 所以在你的情況下 df.map() 應該有代碼來計算 YES 的數量並發出一個具有這個附加列的新記錄。

偽代碼如下——

df.map(count number of YES and append that at the end of the string")