PySpark DataFrame: change cell value based on min / max condition in another column

Question

I have the following pySpark data file:

+------------------+------------------+--------------------+--------------+-------+
|              col1|              col2|                col3|             X|      Y|
+------------------+------------------+--------------------+--------------+-------+
|2.1729247374294496| 3.558069532647046|   6.607603368496324|             1|   null|
|0.2654841575294071|1.2633077949463256|0.023578679968183733|             0|   null|
|0.4253301781296708|3.4566490739823483| 0.11711202266039554|             3|   null|
| 2.608497168338446| 3.529397129549324|   0.373034222141551|             2|   null|
+------------------+------------------+--------------------+--------------+-------+

This is a fairly simple operation, and I can easily do it with pandas. However, I need to do this using only pySpark.

I want to do the following (I will write in the form of pseudocode):

In the line where col3 == max (col3), change Y from zero to 'K'

On the remaining lines in the line, where col1 == max (col1), change Y from zero to 'Z'

On the remaining lines in the line where col1 == min (col1), change Y from zero to 'U'

In the remaining line: change the value of Y from zero to "I".

Therefore, the expected result:

+------------------+------------------+--------------------+--------------+-------+
|              col1|              col2|                col3|             X|      Y|
+------------------+------------------+--------------------+--------------+-------+
|2.1729247374294496| 3.558069532647046|   6.607603368496324|             1|      K|
|0.2654841575294071|1.2633077949463256|0.023578679968183733|             0|      U|
|0.4253301781296708|3.4566490739823483| 0.11711202266039554|             3|      I|
| 2.608497168338446| 3.529397129549324|   0.373034222141551|             2|      Z|
+------------------+------------------+--------------------+--------------+-------+

After that, I need to use this table to find another table:

+--------------------+--------+-----+------------------+--------------+------------+
|                  x1|      x2|   x3|                x4|             X|           d|
+--------------------+--------+-----+------------------+--------------+------------+
|0057f68a-6330-42a...|    2876|   30| 5.989999771118164|             0|    20171219|
|05cc0191-4ee4-412...|  108381|   34|24.979999542236328|             3|    20171219|
|06f353af-e9d3-4d0...|  118798|   34|               0.0|             3|    20171219|
|0c69b607-112b-4f3...|   20993|   34|               0.0|             0|    20171219|
|0d1b52ba-1502-4ff...|   23817|   34|               0.0|             0|    20171219|

. Y , X ( Y , X, X ).

UPD: , , , :

+------------------+------------------+--------------------+--------------+-------+
|              col1|              col2|                col3|             X|      Y|
+------------------+------------------+--------------------+--------------+-------+
| 2.608497168338446| 3.558069532647046|   6.607603368496324|             1|   null|
|0.2654841575294071|1.2633077949463256|0.023578679968183733|             0|   null|
|0.4253301781296708|3.4566490739823483| 0.11711202266039554|             3|   null|
|2.1729247374294496| 3.529397129549324|   0.373034222141551|             2|   null|
+------------------+------------------+--------------------+--------------+-------+

0 max ('col3') max ('col1').

, :

0 "K"

3 "Z" ( (0 "K" 3 max ('col1')

1 "U"

2 "I"

1 "" .

Answer 1

:

from pyspark.sql import functions as F

df = spark.createDataFrame([
    (2.1729247374294496,  3.558069532647046,    6.607603368496324, 1),
    (0.2654841575294071, 1.2633077949463256, 0.023578679968183733, 0),
    (0.4253301781296708, 3.4566490739823483,  0.11711202266039554, 3),
    (2.608497168338446,  3.529397129549324,    0.373034222141551, 2)
], ("col1", "col2", "col3", "x"))

min1, max1, max3 = df.select(F.min("col1"), F.max("col1"), F.max("col3")).first()

when:

y = (F.when(F.col("col3") == max3, "K")  # In row where col3 == max(col3), change Y from null to 'K'
    .when(F.col("col1") == max1, "Z")    # In the remaining rows, in the row where col1 == max(col1), change Y from null to 'Z'
    .when(F.col("col1") == min1, "U")    # In the remaining rows, in the row where col1 == min(col1), change Y from null to 'U'
    .otherwise("I"))                     # In the remaining row: change Y from null to 'I'

df_with_y = df.withColumn("y", y)


df_with_y.show()
# +------------------+------------------+--------------------+---+---+
# |              col1|              col2|                col3|  x|  y|
# +------------------+------------------+--------------------+---+---+
# |2.1729247374294496| 3.558069532647046|   6.607603368496324|  1|  K|
# |0.2654841575294071|1.2633077949463256|0.023578679968183733|  0|  U|
# |0.4253301781296708|3.4566490739823483| 0.11711202266039554|  3|  I|
# | 2.608497168338446| 3.529397129549324|   0.373034222141551|  2|  Z|
# +------------------+------------------+--------------------+---+---+

Y , X

df_with_y.select("x", "Y").join(df2, ["x"])

y :

df_ = spark.createDataFrame([
    (2.1729247374294496,  3.558069532647046,    6.607603368496324, 1, "G"),
    (0.2654841575294071, 1.2633077949463256, 0.023578679968183733, 0, None),
    (0.4253301781296708, 3.4566490739823483,  0.11711202266039554, 3, None),
    (2.608497168338446,  3.529397129549324,    0.373034222141551, 2, None)
], ("col1", "col2", "col3", "x", "y"))

min1_, max1_, max3_ = df.filter(F.col("y").isNull()).select(F.min("col1"), F.max("col1"), F.max("col3")).first()

y_ = (F.when(F.col("col3") == max3_, "K") 
    .when(F.col("col1") == max1_, "Z")
    .when(F.col("col1") == min1_, "U") 
    .otherwise("I"))

df_.withColumn("y", F.coalesce(F.col("y"), y_)).show()


# +------------------+------------------+--------------------+---+---+
# |              col1|              col2|                col3|  x|  y|
# +------------------+------------------+--------------------+---+---+
# |2.1729247374294496| 3.558069532647046|   6.607603368496324|  1|  G|
# |0.2654841575294071|1.2633077949463256|0.023578679968183733|  0|  U|
# |0.4253301781296708|3.4566490739823483| 0.11711202266039554|  3|  I|
# | 2.608497168338446| 3.529397129549324|   0.373034222141551|  2|  K|
# +------------------+------------------+--------------------+---+---+

, :

threshold = 0.0000001 # Choose appropriate 

y_t = (F.when(F.abs(F.col("col3") - max3) < threshold, "K")  # In row where col3 == max(col3), change Y from null to 'K'
    .when(F.abs(F.col("col1") - max1) < threshold, "Z")    # In the remaining rows, in the row where col1 == max(col1), change Y from null to 'Z'
    .when(F.abs(F.col("col1") - min1) < threshold, "U")    # In the remaining rows, in the row where col1 == min(col1), change Y from null to 'U'
    .otherwise("I"))                     # In the remaining row: change Y from null to 'I'

df.withColumn("y", y_t).show()
# +------------------+------------------+--------------------+---+---+
# |              col1|              col2|                col3|  x|  y|
# +------------------+------------------+--------------------+---+---+
# |2.1729247374294496| 3.558069532647046|   6.607603368496324|  1|  K|
# |0.2654841575294071|1.2633077949463256|0.023578679968183733|  0|  U|
# |0.4253301781296708|3.4566490739823483| 0.11711202266039554|  3|  I|
# | 2.608497168338446| 3.529397129549324|   0.373034222141551|  2|  Z|
# +------------------+------------------+--------------------+---+---+