CODEWITHSUNDEEP
pythonapache-sparkpyspark
marc
marc

Reputation: 319

spark extract columns from string

Need help in parsing a string, where it contains values for each attribute. below is my sample string...

otherPartofString Name=<Series VR> Type=<1Ac4> SqVal=<34> conn ID=<2>

sometimes, the string can include other values with a different delimiter like

otherPartofString Name=<Series X> Type=<1B3> SqVal=<34> conn ID=<2> conn Loc=sfo dest=chc bridge otherpartofString..

the output columns will be

Name      | Type | SqVal | ID | Loc  | dest 
-------------------------------------------
Series VR | 1Ac4 | 34    | 2  | null | null
Series X  | 1B3  | 34    | 2  | sfo  | chc

Upvotes: 2

Views: 972

Answers (1)

jxc
jxc

Reputation: 13998

As we discussed, to use str_to_map function on your sample data, we can setup pairDelim and keyValueDelim to the following:

pairDelim: '(?i)>? *(?=Name|Type|SqVal|conn ID|conn Loc|dest|$)'
keyValueDelim: '=<?'

Where pariDelim is case-insensitive (?i) with an optional > followed by zero or more SPACEs, then followed by one of the pre-defined keys (we use '|'.join(keys) to generate it dynamically) or the end of string anchor $. keyValueDelim is an '=' with an optional <.

from pyspark.sql import functions as F

df = spark.createDataFrame([                                               
   ("otherPartofString Name=<Series VR> Type=<1Ac4> SqVal=<34> conn ID=<2>",),   
   ("otherPartofString Name=<Series X> Type=<1B3> SqVal=<34> conn ID=<2> conn Loc=sfo dest=chc bridge otherpartofString..",)
],["value"])

keys = ["Name", "Type", "SqVal", "conn ID", "conn Loc", "dest"]

# add the following conf for Spark 3.0 to overcome duplicate map key ERROR
#spark.conf.set("spark.sql.mapKeyDedupPolicy", "LAST_WIN")

df.withColumn("m", F.expr("str_to_map(value, '(?i)>? *(?={}|$)', '=<?')".format('|'.join(keys)))) \
    .select([F.col('m')[k].alias(k) for k in keys]) \
    .show()
+---------+----+-----+-------+--------+--------------------+
|     Name|Type|SqVal|conn ID|conn Loc|                dest|
+---------+----+-----+-------+--------+--------------------+
|Series VR|1Ac4|   34|      2|    null|                null|
| Series X| 1B3|   34|      2|     sfo|chc bridge otherp...|
+---------+----+-----+-------+--------+--------------------+

We will need to do some post-processing to the values of the last mapped-key, since there is no anchor or pattern to distinguish them from other unrelated text (this could be a problem as it might happen on any keys), please let me know if you can specify any pattern.

Edit: If using map is less efficient for case-insensitive search since it requires some expensive pre-processing, try the following:

ptn = '|'.join(keys)
df.select("*", *[F.regexp_extract('value', r'(?i)\b{0}=<?([^=>]+?)>? *(?={1}|$)'.format(k,ptn), 1).alias(k) for k in keys]).show()

In case the angle brackets < and > are used only when values or their next adjacent key contain any non-word chars, it can be simplified with some pre-processing:

df.withColumn('value', F.regexp_replace('value','=(\w+)','=<$1>')) \
    .select("*", *[F.regexp_extract('value', r'(?i)\b{0}=<([^>]+)>'.format(k), 1).alias(k) for k in keys]) \
    .show()

Edit-2: added a dictionary to handle key aliases:

keys = ["Name", "Type", "SqVal", "ID", "Loc", "dest"]

# aliases are case-insensitive and added only if exist
key_aliases = {
    'Type': [ 'ThisType', 'AnyName' ],
    'ID': ['conn ID'],
    'Loc': ['conn Loc']
}

# set up regex pattern for each key differently
key_ptns = [ (k, '|'.join([k, *key_aliases[k]]) if k in key_aliases else k) for k in keys ]  
#[('Name', 'Name'),
# ('Type', 'Type|ThisType|AnyName'),
# ('SqVal', 'SqVal'),
# ('ID', 'ID|conn ID'),
# ('Loc', 'Loc|conn Loc'),
# ('dest', 'dest')]  

df.withColumn('value', F.regexp_replace('value','=(\w+)','=<$1>')) \
    .select("*", *[F.regexp_extract('value', r'(?i)\b(?:{0})=<([^>]+)>'.format(p), 1).alias(k) for k,p in key_ptns]) \
    .show()
+--------------------+---------+----+-----+---+---+----+
|               value|     Name|Type|SqVal| ID|Loc|dest|
+--------------------+---------+----+-----+---+---+----+
|otherPartofString...|Series VR|1Ac4|   34|  2|   |    |
|otherPartofString...| Series X| 1B3|   34|  2|sfo| chc|
+--------------------+---------+----+-----+---+---+----+

Upvotes: 3

Related Questions