I have a large dataset and some columns have String data-type. Because of typo mistake, some of the cells have None values but written in different styles (with small or capital letters, with or without space, with or without bracket, etc).
I want to count the No. of all those values (excluding Null values) in all columns. A sample dataset is below:
data = [("A", "None", 1), \
("A", "(None)", 2), \
("[None", "none", 3), \
("(none]", "[None]", 4), \
("A", "(none)", 5), \
("A", "(none", 6), \
("A", "none ", 7), \
(" NOne ", None, None), \
]
# Create DataFrame
columns= ["col_1", "col_2", "Number"]
df = spark.createDataFrame(data = data, schema = columns)
The expected result is:
{'col_1': 3, 'col_2': 7, 'Number': 0}
Any idea how to do that by PySpark?
CodePudding user response:
The logic is:
- Use regex to remove all kinds of opening brackets and closing brackets from start and end of the column value.
- Trim extra spaces, convert to lower and compare to "none".
- Count the filtered records for each column.
count_result = {}
for c in df.columns:
count_result[c] = df.select(c).filter(F.lower(F.trim(F.regexp_replace(c, r"(?:^\[|^\(|^\<|^\{|\]$|\)$|\>$|\}$)", ""))) == "none") \
.count()
print(count_result)
Output:
{'col_1': 3, 'col_2': 7, 'Number': 0}