SSB Fag-fellesfunksjoner i Python¶
A place for “loose, small functionality” produced at Statistics Norway in Python. Functionality might start here, if it is to be used widely within the organization, but later be moved to bigger packages if they “grow out of proportions”.
Team: ssb-pythonistas¶
We are a team of statisticians which hope to curate and generalize functionality which arizes from specific needs in specific production-environments. We try to take responsibility for this functionality to be generalized and available to all of statistics Norway through this package.
Contributing¶
Please make contact with one of our team members, to see if you can join, or how to send in a PR for approval into the package.
Installing¶
poetry add ssb-fagfunksjoner
Usage¶
Environment / Pathing¶
Check if you are on Dapla or in prodsone.
from fagfunksjoner import check_env
check_env()
Navigate to the root of your project and back again. Do stuff while in root, like importing local functions.
from fagfunksjoner import ProjectRoot
with ProjectRoot():
... # Do your local imports here...
Sasfiles¶
Setting up password with saspy
from fagfunksjoner.prodsone import saspy_ssb
saspy_ssb.set_password() # Follow the instructions to set the password
saspy_ssb.saspy_df_from_path("path")
Logger that follows SSB standards¶
import logging
from fagfunksjoner import StatLogger
# Ved å opprette StatLogger så "hijacker" den den vanlige loggeren
root_logger = StatLogger(log_file="custom_log_file.log")
# I tillegg sørger vi for at den ikke blir ryddet bort av Python, ved å assigne den til en variabel?
logger = logging.getLogger(__name__)
logger.info("This is an info message")
Export XMLs that can be imported into the KLASS UI¶
from fagfunksjoner import make_klass_xml_codelist
make_klass_xml_codelist(path="kjoenn.xml",
codes=["1", "2"],
names_bokmaal=["Mann", "Kvinne"])
Round data UP¶
import pandas as pd
from fagfunksjoner import round_up
print(round(2.5, 0), round_up(2.5, 0))
round_up(pd.Series([1.5, 2.5, 3.5]), 0) # Datatype blir Int64 når man runder til 0 desimaler
round_up(pd.Series([1.15, 2.15, 3.15]), 1) # Datatype blir Float64 når man runder til mer enn 0 desimaler
df = pd.DataFrame(
{"col1": [1.5, 2.5, 1.2345, 1.2355],
"col2": [3.5, 4.5, 5.6789, 6.7891]}
).astype({"col1": "Float64", "col2": "Float64"})
rounded = round_up(df, decimal_places=0, col_names="col1") # Avrunder kun col1, den endrer datatype til Int64
rounded2 = round_up(df, col_names={"col1": 1, "col2": 2}) # Avrunder col1 til 1 desimal, col2 til 2 desimaler
Aggregation / Categories¶
Aggregate on all exclusive combinations of codes in certain columns (maybe before sending to statbank? Like proc means?)
from fagfunksjoner import all_combos_agg
ialt_koder = {
"skolefylk": "01-99",
"almyrk": "00",
"kjoenn_t": "0",
"sluttkomp": "00",
}
kolonner = list(ialt_koder.keys())
tab = all_combos_agg(vgogjen,
groupcols=kolonner,
aggargs={'antall': sum},
fillna_dict=ialt_koder)
To aggregate on NON-EXCLUSIVE combinations of codes in certain columns, use the slightly less process-effective
from fagfunksjoner import all_combos_agg_inclusive
category_mappings = {
"Alder": {
"15-24": range(15, 25),
"25-34": range(25, 35),
"35-44": range(35, 45),
"45-54": range(45, 55),
"55-66": range(55, 67),
"15-21": range(15, 22),
"22-30": range(22, 31),
"31-40": range(31, 41),
"41-50": range(41, 51),
"51-66": range(51, 67),
"15-30": range(15, 31),
"31-45": range(31, 46),
"46-66": range(46, 67),
},
"syss_student": {
"01": ["01", "02"],
"02": ["03", "04"],
"03": ["02"],
"04": ["04"],
},
"Kjonn": {
"Menn": ["1"],
"Kvinner": ["2"],
}
}
totalcodes = {
"Alder": "Total",
"syss_student": "Total",
"Kjonn": "Begge"
}
all_combos_agg_inclusive(
synthetic_data,
groupcols = [],
category_mappings=category_mappings,
totalcodes=totalcodes,
valuecols = ["n"],
aggargs={"n": "sum"},
grand_total=True)
“Formats” like in SAS¶
Perform mapping using SsbFormat. Behaves like a dictionary. Has functionality for mapping ranges and ‘other’-category and detecting different types of NaN-values. Does not handle non-exclusive / overlapping categories, please only use for exclusive categories.
from fagfunksjoner import SsbFormat
age_frmt = {
'low-18': '-18',
'19-25': '19-25',
'26-35': '26-35',
'36-45': '36-45',
'46-55': '46-55',
'56-high': '56+',
'other': 'missing'
}
# convert dictionary to SsbFormat
ssb_age_frmt = SsbFormat(age_frmt)
# perform mapping of age using ranges in format.
df['age_group'] = df['age'].map(ssb_age_frmt)
print(df['age_group'].value_counts())
# save format
from fagfunksjoner.formats import store_format
store_format(path+'format_name_p2025-02.json')
# or
# NB! after performing range mapping using SsbFormat. The dictionary will be long. You should save a short version. Inspect the dictionary before saving/storing.
ssb_age_frmt.store(path + 'format_name_p2025-02.json', force=True)
# read format/import format (dictionary saved as .json) as SsbFormat
from fagfunksjoner.formats import get_format
some_frmt = get_format(path+'format_name.json')
Opening archive-files based on Datadok-api in prodsone¶
We have “flat files”, which are not comma seperated. These need metadata to correctly open. In SAS we do this with “lastescript”. But there is an API to old Datadok in prodsone, so these functions let you just specify a path, and attempt to open the flat files directly into pandas, with the metadata also available.
from fagfunksjoner import open_path_datadok
archive_object = open_path_datadok("$TBF/project/arkiv/filename/g2022g2023")
# The object now has several important attributes
archive_object.df # The Dataframe of the archived data
archive_object.metadata_df # Dataframe representing metadata
archive_object.codelist_df # Dataframe representing codelists
archive_object.codelist_dict # Dict of codelists
archive_object.names # Column names in the archived data
archive_object.datatypes # The datatypes the archivdata ended up having?
archive_object.widths # Width of each column in the flat file
Operation to Oracle database¶
Remember that any credidential values to the database should not be stored in our code. Possibly use python-dotenv package to make this easier.
Example for a normal select query where we expect not too many records:
import os
import pandas as pd
from doteng import load_dotenv
from fagfunksjoner.prodsone import Oracle
load_dotenv()
query = "select vare, pris from my_db_table"
ora = Oracle(pw=os.getenv("my-secret-password"),
db=os.getenv("database-name"))
df = pd.DataFrame(ora.select(sql=query))
ora.close()
Example for a select query where possibly many records:
import os
import pandas as pd
from doteng import load_dotenv
from fagfunksjoner.prodsone import Oracle
load_dotenv()
query = "select vare, pris from my_db_table"
ora = Oracle(pw=os.getenv("my-secret-password"),
db=os.getenv("database-name"))
df = pd.DataFrame(ora.selectmany(sql=query, batchsize=10000))
ora.close()
Example for inserting new records into database(note that ordering of the columns in sql query statement and data are important):
import os
import pandas as pd
from doteng import load_dotenv
from fagfunksjoner.prodsone import Oracle
load_dotenv()
df = pd.DataFrame(
{
"vare": ["banan", "eple"],
"pris": [11, 10]
}
)
data = list(df.itertuples(index=False, name=None))
query = "insert into my_db_table(vare, pris) values(:vare, :pris)"
ora = Oracle(pw=os.getenv("my-secret-password"),
db=os.getenv("database-name"))
ora.insert_or_update(sql=query, update=data)
ora.close()
Example for updating records in the database(note that ordering of the columns in sql query statement and data are important. It is also important that the query doesn’t update other records than it should. Having some kind of ID to the records will be very usefull!):
import os
import pandas as pd
from doteng import load_dotenv
from fagfunksjoner.prodsone import Oracle
load_dotenv()
df = pd.DataFrame(
{
"id": ["12345", "54321"]
"vare": ["banan", "eple"],
"pris": [11, 10]
}
)
data = list(df[["vare", "pris", "id"]].itertuples(index=False, name=None))
query = "update my_db_table set vare = :vare, pris = :pris where id = :id"
ora = Oracle(pw=os.getenv("my-secret-password"),
db=os.getenv("database-name"))
ora.insert_or_update(sql=query, update=data)
ora.close()
It also support context manager. This is handy when working with big data, and you then have to work more lazy. Or you want to do multiple operations to several tables without closing the connections. Or any other reasons… An easy case; reading large data from database and write it to a parquet file, in batches:
import os
import pandas as pd
import pyarrow as pa
import pyarrow.parquet as pq
from doteng import load_dotenv
from fagfunksjoner.prodsone import Oracle, OraError
load_dotenv()
select_query = "select vare, pris from my_db_table"
parquet_write_path = "write/to/path/datafile.parquet"
with pq.ParquetWriter(parquet_write_path) as pqwriter: # pyarrow schema might be needed
try:
# will go straight to cursor
with Oracle(pw=os.getenv("my-secret-password"),
db=os.getenv("database-name")) as concur:
concur.execute(select_query)
cols = [c[0].lower() for c in cur.description]
while True:
rows = cur.fetchmany(10_000) # 10.000 rows per batch
if not rows:
break
else:
data = [dict(zip(cols, row)) for row in rows]
tab = pa.Table.from_pylist(data)
# this will write data to one row group per batch
pqwriter.write_table(tab)
except OraError as error:
raise error
Contributing¶
Contributions are very welcome. To learn more, see the Contributor Guide.
License¶
Distributed under the terms of the MIT license, SSB Fagfunksjoner is free and open source software.
Issues¶
If you encounter any problems, please file an issue along with a detailed description.
Credits¶
This project was generated from Statistics Norway’s SSB PyPI Template.