Checks

Checks are data integrity tests that are defined in build.yml. They are run at package build time to ensure that all consumers of the data package only receive data that comply to the given checks.

Checks can be used to prevent model drift and data deployment errors that result from using data that do not fit an expected profile.

Known issues

• support data that are larger than a pandas data frame (1GB to 10GB)

• display progress bars during checks process

• print offending line number when a check fails

• allow package users (other than the owner) to see checks and build.yml source

Syntax

• Checks are defined in a top-level dictionary called checks:

• qc.data is an automatic variable that contains the node's data in pandas data frame

• The full pandas expression syntax is supported

• Standard Python can be inlined with YAML's | operator (see below)

Functions (qc.*)

Signature	Description
check(COND)	Check that COND == true
check_column_enum(COL_REGEX, LIST_OR_LAMBDA)	Checks that all column values are in the list (and vice versa), or calls a lambda on the column
print_recnums(COL_REGEX, EXPR)	Print line numbers of rows that match EXPR.
check_column_valrange(COL_REGEX, minval=None, maxval=None, lambda_or_name=None)	Check that column values fall within [minval, maxval]. lambda_or_name is either a lambda expression applied to the matching column(s) or one of 'abs', 'count', 'mean', 'median', 'mode', 'stddev', or 'sum
check_column_regexp(COL_REGEX, REGEX)	Check that all column values match REGEX
check_column_substr(COL_REGEX, SUBSTR)	Check that all column values contain substing SUBSTR
check_column_datetime(COL_REGEX, FORMAT)	Not yet supported. Check that all column datetimes conform to FORMAT

COL_REGEX is a string literal or regular expression that matches one or more columns; the corresponding check is applied to each matching column

Example

Source data: sales.xls from Tableau Community

contents:
  transactions:
    file: sales.xls
    transform: xls
    checks: cardinality labels stats range price dates

checks:
  cardinality: |
    # verify column cardinality
    symbols = qc.data['Order Priority'].nunique()
    qc.check(symbols == 5)
  labels: |
    qc.check_column_enum(r'Order Priority', ['Low', 'High', 'Medium', 'Not Specified', 'Critical'])
    qc.print_recnums("Critical orders",  qc.data['Order Priority'] == 'Critical')
  stats: |
    # standard deviation
    stdev = qc.data['Sales'].std()
    qc.check(stdev < 3586)
  range: |
    # ensure average discount is no more than 20%
    qc.check_column_valrange('Discount', maxval=0.2, lambda_or_name='avg')
  price: |
    # check that prices are formatted properly
    qc.check_column_regexp('Unit Price','\d+\.\d+')