Goodtables is a framework to validate tabular data. It can check the structure of your data (e.g. all rows have the same number of columns), and its contents (e.g. all dates are valid).
[Important Notice]
goodtables
was renamed tofrictionless
since version 3. The framework got various improvements and was extended to be a complete data solution. The change in not breaking for the existing software so no actions are required. Please read the Migration Guide to start working with Frictionless for Python.
- Structural checks: Ensure that there are no empty rows, no blank headers, etc.
- Content checks: Ensure that the values have the correct types ("string", "number", "date", etc.), that their format is valid ("string must be an e-mail"), and that they respect the constraints ("age must be a number greater than 18").
- Support for multiple tabular formats: CSV, Excel files, LibreOffice, Data Package, etc.
- Parallelized validations for multi-table datasets
- Command line interface
For faster goodtables-combatible Pandas dataframes validation take a look at https://github.com/ezwelty/goodtables-pandas-py
pip install goodtables
pip install goodtables[ods] # If you need LibreOffice's ODS file support
goodtables data.csv
Use goodtables --help
to see the different options.
from goodtables import validate
report = validate('invalid.csv')
report['valid'] # false
report['table-count'] # 1
report['error-count'] # 3
report['tables'][0]['valid'] # false
report['tables'][0]['source'] # 'invalid.csv'
report['tables'][0]['errors'][0]['code'] # 'blank-header'
You can read a more in depth explanation on using goodtables with Python on the developer documentation section. Check also the examples folder for other examples.
Goodtables validates your tabular dataset to find structural and content
errors. Consider you have a file named invalid.csv
. Let's validate it:
report = validate('invalid.csv')
We could also pass a remote URI instead of a local path. It supports CSV, XLS, XLSX, ODS, JSON, and all other formats supported by the tabulator library.
The validation report follows the JSON Schema defined on goodtables/schemas/report.json.
The output of the validate()
method is a report dictionary. It includes
information if the data was valid, count of errors, list of table reports, which
individual checks failed, etc. A report will be looking like this:
{
"time": 0.009,
"error-count": 1,
"warnings": [
"Table \"data/invalid.csv\" inspection has reached 1 error(s) limit"
],
"preset": "table",
"valid": false,
"tables": [
{
"errors": [
{
"row-number": null,
"message": "Header in column 3 is blank",
"row": null,
"column-number": 3,
"code": "blank-header"
}
],
"error-count": 1,
"headers": [
"id",
"name",
"",
"name"
],
"scheme": "file",
"row-count": 2,
"valid": false,
"encoding": "utf-8",
"time": 0.007,
"schema": null,
"format": "csv",
"source": "data/invalid"
}
],
"table-count": 1
}
The errors are divided in one of the following categories:
source
- data can't be loaded or parsedstructure
- general tabular errors like duplicate headersschema
- error of checks against Table Schemacustom
- custom checks errors
Check is a main validation actor in goodtables. The list of enabled checks can
be changed using checks
and skip_checks
arguments. Let's explore the options
on an example:
report = validate('data.csv') # by default structure and schema (if available) checks
report = validate('data.csv', checks=['structure']) # only structure checks
report = validate('data.csv', checks=['schema']) # only schema (if available) checks
report = validate('data.csv', checks=['bad-headers']) # check only 'bad-headers'
report = validate('data.csv', skip_checks=['bad-headers']) # exclude 'bad-headers'
By default a dataset will be validated against all available Data Quality Spec
errors. Some checks can be unavailable for validation. For example, if the
schema isn't provided, only the structure
checks will be done.
Goodtables support different formats of tabular datasets. They're called presets. A tabular dataset is some data that can be split in a list of data tables, as:
We can change the preset using the preset
argument for validate()
. By
default, it'll be inferred from the source, falling back to table
. To validate
a data package, we can do:
report = validate('datapackage.json') # implicit preset
report = validate('datapackage.json', preset='datapackage') # explicit preset
This will validate all tabular resources in the datapackage.
It's also possible to validate a list of files using the "nested" preset. To do
so, the first argument to validate()
should be a list of dictionaries, where
each key in the dictionary is named after a parameter on validate()
. For example:
report = validate([{'source': 'data1.csv'}, {'source': 'data2.csv'}]) # implicit preset
report = validate([{'source': 'data1.csv'}, {'source': 'data2.csv'}], preset='nested') # explicit preset
Is similar to:
report_data1 = validate('data1.csv')
report_data2 = validate('data2.csv')
The difference is that goodtables validates multiple tables in parallel, so calling using the "nested" preset should run faster.
Base report errors are standardized and described in Data Quality Spec.
The basic checks can't be disabled, as they deal with goodtables being able to read the files.
check | description |
---|---|
io-error | Data reading error because of IO error. |
http-error | Data reading error because of HTTP error. |
source-error | Data reading error because of not supported or inconsistent contents. |
scheme-error | Data reading error because of incorrect scheme. |
format-error | Data reading error because of incorrect format. |
encoding-error | Data reading error because of an encoding problem. |
These checks validate that the structure of the file are valid.
check | description |
---|---|
blank-header | There is a blank header name. All cells in the header row must have a value. |
duplicate-header | There are multiple columns with the same name. All column names must be unique. |
blank-row | Rows must have at least one non-blank cell. |
duplicate-row | Rows can't be duplicated. |
extra-value | A row has more columns than the header. |
missing-value | A row has less columns than the header. |
These checks validate the contents of the file. To use them, you need to pass a Table Schema. If you don't have a schema, goodtables can infer it if you use the infer_schema
option.
If your schema only covers part of the data, you can use the infer_fields
to infer the remaining fields.
Lastly, if the order of the fields in the data is different than in your schema, enable the order_fields
option.
check | description |
---|---|
schema-error | Schema is not valid. |
non-matching-header | The header's name in the schema is different from what's in the data. |
extra-header | The data contains a header not defined in the schema. |
missing-header | The data doesn't contain a header defined in the schema. |
type-or-format-error | The value can’t be cast based on the schema type and format for this field. |
required-constraint | This field is a required field, but it contains no value. |
pattern-constraint | This field value's should conform to the defined pattern. |
unique-constraint | This field is a unique field but it contains a value that has been used in another row. |
enumerable-constraint | This field value should be equal to one of the values in the enumeration constraint. |
minimum-constraint | This field value should be greater or equal than constraint value. |
maximum-constraint | This field value should be less or equal than constraint value. |
minimum-length-constraint | A length of this field value should be greater or equal than schema constraint value. |
maximum-length-constraint | A length of this field value should be less or equal than schema constraint value. |
check | description |
---|---|
blacklisted-value | Ensure there are no cells with the blacklisted values. |
deviated-value | Ensure numbers are within a number of standard deviations from the average. |
foreign-key | Ensure foreign keys are valid within a data package |
sequential-value | Ensure numbers are sequential. |
truncated-value | Detect values that were potentially truncated. |
custom-constraint | Defines a constraint based on the values of other columns (e.g. value * quantity == total ). |
Sometimes we have to check for some values we don't want to have in out dataset. It accepts following options:
option | type | description |
---|---|---|
column | int/str | Column number or name |
blacklist | list of str | List of blacklisted values |
Consider the following CSV file:
id,name
1,John
2,bug
3,bad
5,Alex
Let's check that the name
column doesn't contain rows with bug
or bad
:
from goodtables import validate
report = validate('data.csv', checks=[
{'blacklisted-value': {'column': 'name', 'blacklist': ['bug', 'bad']}},
])
# error on row 3 with code "blacklisted-value"
# error on row 4 with code "blacklisted-value"
This check helps to find outlines in a column containing positive numbers. It accepts following options:
option | type | description |
---|---|---|
column | int/str | Column number or name |
average | str | Average type, either "mean", "median" or "mode" |
interval | int | Values must be inside range average ± standard deviation * interval |
Consider the following CSV file:
temperature
1
-2
7
0
1
2
5
-4
100
8
3
We use median
to get an average of the column values and allow interval of 3 standard deviations. For our case median is 2.0
and standard deviation is 29.73
so all valid values must be inside the [-87.19, 91.19]
interval.
report = validate('data.csv', checks=[
{'deviated-value': {'column': 'temperature', 'average': 'median', 'interval': 3}},
])
# error on row 10 with code "deviated-value"
We support here relative paths. It MUST be used only for trusted data sources.
This check validate foreign keys within a data package. Consider we have a data package defined below:
DESCRIPTOR = {
'resources': [
{
'name': 'cities',
'data': [
['id', 'name', 'next_id'],
[1, 'london', 2],
[2, 'paris', 3],
[3, 'rome', 4],
# [4, 'rio', None],
],
'schema': {
'fields': [
{'name': 'id', 'type': 'integer'},
{'name': 'name', 'type': 'string'},
{'name': 'next_id', 'type': 'integer'},
],
'foreignKeys': [
{
'fields': 'next_id',
'reference': {'resource': '', 'fields': 'id'},
},
{
'fields': 'id',
'reference': {'resource': 'people', 'fields': 'label'},
},
],
},
}, {
'name': 'people',
'data': [
['label', 'population'],
[1, 8],
[2, 2],
# [3, 3],
# [4, 6],
],
},
],
}
Running goodtables
on it will raise a few foreign-key
errors because we have commented some rows in the data package's data:
report = validate(DESCRIPTOR, checks=['structure', 'schema', 'foreign-key'])
print(report)
{'error-count': 2,
'preset': 'datapackage',
'table-count': 2,
'tables': [{'datapackage': '...',
'error-count': 2,
'errors': [{'code': 'foreign-key',
'message': 'Foreign key "[\'next_id\']" violation in '
'row 4',
'message-data': {'fields': ['next_id']},
'row-number': 4},
{'code': 'foreign-key',
'message': 'Foreign key "[\'id\']" violation in row 4',
'message-data': {'fields': ['id']},
'row-number': 4}],
'format': 'inline',
'headers': ['id', 'name', 'next_id'],
'resource-name': 'cities',
'row-count': 4,
'schema': 'table-schema',
'source': 'inline',
'time': 0.031,
'valid': False},
{'datapackage': '...',
'error-count': 0,
'errors': [],
'format': 'inline',
'headers': ['label', 'population'],
'resource-name': 'people',
'row-count': 3,
'source': 'inline',
'time': 0.038,
'valid': True}],
'time': 0.117,
'valid': False,
'warnings': []}
It experimetally supports external resource checks, for example, for a foreignKey
definition like these:
{"package": "../people/datapackage.json", "resource": "people", "fields": "label"}
{"package": "http:/example.com/datapackage.json", "resource": "people", "fields": "label"}
This checks is for pretty common case when a column should have integers that sequentially increment. It accepts following options:
option | type | description |
---|---|---|
column | int/str | Column number or name |
Consider the following CSV file:
id,name
1,one
2,two
3,three
5,five
Let's check if the id
column contains sequential integers:
from goodtables import validate
report = validate('data.csv', checks=[
{'sequential-value': {'column': 'id'}},
])
# error on row 5 with code "sequential-value"
Some database or spreadsheet software (like MySQL or Excel) could cutoff values on saving. There are some well-known heuristics to find this bad values. See https://github.com/propublica/guides/blob/master/data-bulletproofing.md for more detailed information.
Consider the following CSV file:
id,amount,comment
1,14000000,good
2,2147483647,bad
3,32767,bad
4,234234234,bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbad
To detect all probably truncated values we could use truncated-value
check:
report = validate('data.csv', checks=[
'truncated-value',
])
# error on row 3 with code "truncated-value"
# error on row 4 with code "truncated-value"
# error on row 5 with code "truncated-value"
With Table Schema we could create constraints for an individual field but sometimes it's not enough. With a custom constraint check every row could be checked against given limited python expression in which variable names resolve to column values. See list of available operators. It accepts following options:
- constraint (str)
- Constraint definition (e.g.
col1 + col2 == col3
)
Consider csv file like this:
id,name,salary,bonus
1,Alex,1000,200
2,Sam,2500,500
3,Ray,1350,500
4,John,5000,1000
Let's say our business rule is to be shy on bonuses:
report = validate('data.csv', checks=[
{'custom-constraint': {'constraint': 'salary > bonus * 4'}},
])
# error on row 4 with code "custom-constraint"
To create a custom check user could use a check
decorator. This way the builtin check could be overridden (use the spec error code like duplicate-row
) or could be added a check for a custom error (use type
, context
and position
arguments):
from goodtables import validate, check, Error
@check('custom-check', type='custom', context='body')
def custom_check(cells):
errors = []
for cell in cells:
message = 'Custom error on column {column_number} and row {row_number}'
error = Error(
'custom-error',
cell,
message
)
errors.append(error)
return errors
report = validate('data.csv', checks=['custom-check'])
Recommended steps:
- let's discuss in comment proposed checks first
- select name for a new check like
possible-noise-text
- copy https://github.com/frictionlessdata/goodtables-py/blob/master/goodtables/contrib/checks/blacklisted_value.py to new check module
- add new check module to configuration - https://github.com/frictionlessdata/goodtables-py/blob/master/goodtables/config.py
- write actual code for the new check
- write tests and readme for the new check
To create a custom preset user could use a preset
decorator. This way the builtin preset could be overridden or could be added a custom preset.
from tabulator import Stream
from tableschema import Schema
from goodtables import validate
@preset('custom-preset')
def custom_preset(source, **options):
warnings = []
tables = []
for table in source:
try:
tables.append({
'source': str(source),
'stream': Stream(...),
'schema': Schema(...),
'extra': {...},
})
except Exception:
warnings.append('Warning message')
return warnings, tables
report = validate(source, preset='custom-preset')
For now this documentation section is incomplete. Please see builtin presets to learn more about the dataset extraction protocol.
cli()
Command-line interface
Usage: cli.py [OPTIONS] COMMAND [ARGS]...
Options:
--version Show the version and exit.
--help Show this message and exit.
Commands:
validate* Validate tabular files (default).
init Init data package from list of files.
validate(source, **options)
Validates a source file and returns a report.
Arguments
-
source (Union[str, Dict, List[Dict], IO]): The source to be validated. It can be a local file path, URL, dict, list of dicts, or a file-like object. If it's a list of dicts and the
preset
is "nested", each of the dict key's will be used as if it was passed as a keyword argument to this method.The file can be a CSV, XLS, JSON, and any other format supported by `tabulator`_.
-
checks (List[str]): List of checks names to be enabled. They can be individual check names (e.g.
blank-headers
), or check types (e.g.structure
). -
skip_checks (List[str]): List of checks names to be skipped. They can be individual check names (e.g.
blank-headers
), or check types (e.g.structure
). -
infer_schema (bool): Infer schema if one wasn't passed as an argument.
-
infer_fields (bool): Infer schema for columns not present in the received schema.
-
order_fields (bool): Order source columns based on schema fields order. This is useful when you don't want to validate that the data columns' order is the same as the schema's.
-
error_limit (int): Stop validation if the number of errors per table exceeds this value.
-
table_limit (int): Maximum number of tables to validate.
-
row_limit (int): Maximum number of rows to validate.
-
preset (str): Dataset type could be
table
(default),datapackage
,nested
or custom. Usually, the preset can be inferred from the source, so you don't need to define it. -
Any (Any): Any additional arguments not defined here will be passed on, depending on the chosen
preset
. If thepreset
istable
, the extra arguments will be passed on totabulator
, if it isdatapackage
, they will be passed on to thedatapackage
constructor.
Raises
GoodtablesException
: Raised on any non-tabular error.
Returns
dict
: The validation report.
preset(name)
Register a custom preset (decorator)
Example
@preset('custom-preset')
def custom_preset(source, **options):
# ...
Arguments
- name (str): preset name
check(name, type=None, context=None, position=None)
Register a custom check (decorator)
Example
@check('custom-check', type='custom', context='body')
def custom_check(cells):
# ...
Arguments
- name (str): preset name
- type (str): has to be
custom
- context (str): has to be
head
orbody
- position (str): has to be
before:<check-name>
orafter:<check-name>
Error(self, code, cell=None, row_number=None, message=None, message_substitutions=None)
Describes a validation check error
Arguments
- code (str): The error code. Must be one in the spec.
- cell (dict, optional): The cell where the error occurred.
- row_number (int, optional): The row number where the error occurs.
- message (str, optional): The error message. Defaults to the message from the Data Quality Spec.
- message_substitutions (dict, optional): Dictionary with substitutions to be used when generating the error message and description.
Raises
KeyError
: Raised if the error code isn't known.
dict() -> new empty dictionary dict(mapping) -> new dictionary initialized from a mapping object's (key, value) pairs dict(iterable) -> new dictionary initialized as if via: d = {} for k, v in iterable: d[k] = v dict(**kwargs) -> new dictionary initialized with the name=value pairs in the keyword argument list. For example: dict(one=1, two=2)
GoodtablesException(self, /, *args, **kwargs)
Base goodtables exception
The project follows the Open Knowledge International coding standards.
Recommended way to get started is to create and activate a project virtual environment. To install package and development dependencies into active environment:
$ make install
To run tests with linting and coverage:
$ make test
Here described only breaking and the most important changes. The full changelog and documentation for all released versions could be found in nicely formatted commit history.
- Added
check.check_headers_hook
to support headers check for body-contexted checks (see https://github.com/frictionlessdata/goodtables-py/tree/v3 for native support)
- Added integrity checks for data packages. If
resource.bytes
orresource.hash
(sha256) is provided it will be verified against actual values
- Added a foreign keys check
- Improved missing/non-matching-headers detection (#298)
- A new key added to the
error.to_dict
return:message-data
Breaking changes:
- Checks method signature now only receives the current row's
cells
list - Checks raise errors by returning an array of
Error
objects - Cells have the row number in the
row-number
key - Files with ZIP extension are presumed to be datapackages, so
goodtables mydatapackage.zip
works - Improvements to goodtables CLI (#233)
- New
goodtables init <data paths>
command to create a newdatapackage.json
with the files passed as parameters and their inferred schemas.
Bug fixes:
- Fix bug with
truncated-values
check on date fields (#250)
New API added:
- Validation
source
now could be apathlib.Path
Improved behaviour:
- rebased on Data Quality Spec v1
- rebased on Data Package Spec v1
- rebased on Table Schema Spec v1
- treat primary key as required/unique field
New advanced checks added:
blacklisted-value
custom-constraint
deviated-value
sequential-value
truncated-value
New API added:
report.preset
report.tables[].schema
New API added:
report.tables[].scheme
report.tables[].format
report.tables[].encoding
This version includes various big changes. A migration guide is under development and will be published here.
First version of goodtables
.