Skip to main content

This module provides an easy way to manipulate and inspect lists of data. It was designed to handle plain data types, especially the output from parsing JSON. It allows simple operations to be done on the items in parallel in a concise fashion. Many features will also work on other data types.

Project description

Version 1.0.2

“When all you have is a hammer, everything looks like a nail.” - Anonymous

Note that although the version is barely past 1.0, the code is production-ready. It has 100% unit test coverage, is extensively document and has considerable examples.

TL;DR: Too anxious to see just what DataHammer can do? Check out the DataHammer Examples.


This module provides an easy way to filter, inspect, analyze and manipulate many similar data items. It was designed to handle plain data types, especially the output from parsing JSON. It is designed to allow operations to be done a concise fashion, and on all items in a simple parallel manner.

It works on many data types, but excels at handling of list, tuple and dict values - and was designed primarily for handling similar result objects from parsing JSON or XML.

In the design, concise usages was favored over speed of performance - although for data sets of less than a million records it is still quite performant. It was inspired by a need for a concise data manipulation syntax and by the projects jQuery and jq, and as such it also has no dependencies beyond a standar Python instance.


  • Most operations on a DataHammer instance return a value or a new instance, they do not mutate the contained data, although a returned ITEM could be mutated by the calling code.

  • The contained data can be retrieved with the invert operator (~). It will be a list unless constructed with a single ITEM, in which case that ITEM will be returned.

  • In order to allow accessing arbitrary ITEM attributes uses the dot notation, public functions start with a single underscore, in contrast to typical Python conventions. See Functions.

  • It uses a list as its top-level container, and will convert a tuple and some generators into a list.

  • When constructed with a single ITEM, that item will be wrapped in a list and most operations will be identitical to having been constructed with a list with that single ITEM.

  • It uses ‘.’ to access dict members or object attributes, using None for items where there is no key or attribute with the specified name, thus no KeyError or AttributeError will be raised.

  • Almost all operations will silently ignore items that do not have a member with the “intended” key, attribute or index.

  • There is a Mutator class returned by the _mutator() function that is designed to allow modifying the data in-place for some of the Augmented Assignment statements.

Build Status

Current status:

Build Status

Known Issues

  • Using “ITEM in OBJ” works as you probably expect, but avoid using “OBJ in OTHER” for iterable containers. [6]

  • By design and intent, the bitwise operators (&, |, ^) actually create a new instance by applying the and, or and xor operators, respectively. This is because those keyword operators cannot be overridden to return an object as we wish.

  • There are missing operators that could be added. Among these are del (attribute or key), and the bitwise math operators.

  • The OBJ._long() method was removed in version 0.9.6, since it was complicating the code and testing, and is identical to using: OBJ._apply(long) Apologies for the breakage, if you hit that.

  • The OBJ._toCSV() method changed in version 0.9.7, since it wasn’t using the builtin csv package, and was using a backslash to escape double-quotes within the quoted text, rather than a pair of consecutive double-quotes. For example, the previous output "Derrial \"Preacher\" Book" is now output as "Derrial ""Preacher"" Book".


Creating a DataHammer can take several sources for its input. It is designed for use on a list of items with the same schema.




This must be one of:

  • A list of ITEMS.

  • A single, non-list ITEM.

If the json value is true, then data can be either of:

  • A file object, from which all data is read, and the results are treated as TEXT, or…

  • TEXT to be parsed as JSON.


If given and true, then a deepcopy will be made of data.


If provided, it should either be True or a dict of arguments to be passed to JSON.loads() for when data is of either the file or TEXT forms.


This is a list of supported operations, including applying builtin Python functions. [1]




Returns the contained data.


Creates a list by applying the index (an int for list items, a key for dict items, or the name of an attribute or property), returning a DataHammer instance created using that list. [2]

op can be: < <= == != >= >

Return a DataHammer instance with a bool result from the comparison of each ITEM with OTHER. [3]

To test equality of contents, use: ~OBJ == OTHER

bitop can be: & ^ |

Return a DataHammer instance with the results of applying and, or and a “bool-xor” to each ITEM and OTHER, or (OTHER and ITEM). These are needed since those keywords cannot be overridden in the desired fashion. [4]

OBJ mathop OTHER
mathop can be: + - * / // ** %

Return a DataHammer instance with the results of applying a math operators in: ITEM mathop OTHER. [3]

OTHER mathop OBJ
mathop can be: + - * / // ** %

Return a DataHammer instance with the results of applying a math operators in: OTHER mathop ITEM. [3]


Depending on the argument, returns a DataHammer instance, a single contained ITEM, or a list of ITEMs. [4] See Indexing, for more information.


Return a DataHammer instance with the results of applying the builtin type (of the same name w/o the underscore) to each item in the list.


Return a DataHammer instance with the contained data in reversed order.


Return an int for the number of contained data ITEMs.


Return an int that is the hash of the tuple of the hash of every ITEM. This will raise an exception if any ITEM cannot be hashed.


Return a bool, which is True if any ITEM == OBJ. With regard to limiting the items tested. [3]


This is almost never what you want! Return a single bool, ignoring of contents of ARG or OBJ. The result is True if neither ARG nor OBJ are empty, and False if they both are.

-OBJ (unary minus)

Return a DataHammer instance with the results of applying not ITEM on each item.

NOTE: The OBJ._long() method was removed in version 0.9.6, since it was complicating the code and testing, and is identical to using: OBJ._apply(long)


This is a list of supported functions. [1]




Attribute, index or dict key dereference. [2]


Returns a JSON dump of the contained data.


Return a DataHammer instance with the results of applying ARG in ITEM for each item.


Return a DataHammer instance with the results of applying ITEM in ARG for each item.

OBJ._apply(FUNC, ARG, *ARGS, **KWDS)

Return a DataHammer instance with the results of applying FUNC(ITEM, ARG, *ARGS, **KWDS) to each item. [3]


Return a DataHammer instance with only the desired items. Based on the type of ARG given, the new instance has only the items for which the result is true of: 1. If ARG is not given: bool(ITEM) 2. If ARG is a callable: ARG(ITEM) 3. If ARG is a list, tuple or set: (ITEM in ARG) 4. Otherwise: ITEM == ARG

OBJ._insert(INDEX, ITEM)

Return a DataHammer instance with ITEM inserted at INDEX.


Return a DataHammer instance with ITEMS added at the end.


Return a DataHammer instance with DELNUM items deleted at INDEX, and with ITEM(s) inserted there. [5]

OBJ._slice(START [, END [, STEP ] ])

Return a DataHammer instance with the list sliced according to the given indices (like list slicing works).


Return a DataHammer instance with contained items that are the result of flattening this instance’s contained items by one level. Sub-items are added in iteration-order for items that are a set, list or tuple and for values from a dict.

Other types are not flattened, and are added as-is.


Return a tuple of results for each contained item, the result will be a tuple of values from the items, dereferenced by the SELECTOR parameters, in the same order. See [8]

Only named SELECTOR parameters are allowed.


Return a tuple of str like a Comma Separated Values file, the first str represents the headers for each column, and each subsequent contains a CSV-style representation of the requested values from each item (which must be serializable). See [8]

Both positional and named SELECTOR parameters are allowed.


Return a DataHammer instance of dict items made from one or more sub-items specified by the SELECTOR, as either positional or named parameters. Parameters dictate the keys in the resulting items. See [8]

Both positional and named SELECTOR parameters are allowed.

OBJ._groupby(GRP, VALS [, POST])

Return a DataHammer instance of dict items made by taking all sub-items specified by VALS and combine them with other items with the same GRP values. It is similar to the SQL GROUP BY clause. See [8] and Grouping.

Both positional and named SELECTOR parameters are allowed.


Return a DataHammer instance of dict items from merging items from this instance and OBJ, joining on the values corresponding to the KEYS. The FLAGS parameter controls specifics. Somewhat similar to the SQL JOIN operations. See Joining and the Deeper Examples.

OBJ._unique(KEYS [,UNIQUE])

Return a DataHammer instance with items from this instance, based on the uniqueness of the values for KEYS. The UNIQUE parameter sets handling for items with duplicate key values.

See Unique and the Deeper Examples.


Returns a DataHammer.Mutator instance to be used for making modifications to the contained data. See Mutators.


Indexing a DataHammer instance with [] allows simple access to items from the contained data, but there are various types of parameters types allowed. [4]

  1. Indexing with an int or an implicit or explicit slice object works like indexing list; the result is identical to (~OBJ)[…].

    • A single item is returned with an int argument, and can raise an IndexError.

    • A (possibly empty) list of items is returned with either:

      • An explicit slice argument, eg: OBJ[slice(1, None, 5)]

      • An implicit slice argument, eg: OBJ[1::5]

  2. Indexing with a list, tuple or a DataHammer instance, will return another DataHammer instance. [3] The parameter must either be all bool or all int, and they dictate which items are used to construct the new instance:

    • For bool indexes, each bool in the argument indicates if the corresponding item in the DataHammer is included in the new instance.

    • For int indexes, each int is used to index into the contained data, and which item is include in the new instance. This allows both filtering and reordering of data.

Indexing Examples:

>>> OBJ = DataHammer(list(range(10, 15)))

# Note that the following dereference the instance with "~" to show the contents:

>>> ~OBJ
[10, 11, 12, 13, 14]
>>> ~OBJ[(True, False, True, True, False, True)]
[10, 12, 13]      # The last/6th `True` is ignored since len(OBJ)==5
>>> ~OBJ[(4, 2, 1, 40, -1, 3, 1)]
[14, 12, 11, 14, 13, 11]    # 40 is ignored.

# Note these DO NOT dereference the result, they are not a DataHammer instance.

>>> type(OBJ[1])
<type 'int'>
>>> type(OBJ[:5])
<type 'list'>
>>> type(OBJ[slice(3)])
<type 'list'>
>>> OBJ[::3]
[10, 13]


The _groupby(GROUP, VALUES [, POSTPROC]) method creates a new DataHammer instance, grouping values from multiple source items. It functions somewhat like the GROUP BY feature of SQL, however rather than necessarily combining column values, a the list of values is created.

The GROUP and VALUES parameters should be either a list/tuple or a dict.

  • Strings in the list/tuple are treated like named SELECTOR parameters

  • Items in a dict are treated like named SELECTOR parameters.

For each unique sets of values for the GROUP keys, one item will exist in the resulting instance. Each of the new items will contain the grouping values and a value per VALUES key. The GROUP and VALUES parameters may be either a list/tuple or a dict of SELECTOR parameters (see above).

For every key in the VALUES parameter, a list is built with the corresponding values, one list for each set of GROUP values.

The POSTPROC parameter parameter, is optional and unless provided: each resulting item will contain the corresponding list for each key in VALUES. If FUNC is provided, it will be called once per resulting item. The lists are passed parameters in the same order as the keys in VALUES.

Note that the order of the resulting items will be the same as the order of the first occurence of that set of GROUP keys in the source items. And the order of the list of values for each VALUES key is the same as the order that those occurred in the source items.


There is a method for joining two DataHammer instances, combining items for which the specified key values match, this is partly inspired by the JOIN feature of SQL (JOIN_PRODUCT), and partly inspired by a use case where one-to-one matches were needed.

With the two (2) flags [JOIN_PRODUCT or JOIN_ORDERED] for handling duplicates and the four (4) flags [JOIN_KEEP_NEITHER, JOIN_KEEP_LEFT, JOIN_KEEP_RIGHT or JOIN_KEEP_BOTH] for handling unmatched items, there are eight (8) different flags combinations.


Here, “duplicate” key-values means that a set of key-values occurs more than once in the same instance.

“Mode” Flag Name



Results are somewhat similar to SQL joins. The name comes from the “Cartesian Product” since the output contains an item produced from the each matching item the left input and the right input.


This pairs matching items from the left and the right, one-by-one. Pairing is in the same order as they were found in the input instances, and matching stops after exhausting the matching items in either the right input or left input.

If there are no duplicate entries in either input, then these modes function identically.


An “unmatched” item is one whose key-values never occur in the items from the other instance.

Here, the INNER and OUTER join terminology is a remnant from SQL, the “KEEP” flags are equivalent and provided since they describe the intended action. These can be summarized thus:

“Keep” Flag Name

Inner/Outer Name




Discard unmatched items from left and from the right.



Discard unmatched items from the right.



Discard unmatched items from the left.



Keep unmatched items from the left and from the right.


The order of items in the inputs dictates the order in the output. The algorithm simply iterates over the left input, producing zero or more outputs depending on the flags and presence of any matching items in the right input. It then appends unmatched items from the right, if desired.

See the examples, or use it for yourself, if this is not sufficiently clear.


  • With JOIN_PRODUCT, each matched item from the left will be paired with every matching item from the right, in the order that the right items occurred.

  • With JOIN_ORDERED, each item in the left will be paired with the corresponding order of the matching items in the right input. After the items from the right are exhausted, the remaining items from the left input with that set of key-values are considered unmatched. In addition, any items from the right input that are not consumed in this way are also considered unmatched.

  • With JOIN_KEEP_LEFT or JOIN_KEEP_BOTH, unmatched items from the left input will appear in the same order as they are found in the left input.

  • With JOIN_KEEP_RIGHT or JOIN_KEEP_BOTH, unmatched items from the right input will appear after all items produced from items in the left input. They will be in the same order as they occurred the right input.


The _unique() method allows eliminating items based on the uniqueness / duplication of key values.

Parameters are:

  • KEYS should be a list/tuple of strings which are used as a SELECTOR into each item. The associated values are used for the uniqueness test. (If KEYS is a single string, it is handled as expected.)

    There is a special case when KEYS is None: if so, the hash of the item is used in lieu of key values. Obviously, all items must be hashable.

  • UNIQUE determines which items to keep, based on key values. UNIQUE may be:

    • 0 = Keep only those items that are unique, with no duplicates.

    • 1 = Keep the first item with key values, ignore subsequent duplicates.

    • 2 = Keep all instances of items that have duplicate key values.

In each case, the order of the items is preserved from the original data.

Note that with unique = 2, there will be multiple items that have the same key values; to remove those you filter them a second time with the same KEYS:

OBJ._unique(KEYS, 2)._unique(KEYS)


There is some support for making modifications to the data contained within a DataHammer, beyond direct access. This is done with the DataHammer._mutator method on the instance.

Here MUT is used as a shorthand for OBJ._mutator() - which returns a DataHammer.Mutator instance, and the name Mutator is also used for DataHammer.Mutator.

Functions and Operation


MUT = OBJ._mutator()

Returns a new Mutator for the given DataHammer instance.


Returns the DataHammer instance for this Mutator.


Returns a new Mutator instance useful for modifying the key, attribute or list item at index. [7]

Note that all of these forms work identically, though the first form can only be used with valid identifier names. This is in contrast with [] on a DataHammer instance where it returns an item from the contained data.

op can be: += -= *= /= **= //=

Update the item member for the given Mutator instance, with the given operation, which should be number (or object that supports that operation).


Update the value designated by the given Mutator instance, overwriting with the given value(s). If OTHER is a list, tuple or DataHammer instance, then an interator is used, and application stops when the end is reached. [3]


Like MUT._set(OTHER) but regardless of the type, OTHER is used without iterating. Used to set all rows to the same list or tuple value, but can be used with any value/type.

MUT._apply(FUNC, *ARGS, **KWDS)

Update the value designated by the given Mutator instance, overwriting with the the return value from calling: ``FUNC(VALUE, *ARGS, **KWDS)``.


Install the package using pip, eg:

pip install –user datahammer

Or for a specific version of Python:

python3 -m pip –user install datahammer

To the source git repository, use:

git clone





Initial release, documentation prototyping.


Addition of “_pick” method.


Addition of “_flatten” and “_toCSV” methods.


Addition of “_groupby” and “_tuples” methods.


Moved EXAMPLES into (and reorganized) the README file. Configured for tests, coverage and style on Travis CI.


Removed ‘OBJ._long()’ method, as it was Python2-only.


Added the ‘OBJ._join()’ and ‘OBJ._fromCSV()’ methods.


Added the ‘OBJ._unique()’ and ‘OBJ._in()’ methods.


Moved to, including GitLab-CI.


Minor changes post move to GitLab.


Final changes before pushing to PyPi.

Reporting Issues, Contributing

As an open source project, DataHammer welcomes contributions and feedback.

  1. Report any issues, including with the functionality or with the documentation via the GitLab project:

  2. To contribute to the source code, please use a GitLab pull request for the project, making sure to include full/extensive unit tests for any changes. Note that if you cannot create a PR, then open an issue and attach a diff output there.

  3. To translate the documentation, please follow the same process as for source code contributions.

DataHammer Examples

It is probably easier to show the utility of DataHammer with some examples.

Simple Examples

  1. To construct a DataHammer instance you generally a list/tuple/iterable of items. Many builtin functions operate on the DataHammer instance as it would on the list of objects. The original data can be returned using the tilde operator (~).

    See Sample Data for the data used here.

>>> dh = DataHammer(data)
>>> len(dh)
>>> dh
<datahammer.DataHammer object at 0x7f258fac34e0>
>>> type(~dh)
<type 'list'>
>>> type(dh[0])
<type 'dict'>
>>> type(dh[:3])
<type 'list'>
>>> ~dh == dh[:]
>>> bool(dh)
  1. Accessing the sub-items uses a simple dot notation. To allow irregular data, a None will represent a member that was not present – no KeyError, AttributeError or IndexError are raised.

>>> ~dh.age
[45, 57, 33, 21, 24, 60, 63, 33]
['Stewart', 'Perry', 'Young', 'Lewis', 'Ward', 'Martinez', 'Evans', 'Moore']
# No KeyError
>>> ~dh.missingMember
[None, None, None, None, None, None, None, None]
  1. Indexing into a list sub-item cannot be done with dot notation or slicing (eg: with []), so the _ind() method is provided for this reason. As for dot notation, if an index is out of range then the value will be None.

# This is not a DataHammer instance, it is just the `rank` member of the fourth item.
>>> dh.ranks[3]
[180, 190, 111]

# This is a DataHammer instance with the fourth item from each `rank` member, or `None`.
>>> ~dh.ranks._ind(3)
[None, 18, 155, None, None, 24, 64, None]
  1. To avoid collisions with item members, the public methods of a DataHammer instance are all prefixed with a single underscore, which may be confusing at first, but this is also done for collections.namedtuple instances. Methods that begin with a double underscore are not public.

# This 'mean' function is defined in the Sample Data section, below.
>>> ~dh.ranks._apply(mean)
[None, 70.33333333333333, 114.875, 160.33333333333334, 139.0, 40.2, 94.83333333333333, 97.0]

>>> ~dh._splice(2, 4).name.first
['Addison', 'Katherine', 'Grace', 'Sophia']

>>> print("\n".join(dh._toCSV(FIRST='name.first', LAST='name.last', AGE='age')))
  1. Many operators are overridden to allow operating on the item with a simple syntax, returning a new DataHammer instance with the results. Most operators work with another DataHammer instance, a list/tuple or scalar values. In the case of a list/tuple, the length of the resulting instance will be the shorter of the two arguments.

>>> ~(dh.gender == 'F')
[True, True, False, True, False, False, True, True]
>>> ~(dh.salary / 1000.0)
[10.0, 18.59, 28.64, 8.0, 8.0, 33.7, 26.22, 14.12]
>>> ~(dh.age > [50, 40, 30])
[False, True, True]
>>> ~(dh.salary * 1.0 / dh.age)   # Avoid integer math.
[222.22222222222223, 326.140350877193, 867.8787878787879, 380.95238095238096,
 333.3333333333333, 561.6666666666666, 416.1904761904762, 427.8787878787879]
  1. Using many builtin operations work as you would expect, as if passing a list/tuple of the item data instead.

>>> min(dh.age), max(dh.age)
(21, 63)
>>> sorted(dh.location.state)
['Maryland', 'Maryland', 'New Jersey', 'Oklahoma', 'Oregon', 'Oregon', 'Texas', 'Texas']
>>> sum(dh.salary)
>>> min(dh.salary), mean(dh.salary), max(dh.salary)
(8000, 18408.75, 33700)

# This gives number of females, by counting occurences of `True`.
>>> sum(dh.gender == 'F')
  1. Indexing with another DataHammer instance is another powerful feature. Also, indexing with integers allows arbitrary keeping a subset of, or reordering of, the items.

>>> len(dh.age < 30), sum(dh.age < 30)
(8, 2)
>>> twenties = (20 <= dh.age < 30)
>>> ~twenties
[False, False, False, True, True, False, False, False]
>>> ~dh[twenties].name
[{'first': 'Brianna', 'last': 'Lewis'}, {'first': 'Logan', 'last': 'Ward'}]
['Stewart', 'Perry', 'Young', 'Lewis', 'Ward', 'Martinez', 'Evans', 'Moore']
>>> ~dh[(0, 5, 3, 4)].name.last
['Stewart', 'Martinez', 'Lewis', 'Ward']

Deeper Examples

These demonstrate the extracting and manipulating power of DataHammer instances. Note that these examples and notes are not trivial, so please read carefully so you can understand the functionality as it is designed.

  1. There are methods for extracting parts of each item, including _pick(), _tuples() and _toCSV(). In addition the _groupby() method allows extracting only certain parts and combining them across the items that share certain values, similar to the GROUP BY syntax in SQL.

    See the main README section for detailed SELECTOR Syntax, but the methods are demonstrated here:

    1. The _tuples(SELECTOR [, SELECTOR …]) method returns a tuple of tuples with extracted values in the same order as the names. Only positional SELECTOR parameters are allowed.

    >>> dh._tuples('', 'name.last', 'age')
    (('Baltimore', 'Stewart', 45),
     ('Baltimore', 'Perry', 57),
     ('Portland', 'Young', 33),
     ('San Antonio', 'Lewis', 21),
     ('Oklahoma ', 'Ward', 24),
     ('Portland', 'Martinez', 60),
     ('Jersey City', 'Evans', 63),
     ('San Antonio', 'Moore', 33))
    1. The _toCSV(SELECTOR [, SELECTOR …]) method returns a tuple of strings in a Comma Separated Values format. The first string is a header of the column names in order. Each subsequent string represents the corresponding item in the data, in order. Both positional and named SELECTOR parameters are allowed.

    >>> dh._toCSV('', lname='name.last', yrs='age')
     '"San Antonio","Lewis",21',
     '"Oklahoma ","Ward",24',
     '"Jersey City","Evans",63',
     '"San Antonio","Moore",33')
    1. The _pick(SELECTOR [, SELECTOR …]) method returns a new DataHammer instance where each item is a dictionary with only the requested members. Positional and named SELECTOR parameters are allowed.

    >>> ~dh._pick('location.state', ln='name.last', fn='name.first', years='age')
    [{'state': 'Maryland', 'ln': 'Stewart', 'fn': 'Addison', 'years': 45},
     {'state': 'Maryland', 'ln': 'Perry', 'fn': 'Katherine', 'years': 57},
     {'state': 'Oregon', 'ln': 'Young', 'fn': 'Jack', 'years': 33},
     {'state': 'Texas', 'ln': 'Lewis', 'fn': 'Brianna', 'years': 21},
     {'state': 'Oklahoma', 'ln': 'Ward', 'fn': 'Logan', 'years': 24},
     {'state': 'Oregon', 'ln': 'Martinez', 'fn': 'Logan', 'years': 60},
     {'state': 'New Jersey', 'ln': 'Evans', 'fn': 'Grace', 'years': 63},
     {'state': 'Texas', 'ln': 'Moore', 'fn': 'Sophia', 'years': 33}]
    1. The _groupby(GROUP, VALUES [, POSTPROC]) method returns a new DataHammer instance, using the first list of keys for grouping by value, and the second list as the values to groupby. Like the GROUP BY functionality in SQL, there will be one item in the resulting instance for each unique set of values of the GROUP keys.

      Remember: even if passing a single key for GROUP or VALUES, it must be in a tuple or list.

       # An empty second parameter is allowed, too, the results is just the unique GROUP keys.
       >>> ~dh._groupby(['gender', 'title'], [])
       [{'gender': 'F', 'title': 'Systems Administrator'},
       {'gender': 'F', 'title': 'Bookkeeper'},
       {'gender': 'M', 'title': 'Controller'},
       {'gender': 'F', 'title': 'UX Designer'},
       {'gender': 'M', 'title': 'Web Developer'},
       {'gender': 'M', 'title': 'Assessor'},
       {'gender': 'F', 'title': 'Mobile Developer'}]
       >>> ~dh._groupby(['gender'], ('age', 'salary'))
       [{'gender': 'F', 'age': [45, 57, 21, 63, 33], 'salary': [10000, 18590, 8000, 26220, 14120]},
        {'gender': 'M', 'age': [33, 24, 60], 'salary': [28640, 8000, 33700]}]
    The third parameter is a callable that takes the constructed lists in `VALUES` key order, and
    returns a tuple with same number of items, in the same order.
    >>> def reductor(ages, salaries):
    ...    return (min(ages), max(ages)), (min(salaries), max(salaries))
    >>> ~dh._groupby(['gender'], ('age', 'salary'), reductor)
    [{'gender': 'F', 'age': (21, 63), 'salary': (8000, 26220)},
     {'gender': 'M', 'age': (24, 60), 'salary': (8000, 33700)}]
  2. There is a method for joining two DataHammer instances, combining items for which the specified key values match. The JOIN_PRODUCT mode is inspired by the JOIN feature of SQL, whiel JOIN_ORDERED was inspired by a use case where one-to-one matches were needed.

    >>> left = DataHammer([{"k": "A", "x": 1}, {"k": "B", "x": 2}, {"k": "C", "x": 3},
    ...     {"k": "C", "x": 4}, {"k": "D", "x": 5}])
    >>> right = DataHammer([{"k": "A", "y": 1}, {"k": "A", "y": 2}, {"k": "C", "y": 3},
    ...     {"k": "C", "y": 4}, {"k": "E", "y": 5}])
    # For JOIN_PRODUCT, each matched item from the left is paired with each the corresponding item
    # from the right.  Then the JOIN_KEEP_{...} flag determines unmatched item retention.
    # Default is ORDERED + NEITHER
    >>> ~left._join("k", right)
    [{'k': 'A', 'x': 1, 'y': 1},
     {'k': 'A', 'x': 1, 'y': 2},
     {'k': 'C', 'x': 3, 'y': 3},
     {'k': 'C', 'x': 3, 'y': 4},
     {'k': 'C', 'x': 4, 'y': 3},
     {'k': 'C', 'x': 4, 'y': 4}]
    >>> ~left._join("k", right, flags=DataHammer.JOIN_PRODUCT + DataHammer.JOIN_KEEP_NEITHER)
    [{'k': 'A', 'x': 1, 'y': 1},
     {'k': 'A', 'x': 1, 'y': 2},
     {'k': 'C', 'x': 3, 'y': 3},
     {'k': 'C', 'x': 3, 'y': 4},
     {'k': 'C', 'x': 4, 'y': 3},
     {'k': 'C', 'x': 4, 'y': 4}]
    >>> ~left._join("k", right, flags=DataHammer.JOIN_PRODUCT + DataHammer.JOIN_KEEP_RIGHT)
    [{'k': 'A', 'x': 1, 'y': 1},
     {'k': 'A', 'x': 1, 'y': 2},
     {'k': 'C', 'x': 3, 'y': 3},
     {'k': 'C', 'x': 3, 'y': 4},
     {'k': 'C', 'x': 4, 'y': 3},
     {'k': 'C', 'x': 4, 'y': 4},
     {'k': 'E', 'y': 5}]
    >>> ~left._join("k", right, flags=DataHammer.JOIN_PRODUCT + DataHammer.JOIN_KEEP_LEFT)
    [{'k': 'A', 'x': 1, 'y': 1},
     {'k': 'A', 'x': 1, 'y': 2},
     {'k': 'B', 'x': 2},
     {'k': 'C', 'x': 3, 'y': 3},
     {'k': 'C', 'x': 3, 'y': 4},
     {'k': 'C', 'x': 4, 'y': 3},
     {'k': 'C', 'x': 4, 'y': 4},
     {'k': 'D', 'x': 5}]
    >>> ~left._join("k", right, flags=DataHammer.JOIN_PRODUCT + DataHammer.JOIN_KEEP_BOTH)
    [{'k': 'A', 'x': 1, 'y': 1},
     {'k': 'A', 'x': 1, 'y': 2},
     {'k': 'B', 'x': 2},
     {'k': 'C', 'x': 3, 'y': 3},
     {'k': 'C', 'x': 3, 'y': 4},
     {'k': 'C', 'x': 4, 'y': 3},
     {'k': 'C', 'x': 4, 'y': 4},
     {'k': 'D', 'x': 5},
     {'k': 'E', 'y': 5}]
    # For JOIN_ORDERED, matched items from the left and right are paired, one-by-one, but only as
    # until either side is exhausted, the remaining items are 'unmatched' and the JOIN_KEEP_{...}
    # flag determines unmatched item retention.
    >>> ~left._join("k", right, flags=DataHammer.JOIN_ORDERED + DataHammer.JOIN_KEEP_NEITHER)
    [{'k': 'A', 'x': 1, 'y': 1},
     {'k': 'C', 'x': 3, 'y': 3},
     {'k': 'C', 'x': 4, 'y': 4}]
    >>> ~left._join("k", right, flags=DataHammer.JOIN_ORDERED + DataHammer.JOIN_KEEP_RIGHT)
    [{'k': 'A', 'x': 1, 'y': 1},
     {'k': 'C', 'x': 3, 'y': 3},
     {'k': 'C', 'x': 4, 'y': 4},
     {'k': 'A', 'y': 2},
     {'k': 'E', 'y': 5}]
    >>> ~left._join("k", right, flags=DataHammer.JOIN_ORDERED + DataHammer.JOIN_KEEP_LEFT)
    [{'k': 'A', 'x': 1, 'y': 1},
     {'k': 'B', 'x': 2},
     {'k': 'C', 'x': 3, 'y': 3},
     {'k': 'C', 'x': 4, 'y': 4},
     {'k': 'D', 'x': 5}]
    >>> ~left._join("k", right, flags=DataHammer.JOIN_ORDERED + DataHammer.JOIN_KEEP_BOTH)
    [{'k': 'A', 'x': 1, 'y': 1},
     {'k': 'B', 'x': 2},
     {'k': 'C', 'x': 3, 'y': 3},
     {'k': 'C', 'x': 4, 'y': 4},
     {'k': 'D', 'x': 5},
     {'k': 'A', 'y': 2},
     {'k': 'E', 'y': 5}]

(Obviously, the outputs above were reformmated for clarity.)

  1. There is a method for easily dealing with duplicate values on particular keys. Once again the idea of key values is used to determine what is considered.

    Note that if KEYS is a string, it is handled correctly.

    >>> dh = DataHammer(data)
    >>> keys = ''
    >>> Counter(
    Counter({'Baltimore': 2, 'Portland': 2, 'San Antonio': 2, 'Oklahoma ': 1, 'Jersey City': 1})
    # Zero (0) gives only the unique items, where count was 1 -- all but "Mobile Developer".
    >>> ~dh._unique(keys, 0)
    ['Oklahoma City', 'Jersey City']
    # The default (one, 1) gives all key values, but only the first item with the value(s)
    >>> ~dh._unique(keys)
    ['Baltimore', 'Portland', 'San Antonio', 'Oklahoma City', 'Jersey City']
    >>> ~dh._unique(keys, 1).name
    [{'first': 'Addison', 'last': 'Stewart'}, {'first': 'Jack', 'last': 'Young'},
     {'first': 'Brianna', 'last': 'Lewis'}, {'first': 'Logan', 'last': 'Ward'},
     {'first': 'Grace', 'last': 'Evans'}]
    # Two (2) gives only those items that have a duplicate (of '').
    >>> ~dh._unique(keys, 2)._pick('name', keys)
    [{'name': {'first': 'Addison', 'last': 'Stewart'}, 'city': 'Baltimore'},
     {'name': {'first': 'Katherine', 'last': 'Perry'}, 'city': 'Baltimore'},
     {'name': {'first': 'Jack', 'last': 'Young'}, 'city': 'Portland'},
     {'name': {'first': 'Brianna', 'last': 'Lewis'}, 'city': 'San Antonio'},
     {'name': {'first': 'Logan', 'last': 'Martinez'}, 'city': 'Portland'},
     {'name': {'first': 'Sophia', 'last': 'Moore'}, 'city': 'San Antonio'}]
    # To get the unique set of duplicated values for a given set of keys, you can use a second
    # pass, using the default (1) for `unique`.
    >>> ~dh._unique(keys, 2)._unique(keys).location
    [{'city': 'Baltimore', 'state': 'Maryland'},
     {'city': 'Portland', 'state': 'Oregon'},
     {'city': 'San Antonio', 'state': 'Texas'}]

Formatting Specification

  1. An extension is provided for formatting, using the j type. Each item will be printed as JSON using json.dumps(). In particular, the only allowed parts to the format_spec are:

  1. A negative sign will cause a newline to be inserted between the item outputs.

  2. A non-zero width causes the item JSON is used as the indent within the item output

  3. The only type supported is “j”.

>>> dh.location[0:2]
[{'city': 'Baltimore', 'state': 'Maryland'}, {'city': 'Madison', 'state': 'Wisconsin'}]
>>> print("{:-j}".format(dh.location._slice(0,2)))
>>> print("{:-3j}".format(dh.location._slice(0,2)))

Warnings and Caveats

  1. Warning: To combine multiple instances with bool values you must use the & and |, and not use and and or as you would with Python bool values.

>>> dh1 = DataHammer([False, False, True, True])
>>> dh2 = DataHammer([False, True, False, True])

# These are item-wise correct results
>>> ~(dh1 & dh2)
[False, False, False, True]
>>> ~(dh1 | dh2)
[False, True, True, True]

# Since the objects are not empty, 'or' returns the first, 'and' returns the second:
>>> (dh1 or dh2) == dh1
>>> (dh1 and dh2) == dh2

Other Examples

  1. Given a JSON file that has metadata separated from the data values, we can easily combine these, and find the ones which match criteria we want.

>>> from datahammer import DataHammer
>>> from six.moves.urllib import request
>>> from collections import Counter

>>> URL = ''
>>> req = request.urlopen(URL)
>>> jobs = DataHammer(req, json=dict(encoding='utf-8'))

# Grab the contained data in order to find its keys.
>>> (~jobs).keys()
dict_keys(['meta', 'data'])
>>> names =
>>> norm = DataHammer(dict(zip(names, row)) for row in

# Here 'norm' contains 1260 items, each a dict with the same schema.
>>> len(norm)
>>> print(norm[0])
{'sid': 1, 'id': 'A0447302-02D8-4EFD-AB68-777680645F02', 'position': 1,
 'created_at': 1437380960, 'created_meta': '707861', 'updated_at': 1437380960,
 'updated_meta': '707861', 'meta': None, 'Year': '2012', 'Region': 'Capital Region',
 'NAICS Code': '11', 'Industry': 'Agriculture, Forestry, Fishing and Hunting',
 'Jobs': '2183'}

# Use collections.Counter to count the number of instances of values:
>>> Counter(norm.Year)
Counter({'2012': 210, '2013': 210, '2014': 210, '2015': 210, '2017': 210, '2016': 210})
>>> Counter(norm._get('NAICS Code'))
Counter({'11': 60, '21': 60, '22': 60, '23': 60, '42': 60, '51': 60, '52': 60,
         '53': 60, '54': 60, '55': 60, '56': 60, '61': 60, '62': 60, '71': 60,
         '72': 60, '81': 60, '90': 60, '99': 60, '31-33': 30, '44-45': 30,
         '48-49': 30, '31': 30, '44': 30, '48': 30})

# Use '&' to require both conditions, it is a row-wise `and` of the separate tests.
>>> cap2013 = norm[(norm.Year == '2013') & norm.Region._contains('Capital Region')]
>>> len(cap2013)
>>> keepers = norm.Jobs._int() > 500000
>>> sum(keepers)
>>> large = norm[keepers]
>>> len(large)
>>> large[0]
{'sid': 121, 'id': '98A53A4E-712C-47A9-9106-C9062DB8CBBD', 'position': 121,
 'created_at': 1437380961, 'created_meta': '707861', 'updated_at': 1437380961,
 'updated_meta': '707861', 'meta': None, 'Year': '2012', 'Region': 'New York City',
 'NAICS Code': '62', 'Industry': 'Health Care and Social Assistance', 'Jobs': '591686'}
>>> ~norm.Region._unique(None)
['Capital Region', 'Central New York', 'Finger Lakes', 'Long Island', 'Mid-Hudson',
 'New York City', 'North Country', 'Southern Tier ', 'Western New York ', 'Mohawk Valley',
 'Southern Tier', 'Western New York']
>>> Counter(norm.Region)
Counter({'Capital Region': 126, 'Central New York': 126, 'Finger Lakes': 126,
         'Long Island': 126, 'Mid-Hudson': 126, 'New York City': 126, 'North Country': 126,
         'Mohawk Valley': 126, 'Southern Tier ': 63, 'Western New York ': 63,
         'Southern Tier': 63, 'Western New York': 63})
>>> sum(norm.Region._in(['Mohawk Valley', 'Southern Tier']))


  • The positional parameter “b.b1” would dererence a value like OBJ.b.b1, and the resulting key would be the part after the last dot: “b1”.

  • The named parameter animal=”b.b2” would dererence like OBJ.b.b2, and the resulting key would be “animal”.

>>> dh = DataHammer([
      {"a": 100, "b": {"b1": [101, 102, 103], "b2": "ape"}, "c": ["Apple", "Anise"]},
      {"a": 200, "b": {"b1": [201, 202, 203], "b2": "bat"}, "c": ["Banana", "Basil"]},
      {"a": 300, "b": {"b1": [301, 302, 303], "b2": "cat"}, "c": ["Cherry", "Cayenne"]}

>>> ~dh._pick('a', 'b.b1', animal='b.b2', food='c', nil='')
[{'a': 100, 'b1': [101, 102, 103], 'animal': 'ape', 'food': ['Apple', 'Anise'], 'nil': None},
 {'a': 200, 'b1': [201, 202, 203], 'animal': 'bat', 'food': ['Banana', 'Basil'], 'nil': None},
 {'a': 300, 'b1': [301, 302, 303], 'animal': 'cat', 'food': ['Cherry', 'Cayenne'], 'nil': None}]

#### Result is undefined due to the key collision.
>>> ~dh._pick('b.b1', b1='c')

## This '.0' syntax *might* change in future releases.
>>> ~dh._pick(animal='b.b2', fruit='c.0')
[{'animal': 'ape', 'fruit': 'Apple'},
 {'animal': 'bat', 'fruit': 'Banana'},
 {'animal': 'cat', 'fruit': 'Carmel'}]

Sample Data

Note that the data used here is randomly generated, no relationship to anyone living, dead or undead is intended.

>>> from datahammer import DataHammer
>>> mean = lambda nums: (sum(nums) * 1.0 / len(nums)) if nums else None
>>> data = [
        "ranks":[],"salary":10000,"title":"Systems Administrator"
        "age":21,"gender":"F","location":{"city":"San Antonio","state":"Texas"},
        "ranks":[180,190,111],"salary":8000,"title":"UX Designer"
        "age":24,"gender":"M","location":{"city":"Oklahoma City","state":"Oklahoma"},
        "ranks":[116,162],"salary":8000,"title":"Web Developer"
        "age":63,"gender":"F","location":{"city":"Jersey City","state":"New Jersey"},
        "ranks":[123,126,118,64,110,28],"salary":26220,"title":"Mobile Developer"
        "age":33,"gender":"F","location":{"city":"San Antonio","state":"Texas"},
        "ranks":[97],"salary":14120,"title":"Mobile Developer"

Foot Notes

In these examples, OBJ refers to a DataHammer instance, LIST refers to the list of contained items, and ITEM refers to an item in the contained list or directly in the OBJ.

An attribute dereference (eg: OBJ.index) and the methods OBJ._ind(index) and OBJ._get(index) all function identically, returning a new DataHammer instance. The latter are provided for use when index is an int or otherwise not a valid string identifier.

For most operations and functions that return a new instance, when a DataHammer instance is combined with a list, tuple or other DataHammer instance, the length of the new instance will be limited by the length of the shorter of the two operands. For example:

  • Using a shorter operand, the result will be shortened as if the DataHammer instance had only that many items.

  • Using a longer operand, the result will be as if the DataHammer instance had only as many items as that other operand.

>>> dh1 = DataHammer(range(8))
>>> ~(dh1 + (10, 20))
[10, 21]
>>> dh2 = DataHammer((3, 1, 4))
>>> ~(dh1 == dh2)
[False, True, False]
>>> ~(dh1[dh2])
[3, 1, 4]

Because the [] syntax is used for Indexing and returns an ITEM or list, we cannot use this syntax for chaining or to create another instance as we do for dotted-attribute access. This is why there is a _ind() method, to allow

>>> dh = DataHammer([[i, i*i] for i in range(10, 15)])
>>> ~dh
[[10, 100], [11, 121], [12, 144], [13, 169], [14, 196]]
>>> ~dh._ind(1)
[100, 121, 144, 169, 196]
>>> ~(dh._ind(1) > 125)
[False, False, True, True, True]
>>> ~dh[dh._ind(1) > 125]
[[12, 144], [13, 169], [14, 196]]
>>> dh = DataHammer([dict(a=i, b=tuple(range(i, i*2))) for i in range(6)])

# 'dh.b' returns a DataHammer of N-tuples, then '[3]' retrieves the 4th of these tuples as a `tuple`.
>>> dh.b[2]
(2, 3)

# Here 'dh.b' gives a DataHammer instance of N-tuples, but '_ind(2)' returns another DataHammer
# with the 3rd item from those N-tuples.  Note the `None` for slots where the tuple length.
>>> dh.b._ind(2)
<datahammer.DataHammer object at 0x7f79eb1a9c10>
>>> ~dh.b._ind(2)
[None, None, None, 5, 6, 7]

This works similar to the slice method of the Javascript Array class.

Using “ITEM in OBJ” returns True if ITEM matches one of the items in OBJ, using the operator == for the test. However, using OBJ in OTHER for an iterable containers OTHER, is useless. useless.

Using “OBJ in OTHER” will evaluate the expression “X == OBJ” for each item X in OTHER,, resulting in a list of bool. Unless either OTHER or OBJ are empty, this means a non-empty list will be converted to True even if all of the comparisons fail.

Mutator operations dereference items based on the type of an item, regardless of the type of other items in the contained data. Meaning: if a DataHammer with two items contains a dict with a key “foo” and an object with an attribute “foo”, then using OBJ._mutator().foo will update differently.

The value of a SELECTOR must be a str, but depending on the method can be named or positional. See SELECTOR Examples.

  1. For positional parameters, the text after the last dot, if any, is used for the resulting key.

  2. For named parameters, the value will be used to fetch the value, and the parameter name will be used for the key in the resulting item.

  3. For both, a dot (.) indicates a sub-key, like normal dot notation and/or the _ind() method.


  1. If there are multiple parameters that result in the same key, the result is undefined.

  2. Currently, positional parameters are processed in order before the named parameters, but that is not guaranteed to be true in future releases.

  3. Currently, a bare int (in decimal form) is used to index into lists, but that syntax is not guaranteed to be true in future releases. If a bare int is used as the last component of a postitional parameter value, the resulting key will be a str - the decimal value.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

datahammer-1.0.2.tar.gz (45.5 kB view hashes)

Uploaded source

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page