filterbox 0.9.4

Finds Python objects by attribute value

FilterBox

Container for finding Python objects.

Stores objects by their attribute value. Uses B-tree indexes to make finding very fast.

Install

pip install filterbox

Usage

Find a good day for flying a kite. It needs to have sunny skies and a wind speed between 5 and 10.

from filterbox import FilterBox

days = [
    {'day': 'Saturday', 'sky': 'sunny', 'wind_speed': 1},
    {'day': 'Sunday', 'sky': 'rainy', 'wind_speed': 3},
    {'day': 'Monday', 'sky': 'sunny', 'wind_speed': 7},
    {'day': 'Tuesday', 'sky': 'rainy', 'wind_speed': 9},
    {'day': 'Wednesday', 'sky': 'sunny', 'wind_speed': 25}
]

# make a FilterBox
fb = FilterBox(               # make a FilterBox
    days,                     # add objects of any Python type
    on=['sky', 'wind_speed']  # attributes to find by
)

fb.find({
    'sky': 'sunny', 
    'wind_speed': {'>=': 5, '<=': 10}
})  
# result: [{'day': 'Monday', 'sky': 'sunny', 'wind_speed': 7}]

You can also find objects by functions evaluated on the object.

Find palindromes of length 5 or 7:

from filterbox import FilterBox
strings = ['bob', 'fives', 'kayak', 'stats', 'pullup', 'racecar']

def is_palindrome(s):
    return s == s[::-1]

fb = FilterBox(strings, [is_palindrome, len])
fb.find({
    is_palindrome: True, 
    len: {'in': [5, 7]}
})
# result: ['kayak', 'racecar', 'stats']

Functions are evaluated only once, when the object is added to the FilterBox.

Classes

• FilterBox - can add, remove, and update objects after creation.
• ConcurrentFilterBox - Thread-safe version of FilterBox.
• FrozenFilterBox - Cannot be changed after creation. Fastest finds, lower memory usage, and thread-safe.

All three can be pickled using filterbox.save() / filterbox.load().

More Examples

Expand for sample code.

Exclude values

find() takes two arguments, match and exclude. The examples up to this point have used match only, but exclude works the same way.

from filterbox import FilterBox

objects = [
    {'item': 1, 'size': 10, 'flavor': 'melon'}, 
    {'item': 2, 'size': 10, 'flavor': 'lychee'}, 
    {'item': 3, 'size': 20, 'flavor': 'peach'},
    {'item': 4, 'size': 30, 'flavor': 'apple'}
]

fb = FilterBox(objects, on=['size', 'flavor'])

fb.find(
    match={'size': {'<': 30}},                       # match anything with size < 30
    exclude={'flavor': {'in': ['lychee', 'peach']}}  # where flavor is not in ['lychee', 'peach']
)  
# result: [{'item': 1, 'size': 10, 'flavor': 'melon'}]

If match is unspecified, all objects will match, which allows filtering solely by exclude. Try removing the match line in the example above.

Access nested data using functions

Use functions to get values from nested data structures.

from filterbox import FilterBox

objs = [
    {'a': {'b': [1, 2, 3]}},
    {'a': {'b': [4, 5, 6]}}
]

def get_nested(obj):
    return obj['a']['b'][0]

fb = FilterBox(objs, [get_nested])
fb.find({get_nested: 4})  
# result: {'a': {'b': [4, 5, 6]}}

Handle missing attributes

Objects don't need to have every attribute.

• Objects that are missing an attribute will not be stored under that attribute. This saves lots of memory.
• To find all objects that have an attribute, match the special value ANY.
• To find objects missing the attribute, exclude ANY.
• In functions, raise MissingAttribute to tell FilterBox the object is missing.

Example:

from filterbox import FilterBox, ANY
from filterbox.exceptions import MissingAttribute

objs = [{'a': 1}, {'a': 2}, {}]

def get_a(obj):
    try:
        return obj['a']
    except KeyError:
        raise MissingAttribute  # tell FilterBox this attribute is missing

fb = FilterBox(objs, ['a', get_a])

fb.find({'a': ANY})          # result: [{'a': 1}, {'a': 2}]
fb.find({get_a: ANY})        # result: [{'a': 1}, {'a': 2}]
fb.find(exclude={'a': ANY})  # result: [{}]

Note that None is treated as a normal value and is stored.

Recipes

• Auto-updating - Keep FilterBox updated when objects change
• Wordle solver - Solve string matching problems faster than regex
• Collision detection - Find objects based on type and proximity (grid-based)
• Percentiles - Find by percentile (median, p99, etc.)

---

How it works

For each attribute in the FilterBox, it holds a tree that maps every unique value to the set of objects with that value.

This is a rough idea of the data structure:

class FilterBox:
    indexes = {
        'attribute1': BTree({10: set(some_obj_ids), 20: set(other_obj_ids)}),
        'attribute2': BTree({'abc': set(some_obj_ids), 'def': set(other_obj_ids)}),
    }
    'obj_map': {obj_ids: objects}
}

During find(), the object ID sets matching each query value are retrieved. Then set operations like union, intersect, and difference are applied to get the matching object IDs. Finally, the object IDs are converted to objects and returned.

In practice, FilterBox and FrozenFilterBox have a bit more to them, as they are optimized to have much better memory usage and speed than a naive implementation.

See the "how it works" pages for more detail:

• How FilterBox works
• How ConcurrentFilterBox works
• How FrozenFilterBox works

API Reference:

• FilterBox API
• ConcurrentFilterBox API
• FrozenFilterBox API

Why not SQLite?

SQLite is an awesome relational database, and its in-memory storage option allows it to be used as a Python object container. For example, LiteBox is a container that uses SQLite as an index. This is popular, and works fairly well.

But if you don't need a database - and only need to find Python objects - FilterBox is far superior.

The FilterBox containers have many advantages over SQLite:

• They are faster. Finding objects using FilterBox can be 5-10x faster than SQLite.
• They use sparse representations. Objects do not need to fill in "NULL" for missing attributes, those attributes are simply not stored.
• They can query any Python data type, not just numbers and strings. While there are tricks to get around this in SQLite, those tricks incur other costs in flexibility, complexity, and/or speed.
• There is no need to translate datatypes (serialize / deserialize), allowing much faster finds.
• FrozenFilterBox is immutable, and so implements optimizations that are not possible in SQLite.
• They are much simpler. You'll never worry about whether you've VACUUMed a FilterBox.

---