filterbox 0.7.2

Container for finding Python objects by matching attributes. Stores objects by attribute value for fast lookup.

FilterBox

Container for finding Python objects by matching attributes.

Stores objects in hashtables by attribute value, so finds are very fast.

Finding objects using FilterBox can be 5-10x faster than SQLite.

Install

pip install filterbox

Usage

Find which day will be good for flying a kite. It needs to be windy and sunny.

from filterbox import FilterBox

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

def is_windy(obj):
    return obj['wind_speed'] > 5

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

# find objects by function and / or attribute values
fb.find({is_windy: True, 'sky': 'sunny'})  
# result: [{'day': 'Monday', 'wind_speed': 7, 'sky': 'sunny'}]

There are three classes available.

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

More Examples

Expand for sample code.

Match and exclude multiple values

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': [10, 20]},                # match anything with size in [10, 20] 
    exclude={'flavor': ['lychee', 'peach']}  # where flavor is not in ['lychee', 'peach']
)  
# result: [{'item': 1, 'size': 10, 'flavor': 'melon'}]

Accessing 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]}}

Greater than, less than

FilterBox does == very well, but < and > take some extra effort.

Suppose you need to find objects where x >= some number. If the number is constant, a function that returns obj.x >= constant will work.

Otherwise, FilterBox and FrozenFilterBox have a method get_values(attr) which gets the set of unique values for an attribute. Here's how to use it to find objects having x >= 3.

from filterbox import FilterBox

data = [{'x': i} for i in [1, 1, 2, 3, 5]]
fb = FilterBox(data, ['x'])
vals = fb.get_values('x')                # get the set of unique values: {1, 2, 3, 5}
big_vals = [x for x in vals if x >= 3]   # big_vals is [3, 5]
fb.find({'x': big_vals})                 # result: [{'x': 3}, {'x': 5}

If x is a float or has many unique values, consider making a function on x that rounds it or puts it into a bin of similar values. Discretizing x in ths way will make lookups faster.

Handling 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

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

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

Recipes

• Auto-updating - Keep FilterBox updated when attribute values 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 every attribute in FilterBox, it holds a dict that maps each unique value to the set of objects with that value.

This is the rough idea of the data structure:

class FilterBox:
    indices = {
        'attribute1': {val1: set(some_obj_ids), val2: set(other_obj_ids)},
        'attribute2': {val3: set(some_obj_ids), val4: 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 more complexity, 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:

• FilterBox
• ConcurrentFilterBox
• FrozenFilterBox

API Reference:

• FilterBox API
• ConcurrentFilterBox API
• FrozenFilterBox API

Related projects

FilterBox is a type of inverted index. It is optimized for its goal of finding in-memory Python objects.

Other Python inverted index implementations are aimed at things like vector search and finding documents by words. Outside of Python, ElasticSearch is a popular inverted index search tool. Each of these has goals outside of FilterBox's niche; there are no plans to expand FilterBox towards these functions.

---