Madoka is an implementation of a Count-Min sketch data structure for summarizing data streams.
String-int pairs in a Madoka-Sketch may take less memory than in a standard Python dict, Counter, Redis.
Counting error rate is about 0.0911 %
More details are described in Benchmark.ipynb
This module is based on madoka C++ library.
NOTE: Madoka-Sketch does not have index of keys. so Madoka-Sketch can not dump all keys such as Python dict's dict.keys(). However, when set k parameter to costructer, most_common method (returns key and value as many as k) is available.
Contributions are welcome!
$ pip install madoka
Madoka has some classes having same interface. These classes are vary in value data type. So you can choose for your purpose.
For example, if you wants to count float data, it's preferable to choose CroquisFloat class or CroquisDouble class.
- Sketch - storing unsigned long long (64bit) and fast implementation
- CroquisFloat - storing float (32bit)
- CroquisDouble - storing double (64bit)
- CroquisUint8 - storing unsigned char (8bit)
- CroquisUint16 - storing unsigned short (16bit)
- CroquisUint32 - storing unsigned int (32bit)
- CroquisUint64 - storing unsigned long long (64bit)
From here, I will describe about Sketch class. But, Croquis classes have also same interfaces mostly. So you can use other classes by the same way as Sketch class. In that case, you should replace to intended class from "Sketch".
>>> import madoka
>>> sketch = madoka.Sketch()
- Sketch madoka.Sketch([width=1048576, max_value=35184372088831, path='', flags=0, seed=0, k=5])
- width is a size of register. If you are worrying about gap, you should increase width value. The larger width is, the fewer mistakes madoka makes in estimating value. But, the larger width is, the larger memory consumption is.
- Permission of path should be 644
- k means Top-K used by most_common method. if you don't want to use most_common method, then I recommend to set k=0 so it is slightly fast.
- madoka.Sketch() calls madoka.Sketch.create(), so you don't have to explicitly call create() in initialization
>>> sketch['mami'] += 1
or
>>> sketch.inc('mami')
- int inc(key[, key_length=0])
- Note that key_length is automatically determined when not giving key_length. Thus, the order of parameters differs from original madoka C++ library.
>>> sketch['mami'] += 6
or
>>> sketch.add('mami', 6)
- int add(key, value[, key_length=0])
- Note that key_length is automatically determined when not giving key_length. Thus, the order of parameters differs from original madoka C++ library.
>>> sketch['mami'] = 6
or
>>> sketch.set('mami', 6)
- void set(key, value[, key_length=0])
- Note that set() does nothing when the given value is not greater than the current key value.
- Also note that the new value is saturated when the given value is greater than the upper limit.
- Additionally note that key_length is automatically determined when not giving key_length. Thus, the order of parameters differs from original madoka C++ library.
>>> sketch['mami']
or
>>> sketch.get('mami')
- int get(key[, key_length=0])
- Note that key_length is automatically determined when not giving key_length. Thus, the order of parameters differs from original madoka C++ library.
>>> sketch.values()
- generator<int> values()
- Note that processing time increases according to sketch's width. But this method may be slow, so I recommend setting width to less than 1000000 when creating sketch.
>>> sketch.save('example.madoka')
- void save(path)
- Permission of path should be 644
>>> sketch.load('example.madoka')
- void load(path)
- Permission of path should be 644
>>> sketch.clear()
- void clear()
- Delete all key-value pairs. It differs from create() in maintaining current settings.
>>> sketch.create()
- void create([width=0, max_value=0, path=NULL, flags=0, seed=0])
- Permission of file given to path should be 644
>>> sketch.copy(othersketch)
- void copy(Sketch)
>>> sketch += other_sketch
or
>>> sketch.merge(othersketch)
- void merge(Sketch[, lhs_filter=None, rhs_filter=None])
- lhs_filter is applied for self.sketch, rhs_filter is applied for given sketch
>>> sketch.shrink(sketch, width=1000)
- void shrink(Sketch[, width=0, max_value=0, filter=None, path=None, flags=0])
- When width > 0, width must be less than source sketch
- Permission of path should be 644
>>> summed_sketch = sketch + other_sketch
- Create summed sketch, So it does not break original sketches
>>> summed_sketch = sketch + {'mami': 1, 'kyoko': 2}
- Create summed sketch, So it does not break original sketches
>>> 'mami' in sketch
>>> sketch.inner_product(other_sketch)
- list<float> inner_product(Sketch)
- Returns [inner product, square length of left hands sketch (float), square length of right hands sketch (float)]
>>> sketch['madoka'] = 1
>>> sketch['mami'] = 2
>>> sketch['sayaka'] = 3
>>> sketch['kyouko'] = 4
>>> sketch['homura'] = 5
>>> sketch.median() # => 3
- int or float median()
>>> sketch.filter(lambda x: x + 1)
- void filter(Callable[, apply_zerovalue=False])
- If apply_zerovalue = True, filter_method is applied also 0 values (It may be slow) (from version 0.6 or later)
- Note that processing time increases according to sketch's width. If you feel this method is slow, I recommend setting width to less than 1000000 when creating sketch
>>> sketch.fromdict({'mami': 14, 'madoka': 13})
or
>>> sketch += {'mami': 14, 'madoka': 13}
- void fromdict(dict)
>>> sketch.most_common()
- generator most_common([k=5])
- returns key-value pair as many as k
- Note that this method is required to set k parameter in constructer.
- Wrapper code is licensed under New BSD License.
- Bundled madoka C++ library is licensed under the Simplified BSD License.