ezstat 1.4

OO For eazy statistics, inspired by `Statistics` class of the lib `deap` but more user-friendly

Ezstat: Easy statistics

OO For easy statistics, inspired by Statistics class of deap

It is really easy and awesome! Believe me!

Introduction

ezstat is built for easy statistics, esp. for the history of iterations.

Statistics

The main class Statistics just extends dict{str:function} (called statistics dict), function here will act on the object of statistics. The values of dict have not to be a function, if it is a string, then the object of method with the same name is applied.

Frankly, It just borrows the idea from the Statistics class of deap. But unlike the author of deap, I just create it a subclass of dict, need not define strange methods.

See the following example and function _call, the underlying implementation.

Examples

Example:

>>> import numpy as np

>>> T = np.random.random((100,100)) # samples(one hundrand 100D samples)
>>> stat = Statistics({'mean': np.mean, 'max': 'max', 'shape':'shape'}) # create statistics
>>> print(stat(T))
>>> {'mean': 0.5009150557686407, 'max': 0.5748552862392957, 'shape': (100, 100)}

>>> print(stat(T, split=True)) # split the tuple if it needs
>>> {mean': 0.5009150557686407, 'max': 0.5748552862392957, 'shape[0]': 100, 'shape[1]': 100}

# with sub-statistics
s = Statistics({'mean': np.mean,
'extreme': {'max':'max', 'min':np.min},  # as a sub-statistics
'shape':'shape'})
print(s(X))
# dict-valued statistics, equivalent to the above
s = Statistics({'mean': np.mean, 'extreme': lambda x:{'max': np.max(x), 'min': np.min(x)}, 'shape':'shape'})
print(s(X))

#Result: {'mean': 0.49786554518848564, 'extreme[max]': 0.9999761647791217, 'extreme[min]': 0.0001368184546896023, 'shape': (100, 100)}

MappingStatistics

MappingStatistics is a subclass of Statistics. It only copes with iterable object, and maps the obect to an array by funcional attribute key.

Example:

>>> stat = MappingStatistics(key='mean', {'mean':np.mean, 'max':np.max})
>>> print(stat(T))
>>> {'mean': 0.5009150557686407, 'max': 0.5748552862392957}

In the exmaple, 'mean', an attribute of T, maps T to a 1D array.

Advanced Usage

Statistics acts on a list/tuple of objects iteratively, gets a series of results, forming an object of pandas.DataFrame. In fact, it is insprited by Statistics class of third part lib deap. In some case, it collects a list of dicts of the statistics result for a series of objects. It is suggested to transform to DataFrame object.

history = pd.DataFrame(columns=stat.keys())
for obj in objs:
    history = history.append(stat(obj), ignore_index=True)

To Do

• To define tuple of functions for the value of statistics dict.