#!/usr/bin/env python
#-*- coding: utf-8 -*-
"""Regression functions
.. topic:: regression.py summary
A module dedicated to regression functionalities
:Code status: mature
:Documentation status: to be completed
:Author: Thomas Cokelaer <Thomas.Cokelaer@sophia.inria.fr>
:Revision: $Id$
"""
__version__ = "$Id$"
from . import error
from . import interface
from openalea.stat_tool._stat_tool import _Regression
from openalea.stat_tool._stat_tool import _Vectors
from openalea.stat_tool._stat_tool import _DiscreteParametricModel
from openalea.stat_tool._stat_tool import _Compound
from openalea.stat_tool._stat_tool import _Convolution
from openalea.stat_tool._stat_tool import _DiscreteMixture
from .enums import algo_map
from .enums import bool_type
__all__ = ['_Regression',
'Regression']
[docs]
def Regression(vec, utype, explanatory, response,
*args, **kargs):
"""
Simple regression (with a single explanatory variable).
:Parameters:
* vec : vectors
vectors
* type : string
`"Linear"` or `"MovingAverage"` or `"NearestNeighbors"`
* explanatory_variable : int
index of the explanatory variable
* response_variable : int
index of the response variable
* filter : list of float
filter values on the half width i.e. from one extremity to the central value
(with the constraint sum(filter) = 1),
* frequencies : list of float
frequencies defining the filter,
* dist : distribution, mixture, convolution, compound
symmetric distribution, whose size of the support is even, defining the filter
(for instance Distribution("BINOMIAL",0,4,0.5)),
* span : float
proportion of individuals in each neighbourhood.
:Keywords:
* Algorithm : string
- `"Averaging"` (default)
- `"LeastSquares"`
This optional argument can only be used if the second mandatory argument
specifying the regression type is "MovingAverage".
* Weighting : bool
weighting or not of the neighbors according to their distance to the
computed point (default value: True). This optional argument can only
be used if the second mandatory argument specifying the regression type
is "NearestNeighbors".
:Returns:
An object of type regression is returned.
:Examples:
.. doctest::
:options: +SKIP
>>> Regression(vec, "Linear", explanatory_variable, response_variable)
>>> Regression(vec, "MovingAverage", explanatory_variable,
... response_variable, filter, Algorithm="LeastSquares")
>>> Regression(vec, "MovingAverage", explanatory_variable,
.. response_variable, frequencies, Algorithm="LeastSquares")
>>> Regression(vec, "MovingAverage", explanatory_variable,
... response_variable, dist, Algorithm="LeastSquares")
>>> Regression(vec, "NearestNeighbors", explanatory_variable,
... response_variable, span, Weighting=False)
.. seealso::
:func:`~openalea.stat_tool.output.Plot`
"""
STAT_MINIMUM_SPAN = 0.05 # from aml not stat_tool or sequence headers
error.CheckType([vec, utype, explanatory, response],
[_Vectors, str, int, int])
possible_types = ["Linear", "NearestNeighbors",
"NearestNeighbours", "MovingAverage"]
Algorithm = error.ParseKargs(kargs, "Algorithm", 'Averaging', algo_map)
Weighting = error.ParseKargs(kargs, "Weighting", True, bool_type)
if utype == "Linear":
error.CheckArgumentsLength(args, 0, 0)
return vec.linear_regression(explanatory, response)
elif utype == "MovingAverage":
error.CheckArgumentsLength(args, 1, 1)
param = args[0]
#todo add CheckType for int and models
# param is a list of float, int
if isinstance(args[0], list):
# todo: check that sum equals 1
return vec.moving_average_regression_values(explanatory,
response, param, Algorithm)
# or a set of distributions
# todo: test case of compound, convolution, mixture
else:
error.CheckType([param],
[[_DiscreteParametricModel, _DiscreteMixture,
_Convolution, _Compound]])
return vec.moving_average_regression_distribution(
explanatory, response,
param, Algorithm)
elif utype in ["NearestNeighbors", "NearestNeighbours"]:
error.CheckArgumentsLength(args, 1, 1)
span = args[0]
error.CheckType([span], [[float, int]])
assert span >= STAT_MINIMUM_SPAN
#todo: check this assert
return vec.nearest_neighbours_regression(explanatory, response,
float(span), Weighting)
else:
raise TypeError("Bad Regression type. Must be in %s" % possible_types)
# Extend _Regression class dynamically
interface.extend_class( _Regression, interface.StatInterface)
