transform.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
 
 
 
""" Transformation classes
 
In this file all classes for the transformation methods are defined.
The base class is Transform.
 
 
"""
 
 
from alignment.fancystuff.settings import ProTool
 
import numpy as np
import pandas as pd
from scipy.interpolate import InterpolatedUnivariateSpline
 
 
class Transform(ProTool):
 
    """
    Base Class for the transformations.
    The function _fit() is overwritten by the sub classes.
 
    Attributes
    ----------
    n_bins : int, optional
        Binning in x, will be set automatically
    max : float
        Maximum of the fitted distribution
    min : float
        Minimum of the fitted distribution
    is_processed : bool
        Status flag
    name : str
        Name of the transformation
 
    """
 
    def __init__(self, name="Original", n_bins=None):
        """ Init function
 
        :param name:    Name
        :param n_bins:  Binning for the transformations
        """
        
        self.n_binsn_bins = n_bins
 
        
        self.maxmax = 0
 
        
        self.minmin = 0
 
        
        self.is_processedis_processed = False
 
        
        self.namenamename = name
 
        ProTool.__init__(self, "Transform." + self.namenamename)
 
    def _initialise(self, x):
        """
        Sets limits for the data.
        Not called by the user.
 
        :param x: array type
        """
        self.ioio.debug("Initiating " + self.namenamename)
        if self.n_binsn_bins is None:
            self.set_n_binsset_n_bins(len(x))
        self.maxmax = np.max(x)
        self.minmin = np.min(x)
 
    def fit(self, x, y=None):
        """
        The fit function is calls the individual _fit() functions.
 
        :param x:   Distribution to fit, array type
        :param y:   optional for some transformations, sets signal class
        """
        self._initialise_initialise(x)
        self._fit_fit(x, y)
        self.is_processedis_processed = True
 
    def __call__(self, x):
        """ Call function calls transform
        :param x:   Input data
        :return:    Transformed data
        """
        return self.transformtransform(x)
 
    def _fit(self, x, y=None):
        """
        This is defined in the children and overwritten.
        :param x:   array x values
        :param y:   class variable [1,0]
 
        """
 
    def transform(self, x, set_limits=False):
        """
        This is defined in the children and overwritten.
        :param x:           Distribution to transform, array type
        :param set_limits:  Limits the range of the data to the fitted range
        :return:            Transformed data
        """
        if set_limits:
            self.set_limits(x)
        return self._transform(x)
 
    def _transform(self, x):
        """
        This is defined in the children and overwritten.
        In the base class it does nothing and returns the original distribution.
 
        :param x:   Distribution to transform, array type
        :return:    Transformed data
        """
        return x
 
    def set_n_bins(self, n):
        """
        Calculates the optimal size for the binning.
        :param n:   Length of the input data
        """
        self.n_binsn_bins = get_optimal_bin_size(n)
        self.ioio.debug("Bins are set to " + str(self.n_binsn_bins) + "\t " + str(n / float(self.n_binsn_bins)) + "per bin")
 
    def set_limits(self, x):
        """
        Limits the data to the fitted range.
        :param x:   Input data
        :return:    Limited data
        """
        try:
            _ = len(x)  # to catch exception
            x[x > self.maxmax] = self.maxmax
            x[x < self.minmin] = self.minmin
        except TypeError:
            if x < self.minmin:
                x = self.minmin
            if x > self.maxmax:
                x = self.maxmax
        return x
 
 
def get_optimal_bin_size(n):
    """
    This function calculates the optimal amount of bins for the number of events n.
    :param      n:  number of Events
    :return:        optimal bin size
 
    """
    return int(3 * n ** (1 / 3.0))
 
 
def get_average_in_bins(n):
    """
    Returns the expected amount of entries in each bins.
    :param n:   Length of the data
    :return:    Length of the data divided by the optimal bin size
    """
    return n / float(get_optimal_bin_size(n))
 
 
class CDF(Transform):
 
    """
    Calculates the cumulative distribution (CDF)
    Can be used for the flat transformation.
 
    Attributes
    ----------
    spline : InterpolatedUnivariateSpline
        Spline, fitting the CDF
 
    """
 
    def __init__(self, *args):
        """ Init function
 
        :param args: None
        """
        Transform.__init__(self, "CDF", *args)
 
        
        self.splinespline = None
 
    def _fit(self, x, y=None):
        """
        Fit function calculates the cumulative distribution with numpy percentile.
 
        :param x:   Input distribution
        :param y:   Will not be used in this transformation
        """
        self.ioio.debug("Fitting CDF")
        y_ = np.linspace(0, 100, 2 * self.n_binsn_bins)
        x_ = pd.Series(np.percentile(x, list(y_)))
 
        # Count same values
        vc = x_.value_counts()
        vc = vc.sort_index()
 
        # replace same values
        for i, xi in enumerate(vc):
            if xi > 1:
                try:
                    nex_val = vc.index[i + 1]
                except IndexError:
                    nex_val = vc.index[i] + 0.01
                fill = np.linspace(vc.index[i], nex_val, xi)
                x_[x_ == vc.index[i]] = fill
        self.splinespline = InterpolatedUnivariateSpline(x_, y_)
 
    def _transform(self, x):
        """
        Transforms the input data according to the cdf.
        :param x:   Input data
        :return:    Transformed data
        """
        x = self.set_limitsset_limits(x)
        return self.splinespline(x)
 
 
class ToFlat(Transform):
 
    """
    This transformation uses the CDF to transform input data to a
    flat transformation.
 
    Attributes
    ----------
    cdf : Transform.CDF
        Transformation with the CDF
 
    """
 
    def __init__(self, *args):
        """ Init function
 
        :param args: None
        """
        Transform.__init__(self, "Flat", *args)
 
        
        self.cdfcdf = CDF(*args)
 
    def _fit(self, x, y=None):
        """
        Fit function calculates the cumulative distribution with numpy percentile.
 
        :param x:   Inout distribution
        :param y:   Will not be used in this transformation
        """
        self.ioio.debug("Fitting Flat")
        self.cdfcdf.fit(x)
 
    def _transform(self, x):
        """
        Transforms the input data according to the cdf.
        :param x:   Input data
        :return:    Transformed data
        """
        if not self.is_processedis_processed:
            self.fitfit(x)
        return self.cdfcdf.transform(x)
 
    def get_flat_bins(self):
        """
        Returns the binning of the CDF
        :return: Binning for a flat distribution
        """
        return self.cdfcdf.x
 
    def get_x(self, x_flat):
        """
        Dirty version for getting the original x value out of a flat x value.
        :param x_flat:  x value in the flat distribution
        :return:        x value on the original axis (approx)
        """
        x_cumul = np.linspace(self.minmin, self.maxmax, self.n_binsn_bins * 50)
        for xx in x_cumul:
            if self.cdfcdf.spline(xx) > x_flat:
                return xx