Source code for equadratures.scalers
"""
Classes to scale data.
Some of these classes are called internally by other modules, but they can also be used independently as a pre-processing stage.
Scalers can fit to one set of data, and used to transform other data sets with the same number of dimensions.
Examples
--------
Fitting scaler implicitly during transform
>>> # Define some 1D sample data
>>> X = np.random.RandomState(0).normal(2,0.5,200)
>>> (X.mean(),X.std())
>>> (2.0354552465705806, 0.5107113843479977)
>>>
>>> # Scale to zero mean and unit variance
>>> X = eq.scalers.scaler_meanvar().transform(X)
>>> (X.mean(),X.std())
>>> (2.886579864025407e-17, 1.0)
Using the same scaling to transform train and test data
>>> # Define some 5D example data
>>> X = np.random.RandomState(0).uniform(-10,10,size=(50,5))
>>> y = X[:,0]**2 - X[:,4]
>>> # Split into train/test
>>> X_train, X_test,y_train,y_test = eq.datasets.train_test_split(X,y,train=0.7,random_seed=0)
>>> (X_train.min(),X_train.max())
>>> (-9.906090476149059, 9.767476761184525)
>>>
>>> # Define a scaler and fit to training split
>>> scaler = eq.scalers.scaler_minmax()
>>> scaler.fit(X_train)
>>>
>>> # Transform train and test data with same scaler
>>> X_train = scaler.transform(X_train)
>>> X_test = scaler.transform(X_test)
>>> (X_train.min(),X_train.max())
>>> (-1.0, 1.0)
>>>
>>> # Finally, e.g. of transforming data back again
>>> X_train = scaler.untransform(X_train)
>>> (X_train.min(),X_train.max())
>>> (-9.906090476149059, 9.767476761184525)
"""
import numpy as np
[docs]class scaler_minmax(object):
""" Scale the data to have a min/max of -1 to 1. """
def __init__(self):
self.fitted = False
[docs] def fit(self,X):
""" Fit scaler to data.
Parameters
----------
X : numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing data to fit scaler to.
"""
if X.ndim == 1: X = X.reshape(-1,1)
self.Xmin = np.min(X,axis=0)
self.Xmax = np.max(X,axis=0)
self.fitted = True
[docs] def transform(self,X):
""" Transforms data. Calls :meth:`~equadratures.scalers.scaler_minmax.fit` fit internally if scaler not already fitted.
Parameters
----------
X : numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing data to transform.
Returns
-------
numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing transformed data.
"""
if X.ndim == 1: X = X.reshape(-1,1)
if not self.fitted: self.fit(X)
Xtrans = 2.0 * ( (X[:,:]-self.Xmin)/(self.Xmax - self.Xmin) ) - 1.0
return Xtrans
[docs] def untransform(self,X):
""" Untransforms data.
Parameters
----------
X : numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing data to untransform.
Returns
-------
numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing untransformed data.
Raises
------
Exception
scaler has not been fitted
"""
if X.ndim == 1: X = X.reshape(-1,1)
if not self.fitted:
raise Exception('scaler has not been fitted')
Xuntrans = 0.5*(X[:,:]+1)*(self.Xmax - self.Xmin) + self.Xmin
return Xuntrans
[docs]class scaler_meanvar(object):
"""
Scale the data to have a mean of 0 and variance of 1.
"""
def __init__(self):
self.fitted = False
[docs] def fit(self,X):
""" Fit scaler to data.
Parameters
----------
X : numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing data to fit scaler to.
"""
if X.ndim == 1: X = X.reshape(-1,1)
self.Xmean = np.mean(X,axis=0)
self.Xstd = np.std(X,axis=0)
self.fitted = True
[docs] def transform(self,X):
""" Transforms data. Calls :meth:`~equadratures.scalers.scaler_meanvar.fit` fit internally if scaler not already fitted.
Parameters
----------
X : numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing data to transform.
Returns
-------
numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing transformed data.
"""
if X.ndim == 1: X = X.reshape(-1,1)
if not self.fitted: self.fit(X)
eps = np.finfo(np.float64).tiny
Xtrans = (X[:,:]-self.Xmean)/(self.Xstd+eps)
return Xtrans
[docs] def untransform(self,X):
""" Untransforms data.
Parameters
----------
X : numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing data to untransform.
Returns
-------
numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing untransformed data.
Raises
------
Exception
scaler has not been fitted
"""
if X.ndim == 1: X = X.reshape(-1,1)
if not self.fitted:
raise Exception('scaler has not been fitted')
eps = np.finfo(np.float64).tiny
Xuntrans = X[:,:]*(self.Xstd+eps) + self.Xmean
return Xuntrans
[docs]class scaler_custom(object):
""" Scale the data by the provided offset and divisor.
Parameters
----------
offset : float, numpy.ndarray
Offset to subtract from data. Either a float, or array with shape (number_of_samples, number_of_dimensions).
div : float, numpy.ndarray
Divisor to divide data with. Either a float, or array with shape (number_of_samples, number_of_dimensions).
"""
def __init__(self, offset, div):
self.offset = offset
self.div = div
self.fitted = True
[docs] def transform(self,X):
""" Transforms data.
Parameters
----------
X : numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing data to transform.
Returns
-------
numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing transformed data.
"""
if X.ndim == 1: X = X.reshape(-1,1)
eps = np.finfo(np.float64).tiny
Xtrans = (X - self.offset)/(self.div + eps)
return Xtrans
[docs] def untransform(self,X):
""" Untransforms data.
Parameters
----------
X : numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing data to untransform.
Returns
-------
numpy.ndarray
Array with shape (number_of_samples, number_of_dimensions) containing untransformed data.
"""
if X.ndim == 1: X = X.reshape(-1,1)
eps = np.finfo(np.float64).tiny
Xuntrans = X * (self.div + eps) + self.offset
return Xuntrans
