Package gmisclib :: Module g_localfit

Module g_localfit

Fit a linear transform to a bunch of tt input/output vectors.

Notes:

If you compute Pearson's R^2 via 1-localfit()/err_before_fit(), you get the "unadjusted" R^2 value. There is also an "adjusted" R^2 value, which is Ra^2 = 1-(1-R^2)*(n-1)/(n-p-1) where n is the number of data and p is the number of adjustable parameters of the linear regression.
When you are computing Pearson's r^2 by way of localfit()/err_before_fit() the two function calls MUST have the same data and flags. Specifically, constant must be equal in both calls.

Functions

numpy.ndarray

err_before_fit(data, minsv=None, minsvr=None, constant=True)
How much variation did the data have before the fit? This is used to compare with the error after the fit, to allow a F-test or ANOVA.

source code

pack(data, constant=True)
Prepare the data array and (optionally) weight the data.

source code

(A, B, errs, sv, rank) where

A is a 2-D numpy.ndarray matrix.
B is a 1-D numpy.ndarray vector (if constant, else None).
errs is a numpy.ndarray vector, one value for each output coordinate. The length is the same as the out vector (see pack).
sv is a numpy.ndarray vector. The length is the same as the in vector (see pack).

localfit(data, minsv=None, minsvr=None, constant=True)
Does a linear fit to data via a singular value decomposition algorithm.

source code

reg_localfit(data, regstr=0.0, regtgt=None, rscale=None, constant=True)
Does a linear fit to data via a singular value decomposition algorithm.

source code

robust_localfit(data, minsv=None, minsvr=None, constant=True)
Data is [ (input_coordinates, output_coordinates), ...

source code

fit_giving_sigmas(data, minsv=None, minsvr=None, constant=True)
Does a linear fit to data via a singular value decomposition algorithm.

source code

leaktest()

source code

test0()

source code

test_localfit11(r)

source code

test_localfit11e()

source code

test_localfit21()

source code

test_localfit21u()

source code

test_localfit22(r)

source code

test_fgs1()

source code

test_wt()

source code

Variables
	__package__ = `'gmisclib'`

Imports: numpy, warnings, gpk_lsq, gpk_rlsq

Function Details

err_before_fit(data, minsv=None, minsvr=None, constant=True)

source code

How much variation did the data have before the fit? This is used to compare with the error after the fit, to allow a F-test or ANOVA. Strictly speaking, we compute the mean-squared error about a constant.

Parameters:

data (See pack.) - See pack.

Returns: numpy.ndarray

Returns the per-coordinate sum-squared-error of the output coordinates.

pack(data, constant=True)

source code

Prepare the data array and (optionally) weight the data.

Parameters:

data ([(numpy.ndarray, numpy.ndarray), ...] or [(numpy.ndarray, numpy.ndarray, float), ...]) - [(input_coordinates, output_coordinates), ...] or [(input_coordinates, output_coordinates, weight), ...]. A list of (in,out) tuples corresponding to the independent and dependent parameters of the linear transform. Both in and out are one-dimensional numpy vectors. They don't need to have the same length, though (obviously) all instances of "in" need to have the same length and all instances of "out" also need to match one another. If weight is given, it must be a scalar.

localfit(data, minsv=None, minsvr=None, constant=True)

source code

Does a linear fit to data via a singular value decomposition algorithm. It returns the matrix A and vector B such that A*input_coordinates+B is the best fit to output_coordinates.

Parameters:

minsv (float or None) - sets the minimum useable singular value;
minsvr (float or None) - sets the minimum useable s.v. in terms of the largest s.v.
constant (Boolean) - Do you want a constant built into the linear equations?
data (See pack.) - list of tuples. See pack.

Returns: (A, B, errs, sv, rank) where A is a 2-D numpy.ndarray matrix. B is a 1-D numpy.ndarray vector (if constant, else None). errs is a numpy.ndarray vector, one value for each output coordinate. The length is the same as the out vector (see pack). sv is a numpy.ndarray vector. The length is the same as the in vector (see pack).

(A, B, errs, sv, rank) where

A is the matrix of coefficients
B is a vector of constants. (Or None) One constant per component of the out vector in pack.
errs Each component gives the sum of squared residuals for the corresponding component of the out vector.
sv are the singular values, sorted into decreasing order.

reg_localfit(data, regstr=0.0, regtgt=None, rscale=None, constant=True)

source code

Does a linear fit to data via a singular value decomposition algorithm. It returns the matrix A and vector B such that A*input_coordinates+B is the best fit to output_coordinates.

Parameters:

constant (Boolean) - Do you want a constant built into the linear equations?
data (See pack.) - list of tuples. See pack.

Returns:

(A, B, errs, sv, rank) where

A is a 2-D numpy.ndarray matrix.
B is a 1-D numpy.ndarray vector (if constant, else None).
errs is a numpy.ndarray vector, one value for each output coordinate. It gives the sum of squared residuals.
sv are the singular values, sorted into decreasing order.

robust_localfit(data, minsv=None, minsvr=None, constant=True)

source code

Data is [ (input_coordinates, output_coordinates), ... ] Minsv sets the minimum useable singular value; minsvr sets the minimum useable s.v. in terms of the largest s.v.. It returns the matrix A and vector B such that the best fit is A*input_coordinates+B in a tuple (A, B, errs, rank). errs is a vector, one value for each output coordinate. Rank is the minimum rank of the various fits.

Warning! Not tested.

fit_giving_sigmas(data, minsv=None, minsvr=None, constant=True)

source code

Does a linear fit to data via a singular value decomposition algorithm. It returns the matrix A and vector B such that A*input_coordinates+B is the best fit to output_coordinates.

Parameters:

minsv (float or None) - sets the minimum useable singular value;
minsvr (float or None) - sets the minimum useable s.v. in terms of the largest s.v.
constant (Boolean) - Do you want a constant built into the linear equations?
data ([(numpy.ndarray, numpy.ndarray), ...]) - [(input_coordinates, output_coordinates), ...]. A list of (in,out) tuples corresponding to the independent and dependent parameters of the linear transform. Both in and out are one-dimensional numpy vectors. They don't need to have the same length, though (obviously) all instances of "in" need to have the same length and all instances of "out" also need to match one another.

Returns:

(A, B, sigmaA, sigmaB) where

A is a 2-D numpy.ndarray matrix.
B is a 1-D numpy.ndarray vector (if constant, else None). )