Merge branch 'nxt_ver_0_3_2' into 'master'

from sasmodels.core import load_model_info

from sasmodels.sasview_model import make_model_from_info

model_info = load_model_info('fractal+mono_gauss_coil')

model_info.name = 'addition_example'

model_info.description = 'fractal + mono_gauss_coil'

Model = make_model_from_info(model_info)

file: 

    name: sphere_nz_unc.txt

    qmin: 0.01

    qmax: 0.30

output_files: addition_example

sa_parameters: 

    temperatures: 100

    temperature_rate: 0.90

    parameter_rate: 0.95

    iterations: 1000

    models_to_generate: 5

model:

    name: addition_example.py

    category: shape-independent

    scale:

        fixed: [1.00]

    background: 

        linear: [0.000, 0.10]

    A_scale:

        linear: [0.001, 1.0]

    A_radius:

        linear: [100.0, 200.0]

    A_fractal_dim:

        linear: [1.0, 4.0]

    A_cor_length:

        linear: [100.0, 500.0]

    A_sld_block:

        linear: [2.0, 4.0]

    A_sld_solvent:

        fixed: [6.4]

    B_scale:

        linear: [0.001, 0.10]

    B_i_zero:

        fixed: [1.0]

    B_rg:

        linear: [20.0, 50.0]

# perform the traditional estimation of the uncertainties in the

# fitting parameters by looking at the partial derivatives using the

# approach that is used in Paul Kienzle's "bump", as is noted below.    

# Jacobian matrix.  The approach is not perfect, and can give odd

# values that do not seem consistent with what a person using it 

# might expect.  This may be removed from future releases, or it will

# only be available if a single model is being generated.

#

# This approach to the estimation is one of a few that is implemented in bumps.

# bumps is by Paul Kienzle, who works at NIST's NCNR

# https://github.com/bumps/

# see https://github.com/bumps/bumps/blob/master/bumps/lsqerror.py 

# bumps is the fitting engine that Sasview uses with sasmodels,

# but it is an independent package.

def est_uncerts(d, f, modconf, best_model):

    # this is our ugly way of getting at this matrix of derivatives

    loc = modelconfig.ModelConfig(f.name,f.category,f.params,f.sq)

    # print("Jacobian")

    # print(J)

    # this is the singlular value decomposition of J

    # see https://github.com/bumps/bumps/blob/master/bumps/lsqerror.py lines 237-239

    # This is the singlular value decomposition of J.

    # The tolerance shown cuts down on singlular values

    # bumps as a whole may not use this directly.  have to give it a try, though

    # bumps as a whole may not use this directly.

    U, S, V = np.linalg.svd(J, 0)

    tol = 1e-8

    S[S <= tol] = tol