IltData attributes
IltData
holds the data and parameters required for inversion. Immediately after using iltload
, the following data-related attributes
parameters are available :
IltData.data
: Processed data read byiltload
IltData.t
: Arrays stored as a list representing the input sampling vectors corresponding to the dimensions ofIltData.data
.IltData.filepath
: Filepath which was provided toiltload
, in case of loading from numpy arrays, this corresponds to the path of the current working directory.
After a successful inversion, the following additional attributes are defined:
IltData.g
: The distribution (or spectrum) obtained after inversion.IltData.fit
: Fit of the data obtained after inversion.IltData.residuals
: Residuals of the fit obtained after inversion.
Brief descriptions of other attributes are given below. More detailed explanations for adjustable parameters can be found here.
Attributes
----------
alpha_0 : float, optional
Baseline regularization in regions without spectral density of the inverted distribution.
Default is 1e-4.
alpha_00 : float, optional
Global scaling factor for the regularization term. Default is 1.
alpha_p : float or None, optional
Reduces the curvature penalty as a function of the slope, especially for broad features.
Default is None.
alpha_c : float or None, optional
Weight of the curvature penalty in uniform penalty regularization. Default is None.
alpha_d : float or None, optional
Weight of the zero-crossing (ZC) penalty. If None, it is set to 1/alpha_a. Default is None.
alpha_a : float or None, optional
Adjusts contribution of zero crossings to the regularization. Default is 1e5.
alpha_bc : float, optional
Boundary regularization weight to ensure smooth edge behavior. Default is 50.
alpha_nn : float, optional
Weight of the non-negativity penalty. Default is 1000.
alt_g : int, optional
Algorithm choice for time-limiting inversion. Allowed values are {0, 2, 4, 6}. Default is 0.
compress : bool or list of bool, optional
Enable/disable singular value decomposition (SVD) based data compression. Default is False.
use_svds : bool or list of bool, optional
Use sparse algorithm for SVD of highly sparse matrices. Default is False.
s : int or list of int, optional
Number of singular values to retain during compression. Default is 15.
sigma : ndarray, optional
Standard deviation of noise for each data point. Must match shape of data. Default is None.
g_guess : ndarray, optional
Initial guess for the distribution, based on prior knowledge. Default is None.
c_nmax : int, optional
Window size for moving max in curvature term calculation. Default is 3.
p_nmax : int, optional
Window size for moving max in slope term calculation. Default is 3.
reg_bc : int or None, optional
Number of edge points regularized for boundary smoothness. Default is None.
kernel : IltKernel, optional
Kernel function for inversion. Must be an instance of IltKernel. Default is IltKernel().
base_tau : int, {0, 1}, optional
Defines spacing for `tau`. 1 for logarithmic, 0 for linear. Default is 1.
maxloop : int, optional
Maximum number of iterations before termination. Default is 100.
nn : bool, optional
Enable non-negativity penalty. Default is False.
reg_down : int, optional
Iteration after which the regularization update rate starts decreasing. Default is 29.
reg_downrate : float, optional
Rate of decrease of the regularization update after `reg_down`. Default is 0.05.
reg_upd : float, optional
Initial fraction of regularization update. Default is 1.
reg_updmin : float, optional
Minimum update fraction for regularization. Default is 0.2.
sb : bool, optional
Add bias term to the distribution (not used in uniform penalty). Default is False.
reg_zc : bool, optional
Enable zero-crossing (ZC) penalty. Default is True.
zc_max : int, optional
(Deprecated) Used in previous ZC penalty calculations. Default is 1.
zc_down : int, optional
Iteration after which ZC penalty update starts to decrease. Default is 59.
zc_upd : float, optional
Initial update fraction for ZC penalty. Default is 1.
zc_updmin : float, optional
Minimum update fraction for ZC penalty. Default is 0.2.
zc_downrate : float, optional
Rate of decrease for ZC update fraction after `zc_down`. Default is 0.05.
zc_on : int, optional
Iteration index when ZC regularization is turned on. Default is 4.
zc_nmax : int, optional
Window size for moving max in ZC penalty coefficient calculation. Default is 3.
dim_ndim : ndarray, optional
Custom sampling vector for regularized non-inverted dimensions. Default is None.
tau : ndarray or list of ndarray, optional
Independent variable values for each dimension. Default is None.
conv_limit : float, optional
Convergence tolerance for iterative update. Default is 1e-6.
Cz : list of ndarray, optional
Zero-crossing term of the regularization matrix. Used for resuming or inspecting inversion. Default is [np.array([0])].
store_g : bool, optional
If True, the ``g``(distribution or spectrum) is stored in ``IltData.g_list`` after each iteration.
store_gamma : bool, optional
If True, the ``Gamma``(regularization matrix) is stored in ``IltData.gamma_list`` after each iteration.
g_list : bool, optional
If ``store_g`` is True, the ``g``(distribution or spectrum) is stored in ``IltData.g_list`` after each iteration.
gamma_list : bool, optional
If ``store_gamma`` is True, the ``Gamma``(regularization matrix) is stored in ``IltData.gamma_list`` after each iteration.
force_sparse : bool, optional
If True, all operations for alt_g = 0 will use matrices in sparse format irrespective of the sparsity of kernel matrices
sparse_threshold : float, optional
Defines the minimum sparsity level required for kernel matrices to be treated as sparse in case of alt_g = 0.
For example, a value of 0.9 means that a matrix must have more than 90% zero entries to be considered sparse.