pyimcom.routine

Numba version of pyimcom_croutines.c.

Slightly slower than C, used when furry-parakeet is not installed.

Functions

iD5512C_getw

Interpolation code written by Python.

iD5512C

2D, 10x10 kernel interpolation for high accuracy.

iD5512C_sym

iD5512C for symmetrical output.

gridD5512C

iD5512C for rectangular grid.

lakernel1

PyIMCOM linear algebra kernel (eigendecomposition).

lsolve_sps

Routine to solve Ax=b.

build_reduced_T_wrap

Makes coaddition matrix T from a reduced space.

Functions

iD5512C_getw(→ None)	Interpolation code written by Python.
iD5512C(→ None)	2D, 10x10 kernel interpolation for high accuracy
iD5512C_sym(→ None)	2D, 10x10 kernel interpolation for high accuracy
gridD5512C(→ None)	2D, 10x10 kernel interpolation for high accuracy
lakernel1(→ None)	PyIMCOM linear algebra kernel (eigendecomposition).
lsolve_sps(→ None)	Routine to solve Ax=b.
build_reduced_T_wrap(→ None)	Intermediate quantities to build coaddition matrix T from a reduced space.

Module Contents

iD5512C_getw(w: numpy.array, fh: float) → None

Interpolation code written by Python.

Parameters:

• w (np.array) – Array, shape (10,); 1D interpolation weights will be written here.

• fh (float) – ‘xfh’ and ‘yfh’ with 1/2 subtracted.

Return type:

None

iD5512C(infunc: numpy.array, xpos: numpy.array, ypos: numpy.array, fhatout: numpy.array) → None

2D, 10x10 kernel interpolation for high accuracy

Can interpolate multiple functions at a time from the same grid so we don’t have to keep recomputing weights.

Parameters:

• infunc (np.array) – Input function on some grid. Shape (nlayer, ngy, ngx).

• xpos (np.array) – Input x values. Shape (nout,).

• ypos (np.array) – Input y values. Shape (nout,).

• fhatout (np.array) – Location to put the output values. Shape : (nlayer, nout).

Return type:

None

iD5512C_sym(infunc: numpy.array, xpos: numpy.array, ypos: numpy.array, fhatout: numpy.array) → None

2D, 10x10 kernel interpolation for high accuracy

Can interpolate multiple functions at a time from the same grid so we don’t have to keep recomputing weights.

This version assumes the output is symmetrical as a sqrt nout x sqrt nout matrix

Parameters:

• infunc (np.array) – Input function on some grid. Shape (nlayer, ngy, ngx).

• xpos (np.array) – Input x values. Shape (nout,).

• ypos (np.array) – Input y values. Shape (nout,).

• fhatout (np.array) – Location to put the output values. Shape (nlayer, nout).

Return type:

None

gridD5512C(infunc: numpy.array, xpos: numpy.array, ypos: numpy.array, fhatout: numpy.array) → None

2D, 10x10 kernel interpolation for high accuracy

this version works with output points on a rectangular grid so that the same weights in x and y can be used for many output points

Parameters:

• infunc (np.array) – Input function on some grid. Shape (ngy, ngx).

• xpos (np.array) – Input x values. Shape (npi, nxo).

• ypos (np.array) – Input y values. Shape (npi, nyo).

• fhatout (np.array) – Location to put the output values. Shape (npi, nyo*nxo).

Return type:

None

Notes

There are npi*nyo*nxo interpolations to be done in total, but for each input pixel npi, there is an nyo x nxo grid of output points. If you want a 3D array (npi, nyo, nxo), you can reshape fhatout after it is filled.

lakernel1(lam: numpy.array, Q: numpy.array, mPhalf: numpy.array, C: float, targetleak: float, kCmin: float, kCmax: float, nbis: int, kappa: numpy.array, Sigma: numpy.array, UC: numpy.array, T: numpy.array, smax: float) → None

PyIMCOM linear algebra kernel (eigendecomposition).

Performs all the steps with the for loops (i.e., except the matrix diagonalization). In the parameter descriptions, “n” refers to the number of input pixels and “m” to the number of output pixels.

Parameters:

• lam (np.array) – System matrix eigenvalues. Shape (n,).

• Q (np.array) – System matrix eigenvectors. Shape (n, n). (No longer used in this version of the algorithm.)

• mPhalf (np.array) – -P/2 = premultiplied target overlap matrix. Shape (m, n).

• C (float) – Target normalization.

• targetleak (float) – Allowable leakage of target PSF.

• kCmin (float, float) – Range of kappa/C to test.

• kCmax (float, float) – Range of kappa/C to test.

• nbis (int) – Number of bisections.

• kappa (np.array) – Lagrange multiplier per output pixel. Shape (m,). Writes to this array.

• Sigma (np.array) – Output noise amplification. Shape (m,). Writes to this array.

• UC (np.array) – Fractional squared error in PSF. Shape (m,). Writes to this array.

• T (np.array) – Coaddition matrix, needs to be multiplied by Q^T after the return. Shape (m,n). Writes to this array.

• smax (float) – Maximum allowed noise amplification Sigma.

Return type:

None

lsolve_sps(N: int, A: numpy.array, x: numpy.array, b: numpy.array) → None

Routine to solve Ax=b.

Only the lower triangle of A is ever used (the rest need not even be allocated). The matrix A is destroyed.

Parameters:

• N (int) – Number of ‘node’ eigenvalues.

• A (np.array, shape : (N, N)) – Positive definite matrix. Shape (N, N).

• x (np.array) – Vector, output. Shape (N,).

• b (np.array) – Vector. Shape (N,).

Return type:

None

build_reduced_T_wrap(Nflat: numpy.array, Dflat: numpy.array, Eflat: numpy.array, kappa: numpy.array, ucmin: float, smax: float, out_kappa: numpy.array, out_Sigma: numpy.array, out_UC: numpy.array, out_w: numpy.array) → None

Intermediate quantities to build coaddition matrix T from a reduced space.

In the parameter descriptions, “m” refers to the number of output pixels and “nv” to the number of kappa nodes.

Parameters:

• Nflat (np.array) – Input noise array. Shape (m*nv*nv,). Flattened version of (m,nv,nv).

• Dflat (np.array) – Input 1st order signal D/C. Shape (m*nv,). Flattened version of (m,nv).

• Eflat (np.array) – Input 2nd order signal E/C. Shape (m*nv*nv,). Flattened version of (m,nv,nv).

• kappa (np.array) – List of eigenvalues, must be sorted ascending. Shape (nv,).

• ucmin (float) – Min U/C.

• smax (float) – Max Sigma (noise).

• out_kappa (np.array) – Output “kappa” parameter. Shape (m,).

• out_Sigma (np.array) – Output “Sigma”. Shape (m,).

• out_UC (np.array) – Output “U/C”. Shape (m,).

• out_w (np.array) – Output weights for each eigenvalue and each output pixel. Shape (m*nv,). Flattened version of (m,nv).

Return type:

None