Configuration
JenpyROQ reads INI files with six logical sections:
[I/O]
[Parallel]
[Waveform_and_parametrisation]
[ROQ]
[Training_range]
[Test_values]
The implementation defaults below are the values assigned in
JenpyROQ/initialise.py before the config file is read.
I/O Section
Option |
Default |
Meaning |
|---|---|---|
|
|
Run directory where logs, copied config, ROQ arrays and plots are written. |
|
|
Enables stream logging when non-zero. |
|
|
Activates additional basis-matrix inversion checks and creates
|
|
|
Logs wall-clock timings for pre-selection, enrichment and validation steps. |
|
|
Calls |
|
|
Skips basis construction, loads an existing ROQ from |
|
|
Initial seed for pseudo-random training-set generation. |
Parallel Section
Option |
Default |
Meaning |
|---|---|---|
|
|
|
|
|
Number of worker processes requested by the selected pool. |
When parallel is non-zero, n-processes must be at least 2. The code
does not infer this value from mpiexec; keep the config and launch command
aligned.
Waveform & Parametrisation Section
Option |
Default |
Meaning |
|---|---|---|
|
|
Waveform family. Available values depend on installed optional libraries. |
|
|
Selects active spin degrees of freedom: |
|
|
Adds |
|
|
Adds |
|
|
Adds |
|
|
Adds |
|
|
Uses chirp mass and mass ratio as the sampled mass coordinates. |
|
|
Uses total mass and mass ratio as the sampled mass coordinates. |
|
|
Uses spherical spin coordinates. This is accepted only with
|
|
|
Lower edge of the construction frequency grid in Hz. |
|
|
Upper edge of the construction frequency grid in Hz. |
|
|
Segment length in seconds. The frequency spacing is |
mc-q-par and m-q-par are mutually exclusive. If both are set to
0, the code samples m1 and m2 directly.
Frequency Grid Size
The frequency-grid size is controlled by f-min, f-max and seglen.
The frequency spacing is:
The code constructs the grid with:
np.arange(f_min, f_max + deltaF, deltaF)
If (f_max - f_min) / deltaF is an integer, the number of grid points is:
For example, with f-min = 20, f-max = 2048 and seglen = 128:
so the grid has:
frequency samples. This count directly affects memory use, basis construction time and the maximum possible ROQ speedup.
ROQ Section
Option |
Default |
Meaning |
|---|---|---|
|
|
When enabled, orthogonalises new basis elements against the known basis. |
|
|
Builds the linear waveform basis for |
|
|
Builds the quadratic basis for |
|
|
Starting basis strategy. Supported values are |
|
|
Projection-error threshold for the linear pre-selection loop. |
|
|
Projection-error threshold for the quadratic pre-selection loop. |
|
|
Maximum number of linear pre-selection elements, including the two
initial |
|
|
Maximum number of quadratic pre-selection elements, including the two
initial |
|
|
Number of random candidates evaluated when searching for a new pre-selection basis element. |
|
|
Number of enrichment cycles. |
|
|
Comma-separated training-set size for each enrichment cycle. |
|
|
Number of outliers allowed to remain at the end of each cycle. |
|
|
Relative factors multiplying |
|
|
Final interpolation-error scale for the linear basis. |
|
|
Final interpolation-error scale for the quadratic basis. |
|
|
Number of random validation points after construction. |
|
|
Minimum allowed ratio |
The three list options training-set-sizes, training-set-n-outliers and
training-set-rel-tol must have length equal to n-training-set-cycles.
The default training-set-rel-tol currently contains five entries while the
default cycle count is four; production configs should set these lists
explicitly. For example:
n-training-set-cycles = 3
training-set-sizes = 10000,100000,100000
training-set-n-outliers = 0,0,0
training-set-rel-tol = 0.1,1.0,1.0
Training Range Syntax
The training range section uses one -min and one -max key per active
parameter:
[Training_range]
mc-min = 1.1968
mc-max = 1.1988
q-min = 1.0
q-max = 3.0
lambda1-min = 5.0
lambda1-max = 5000.0
The test-values section sets one representative point used in the waveform comparison plots:
[Test_values]
mc = 1.1975
q = 1.1
lambda1 = 200
If a test value is outside its training range, the code logs a warning but does
not abort. If an upper bound is smaller than its lower bound, the run raises
ValueError.
Parameter Activation
Parameter names |
Active when |
Default range |
|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
magnitudes |
|
|
|
|
|
|
|
|
|
|
|
|
Mass Coordinates
When mc-q-par = 1, sampled mc and q are converted to component
masses by:
When m-q-par = 1, sampled total mass m and mass ratio q are
converted by: