Normalized Mean Model and PSF metrics

[jorgemarpa]

This PR adds two new methods:

1. A hidden function to find the scale factor that normalized the mean model to sum 1 when sources are full in the TPF. It creates a high-resolution mean model and integrates the PSF using the trapezoidal rule, the integral is the rescale factor that is applied to the PSF weights. The method also has a diagnostic plot.

2. A visible method to compute PSF metrics such as:

   • source_psf_fraction amount of PSF in data, useful to determine if a source has total or partial flux data.
   • perturbed_ratio_mean how much the perturbed model deviates from the mean model for a given source, useful to find when the time model changes the mean value of the light curve. The expected value is 1.
   • perturbed_std how much variability has the perturbed model in time, helps find when the time model introduces synthetic variability or features. A large value means a poor time model.

Requires #63 to be merged first

[jorgemarpa]

Sanity check, comparing SAP and PSF flux:
before (no normalization)

after (normalized shape model)

Scatter is due to saturated sources, sources with partial data (e.g. outside or near the edge of the TPF), and highly blended sources.

[christinahedges]

This looks very good! Let's add some tests that just check the metrics work and produce reasonable values? Especially the high resolution one would be good to test?

[christinahedges]

@jorgemarpa I reverted these changes because it broke in ways that were hard to debug quickly, you'll need to reopen this pull request!

[jorgemarpa]

This PR adds two new methods:

1. A hidden function to find the scale factor that normalized the mean model to sum 1 when sources are full in the TPF. It creates a high-resolution mean model and integrates the PSF using the trapezoidal rule, the integral is the rescale factor that is applied to the PSF weights. The method also has a diagnostic plot.

2. A visible method to compute PSF metrics such as:

   • source_psf_fraction amount of PSF in data, useful to determine if a source has total or partial flux data.
   • perturbed_ratio_mean how much the perturbed model deviates from the mean model for a given source, useful to find when the time model changes the mean value of the light curve. The expected value is 1.
   • perturbed_std how much variability has the perturbed model in time, helps find when the time model introduces synthetic variability or features. A large value means a poor time model.

Requires #63 to be merged first

[jorgemarpa]

Sanity check, comparing SAP and PSF flux:
before (no normalization)

after (normalized shape model)

Scatter is due to saturated sources, sources with partial data (e.g. outside or near the edge of the TPF), and highly blended sources.

[christinahedges]

This looks very good! Let's add some tests that just check the metrics work and produce reasonable values? Especially the high resolution one would be good to test?

[christinahedges]

@jorgemarpa I reverted these changes because it broke in ways that were hard to debug quickly, you'll need to reopen this pull request!

[jorgemarpa]

This PR adds two new methods:

1. A hidden function to find the scale factor that normalized the mean model to sum 1 when sources are full in the TPF. It creates a high-resolution mean model and integrates the PSF using the trapezoidal rule, the integral is the rescale factor that is applied to the PSF weights. The method also has a diagnostic plot.

2. A visible method to compute PSF metrics such as:

   • source_psf_fraction amount of PSF in data, useful to determine if a source has total or partial flux data.
   • perturbed_ratio_mean how much the perturbed model deviates from the mean model for a given source, useful to find when the time model changes the mean value of the light curve. The expected value is 1.
   • perturbed_std how much variability has the perturbed model in time, helps find when the time model introduces synthetic variability or features. A large value means a poor time model.

Requires #63 to be merged first

[jorgemarpa]

Sanity check, comparing SAP and PSF flux:
before (no normalization)

after (normalized shape model)

Scatter is due to saturated sources, sources with partial data (e.g. outside or near the edge of the TPF), and highly blended sources.

[christinahedges]

This looks very good! Let's add some tests that just check the metrics work and produce reasonable values? Especially the high resolution one would be good to test?

[christinahedges]

@jorgemarpa I reverted these changes because it broke in ways that were hard to debug quickly, you'll need to reopen this pull request!

[jorgemarpa]

This PR adds two new methods:

1. A hidden function to find the scale factor that normalized the mean model to sum 1 when sources are full in the TPF. It creates a high-resolution mean model and integrates the PSF using the trapezoidal rule, the integral is the rescale factor that is applied to the PSF weights. The method also has a diagnostic plot.

2. A visible method to compute PSF metrics such as:

   • source_psf_fraction amount of PSF in data, useful to determine if a source has total or partial flux data.
   • perturbed_ratio_mean how much the perturbed model deviates from the mean model for a given source, useful to find when the time model changes the mean value of the light curve. The expected value is 1.
   • perturbed_std how much variability has the perturbed model in time, helps find when the time model introduces synthetic variability or features. A large value means a poor time model.

Requires #63 to be merged first

[jorgemarpa]

Sanity check, comparing SAP and PSF flux:
before (no normalization)

after (normalized shape model)

Scatter is due to saturated sources, sources with partial data (e.g. outside or near the edge of the TPF), and highly blended sources.

[christinahedges]

This looks very good! Let's add some tests that just check the metrics work and produce reasonable values? Especially the high resolution one would be good to test?

[christinahedges]

@jorgemarpa I reverted these changes because it broke in ways that were hard to debug quickly, you'll need to reopen this pull request!

[jorgemarpa]

This PR adds two new methods:

1. A hidden function to find the scale factor that normalized the mean model to sum 1 when sources are full in the TPF. It creates a high-resolution mean model and integrates the PSF using the trapezoidal rule, the integral is the rescale factor that is applied to the PSF weights. The method also has a diagnostic plot.

2. A visible method to compute PSF metrics such as:

   • source_psf_fraction amount of PSF in data, useful to determine if a source has total or partial flux data.
   • perturbed_ratio_mean how much the perturbed model deviates from the mean model for a given source, useful to find when the time model changes the mean value of the light curve. The expected value is 1.
   • perturbed_std how much variability has the perturbed model in time, helps find when the time model introduces synthetic variability or features. A large value means a poor time model.

Requires #63 to be merged first

[jorgemarpa]

Sanity check, comparing SAP and PSF flux:
before (no normalization)

after (normalized shape model)

Scatter is due to saturated sources, sources with partial data (e.g. outside or near the edge of the TPF), and highly blended sources.

[christinahedges]

This looks very good! Let's add some tests that just check the metrics work and produce reasonable values? Especially the high resolution one would be good to test?

[christinahedges]

@jorgemarpa I reverted these changes because it broke in ways that were hard to debug quickly, you'll need to reopen this pull request!

[jorgemarpa]

This PR adds two new methods:

1. A hidden function to find the scale factor that normalized the mean model to sum 1 when sources are full in the TPF. It creates a high-resolution mean model and integrates the PSF using the trapezoidal rule, the integral is the rescale factor that is applied to the PSF weights. The method also has a diagnostic plot.

2. A visible method to compute PSF metrics such as:

   • source_psf_fraction amount of PSF in data, useful to determine if a source has total or partial flux data.
   • perturbed_ratio_mean how much the perturbed model deviates from the mean model for a given source, useful to find when the time model changes the mean value of the light curve. The expected value is 1.
   • perturbed_std how much variability has the perturbed model in time, helps find when the time model introduces synthetic variability or features. A large value means a poor time model.

Requires #63 to be merged first

[jorgemarpa]

Sanity check, comparing SAP and PSF flux:
before (no normalization)

after (normalized shape model)

Scatter is due to saturated sources, sources with partial data (e.g. outside or near the edge of the TPF), and highly blended sources.

[christinahedges]

This looks very good! Let's add some tests that just check the metrics work and produce reasonable values? Especially the high resolution one would be good to test?

[christinahedges]

@jorgemarpa I reverted these changes because it broke in ways that were hard to debug quickly, you'll need to reopen this pull request!

[jorgemarpa]

This PR adds two new methods:

1. A hidden function to find the scale factor that normalized the mean model to sum 1 when sources are full in the TPF. It creates a high-resolution mean model and integrates the PSF using the trapezoidal rule, the integral is the rescale factor that is applied to the PSF weights. The method also has a diagnostic plot.

2. A visible method to compute PSF metrics such as:

   • source_psf_fraction amount of PSF in data, useful to determine if a source has total or partial flux data.
   • perturbed_ratio_mean how much the perturbed model deviates from the mean model for a given source, useful to find when the time model changes the mean value of the light curve. The expected value is 1.
   • perturbed_std how much variability has the perturbed model in time, helps find when the time model introduces synthetic variability or features. A large value means a poor time model.

Requires #63 to be merged first

[jorgemarpa]

Sanity check, comparing SAP and PSF flux:
before (no normalization)

after (normalized shape model)

Scatter is due to saturated sources, sources with partial data (e.g. outside or near the edge of the TPF), and highly blended sources.

[christinahedges]

This looks very good! Let's add some tests that just check the metrics work and produce reasonable values? Especially the high resolution one would be good to test?

[christinahedges]

@jorgemarpa I reverted these changes because it broke in ways that were hard to debug quickly, you'll need to reopen this pull request!

[jorgemarpa]

This PR adds two new methods:

1. A hidden function to find the scale factor that normalized the mean model to sum 1 when sources are full in the TPF. It creates a high-resolution mean model and integrates the PSF using the trapezoidal rule, the integral is the rescale factor that is applied to the PSF weights. The method also has a diagnostic plot.

2. A visible method to compute PSF metrics such as:

   • source_psf_fraction amount of PSF in data, useful to determine if a source has total or partial flux data.
   • perturbed_ratio_mean how much the perturbed model deviates from the mean model for a given source, useful to find when the time model changes the mean value of the light curve. The expected value is 1.
   • perturbed_std how much variability has the perturbed model in time, helps find when the time model introduces synthetic variability or features. A large value means a poor time model.

Requires #63 to be merged first

[jorgemarpa]

Sanity check, comparing SAP and PSF flux:
before (no normalization)

after (normalized shape model)

Scatter is due to saturated sources, sources with partial data (e.g. outside or near the edge of the TPF), and highly blended sources.

[christinahedges]

This looks very good! Let's add some tests that just check the metrics work and produce reasonable values? Especially the high resolution one would be good to test?

[christinahedges]

@jorgemarpa I reverted these changes because it broke in ways that were hard to debug quickly, you'll need to reopen this pull request!