Recent changes to Prediction

Prediction modified by André Offringa

André Offringa — Fri, 01 Dec 2017 11:17:32 -0000

--- v19
+++ v20
@@ -30,17 +30,17 @@
 WSClean will also not look at the frequency of the FITS files when you make use of the [wide-band modes](WidebandDeconvolution) and create multiple model images. E.g. when you image with:

     :::bash
-    wsclean -channelsout 4 [...] ms-at-100MHz.ms ms-at-110MHz.ms ms-at-120MHz.ms ms-at-130MHz.ms
+    wsclean -channels-out 4 [...] ms-at-100MHz.ms ms-at-110MHz.ms ms-at-120MHz.ms ms-at-130MHz.ms

 and predict those four models into only one MS with:

     :::bash
-    wsclean -channelsout 4 -predict [...] ms-at-100MHz.ms
+    wsclean -channels-out 4 -predict [...] ms-at-100MHz.ms

 this will split the channels in the 100MHz MS into four groups and predict the four channel images into these groups. So one will get different results in this MS compared to what one would get with:

     :::bash
-    wsclean -channelsout 4 -predict [...] ms-at-100MHz.ms ms-at-110MHz.ms ms-at-120MHz.ms ms-at-130MHz.ms
+    wsclean -channels-out 4 -predict [...] ms-at-100MHz.ms ms-at-110MHz.ms ms-at-120MHz.ms ms-at-130MHz.ms

 Which will split the full available bandwidth into four groups and thus predict the first channel image into the first MS and so on.

Prediction modified by André Offringa

André Offringa — Fri, 19 May 2017 10:39:27 -0000

--- v18
+++ v19
@@ -15,7 +15,7 @@
 Prediction is supported since [WSClean version 1.2](Changelog-1.2). Since [WSClean 2.1](Changelog-2.1), it is no longer necessary to manually specify the image dimensions and pixel size. It is still allowed though, in which case the specified dimensions are checked against the image dimensions.

 ###Frequency information during prediction
-Some care should be taken when predicting at a different frequency than the model image, or from image cubes made with WSClean (described in [making image cubes](MakingImageCubes)), because this works slightly different than CASA. At this point, WSClean does not look at the frequency of the FITS file, and therefore won't interpolate or extrapolate the model to the right frequency. For example, if an image is made with
+Some care should be taken when predicting at a different frequency than the model image or when predicting from image cubes made with WSClean (described in [making image cubes](MakingImageCubes)), because this works slightly different than CASA. At this point, WSClean does not look at the frequency of the FITS file, and therefore won't interpolate or extrapolate the model to the right frequency. For example, if an image is made with

     :::bash
     wsclean [...] ms-at-100MHz.ms
@@ -27,7 +27,7 @@

 the model is not extrapolated to the right frequency, so one will end up with the flux levels of the 100 MHz model. 

-It will also not look at the frequency of the FITS files when you make use the [wide-band modes](WidebandDeconvolution) and create multiple model images. E.g. when you image with:
+WSClean will also not look at the frequency of the FITS files when you make use of the [wide-band modes](WidebandDeconvolution) and create multiple model images. E.g. when you image with:

     :::bash
     wsclean -channelsout 4 [...] ms-at-100MHz.ms ms-at-110MHz.ms ms-at-120MHz.ms ms-at-130MHz.ms
@@ -37,7 +37,7 @@
     :::bash
     wsclean -channelsout 4 -predict [...] ms-at-100MHz.ms

-this will split the channels in the 100MHz MS into four groups and predict the four channel images into these groups. So you will get different results in this MS than you would get with:
+this will split the channels in the 100MHz MS into four groups and predict the four channel images into these groups. So one will get different results in this MS compared to what one would get with:

     :::bash
     wsclean -channelsout 4 -predict [...] ms-at-100MHz.ms ms-at-110MHz.ms ms-at-120MHz.ms ms-at-130MHz.ms

Prediction modified by André Offringa

André Offringa — Fri, 19 May 2017 09:22:50 -0000

--- v17
+++ v18
@@ -12,7 +12,7 @@

 The normal use-case of using prediction is for self-calibration or subtraction. For these cases one should use a "model" image, which contains the clean components. If the input image is in absolute flux, it might be necessary to first apply the beam to the model image.

-Prediction is supported since [WSClean version 1.2](Changelog-1.2). Since [WSClean 2.1](Changelog-2.1), it is no longer necessary to manually specify the image dimensions and pixel size. It still allowed though, in which case the specified dimensions are checked against the image dimensions.
+Prediction is supported since [WSClean version 1.2](Changelog-1.2). Since [WSClean 2.1](Changelog-2.1), it is no longer necessary to manually specify the image dimensions and pixel size. It is still allowed though, in which case the specified dimensions are checked against the image dimensions.

 ###Frequency information during prediction
 Some care should be taken when predicting at a different frequency than the model image, or from image cubes made with WSClean (described in [making image cubes](MakingImageCubes)), because this works slightly different than CASA. At this point, WSClean does not look at the frequency of the FITS file, and therefore won't interpolate or extrapolate the model to the right frequency. For example, if an image is made with

Prediction modified by André Offringa

André Offringa — Fri, 19 May 2017 09:22:00 -0000

--- v16
+++ v17
@@ -1,6 +1,6 @@
 WSClean can be used to fill the `MODEL_DATA` column with the visibilities corresponding to an image. This is called 'predicting' visibilities (CASA's corresponding task is 'ft' -- sometimes this is referred to as 'degridding').

-Predicting can be performed by adding '-predict' to the command line. At this point, the the input image needs to be in the exact same projection as that WSClean would output it. If your image is in a different projection, you will have to regrid it first.
+Predicting can be performed by adding '-predict' to the command line. For this to work, the input image needs to be in the exact same projection as that WSClean would output it. If your image is in a different projection, you will have to regrid it first.

 An example prediction run:

Prediction modified by André Offringa

André Offringa — Fri, 19 May 2017 07:17:24 -0000

--- v15
+++ v16
@@ -5,7 +5,7 @@
 An example prediction run:

     :::bash
-    # Predict visibilities from a 3072x3072 Stokes I image with 0.72 arcmin res.
+    # Predict visibilities from a Stokes I image
     wsclean -predict -name my-image obs.ms

 Be aware that the specified name (here '`-name my-image`') still specifies the "prefix" of the file names in the same way as that it is normally used for imaging. That means that in this run, prediction will look for an image called 'my-image-model.fits'. If you predict for multiple polarizations, e.g. with "`-pol xx,xy,yx,yy`", the files should be named accordingly, so the XX name is "my-image-XX-model.fits", etc. The same applies when predicting with multiple frequency intervals with '-channelsout'.

Prediction modified by André Offringa

André Offringa — Sat, 21 Jan 2017 10:03:42 -0000

--- v14
+++ v15
@@ -53,3 +53,5 @@
 The input are absolute Stokes images and the output are apparent flux images with linear polarizations. The 'uncorrect' parameter specifies it should do the opposite of its normal operation, as it would normally make Stokes images out of wsclean's output. For example, if the <stokes-prefix> is "stokes", then pbcorrect will look for stokes-I.fits, stokes-Q.fits, stokes-U.fits and stokes-V.fits. If any of these is not present, it will be assumed zero (and a warning is issued). The 'beam' files are 8 files containing all real/imaginary components for the four linear polarizations. These can be created with the 'beam' tool in my MWA repository.

 If you use pbcorrect to prepare an image for wsclean prediction, you should set 'image-postfix' to "model.fits", and the image prefix is the same prefix you will specify to wsclean.
+
+**Next chapter: [WSClean and self-cal](WSCleanAndSelfcal)**

Prediction modified by André Offringa

André Offringa — Tue, 20 Dec 2016 08:12:54 -0000

--- v13
+++ v14
@@ -1,18 +1,18 @@
 WSClean can be used to fill the `MODEL_DATA` column with the visibilities corresponding to an image. This is called 'predicting' visibilities (CASA's corresponding task is 'ft' -- sometimes this is referred to as 'degridding').

-Predicting can be performed by adding '-predict' to the command line. At this point, the dimensions and scale of the input image(s) have to be set manually, just as in a normal imaging run, and the input image needs to be in the exact same projection as that WSClean would output it. If your image is in a different projection, you will have to regrid it first.
+Predicting can be performed by adding '-predict' to the command line. At this point, the the input image needs to be in the exact same projection as that WSClean would output it. If your image is in a different projection, you will have to regrid it first.

 An example prediction run:

     :::bash
     # Predict visibilities from a 3072x3072 Stokes I image with 0.72 arcmin res.
-    wsclean -predict -size 3072 3072 -scale 0.72amin -name my-image obs.ms
+    wsclean -predict -name my-image obs.ms

 Be aware that the specified name (here '`-name my-image`') still specifies the "prefix" of the file names in the same way as that it is normally used for imaging. That means that in this run, prediction will look for an image called 'my-image-model.fits'. If you predict for multiple polarizations, e.g. with "`-pol xx,xy,yx,yy`", the files should be named accordingly, so the XX name is "my-image-XX-model.fits", etc. The same applies when predicting with multiple frequency intervals with '-channelsout'.

 The normal use-case of using prediction is for self-calibration or subtraction. For these cases one should use a "model" image, which contains the clean components. If the input image is in absolute flux, it might be necessary to first apply the beam to the model image.

-Prediction is supported since [WSClean version 1.2](Changelog-1.2).
+Prediction is supported since [WSClean version 1.2](Changelog-1.2). Since [WSClean 2.1](Changelog-2.1), it is no longer necessary to manually specify the image dimensions and pixel size. It still allowed though, in which case the specified dimensions are checked against the image dimensions.

 ###Frequency information during prediction
 Some care should be taken when predicting at a different frequency than the model image, or from image cubes made with WSClean (described in [making image cubes](MakingImageCubes)), because this works slightly different than CASA. At this point, WSClean does not look at the frequency of the FITS file, and therefore won't interpolate or extrapolate the model to the right frequency. For example, if an image is made with

Prediction modified by André Offringa

André Offringa — Tue, 29 Mar 2016 07:33:34 -0000

--- v12
+++ v13
@@ -1,4 +1,4 @@
-Since [version 1.2](Changelog-1.2), WSClean can be used for filling the MODEL_DATA column with the visibilities corresponding to an image. This is called 'predicting' visibilities (CASA's corresponding task is 'ft' -- sometimes this is referred to as 'degridding').
+WSClean can be used to fill the `MODEL_DATA` column with the visibilities corresponding to an image. This is called 'predicting' visibilities (CASA's corresponding task is 'ft' -- sometimes this is referred to as 'degridding').

 Predicting can be performed by adding '-predict' to the command line. At this point, the dimensions and scale of the input image(s) have to be set manually, just as in a normal imaging run, and the input image needs to be in the exact same projection as that WSClean would output it. If your image is in a different projection, you will have to regrid it first.

@@ -8,9 +8,11 @@
     # Predict visibilities from a 3072x3072 Stokes I image with 0.72 arcmin res.
     wsclean -predict -size 3072 3072 -scale 0.72amin -name my-image obs.ms

-Be aware that the '-name my-image' still specifies the "prefix" of the file names in the same way as that it is normally used for imaging. That means that in this run, prediction will look for an image called 'my-image-model.fits'. If you predict for multiple polarizations, e.g. with "-pol xx,xy,yx,yy", the files should be named accordingly, so like "my-image-XX-model.fits", etc., and similarly when you are predicting with multiple frequency intervals with '-channelsout'.
+Be aware that the specified name (here '`-name my-image`') still specifies the "prefix" of the file names in the same way as that it is normally used for imaging. That means that in this run, prediction will look for an image called 'my-image-model.fits'. If you predict for multiple polarizations, e.g. with "`-pol xx,xy,yx,yy`", the files should be named accordingly, so the XX name is "my-image-XX-model.fits", etc. The same applies when predicting with multiple frequency intervals with '-channelsout'.

-The normal use-case of using prediction is for self-calibration or subtraction. For these cases you should use a "model" image, which contains the clean components. If your image is in absolute flux, you probably need to apply the beam on the model image first.
+The normal use-case of using prediction is for self-calibration or subtraction. For these cases one should use a "model" image, which contains the clean components. If the input image is in absolute flux, it might be necessary to first apply the beam to the model image.
+
+Prediction is supported since [WSClean version 1.2](Changelog-1.2).

 ###Frequency information during prediction
 Some care should be taken when predicting at a different frequency than the model image, or from image cubes made with WSClean (described in [making image cubes](MakingImageCubes)), because this works slightly different than CASA. At this point, WSClean does not look at the frequency of the FITS file, and therefore won't interpolate or extrapolate the model to the right frequency. For example, if an image is made with
@@ -18,7 +20,7 @@
     :::bash
     wsclean [...] ms-at-100MHz.ms

-and the generated model is than predicted with
+and the generated model is then predicted with

     :::bash
     wsclean -predict [...] ms-at-200MHz.ms

Prediction modified by Andreas Horneffer

Andreas Horneffer — Fri, 05 Feb 2016 09:54:55 -0000

--- v11
+++ v12
@@ -40,6 +40,8 @@
     :::bash
     wsclean -channelsout 4 -predict [...] ms-at-100MHz.ms ms-at-110MHz.ms ms-at-120MHz.ms ms-at-130MHz.ms

+Which will split the full available bandwidth into four groups and thus predict the first channel image into the first MS and so on.
+
 ###Some MWA specifics
 Applying the beam to an MWA image is a bit tricky, because the feeds are not orthogonal for anything but zenith. You can use the 'pbcorrect' tool (in my MWA repository) to apply a beam to an image. The basic syntax is:

Prediction modified by Andreas Horneffer

Andreas Horneffer — Fri, 05 Feb 2016 09:51:20 -0000

--- v10
+++ v11
@@ -23,12 +23,22 @@
     :::bash
     wsclean -predict [...] ms-at-200MHz.ms

-the model is not extrapolated to the right frequency, so one will end up with the fluxe levels of the 100 MHz model. Similarly, if you make use the [wide-band modes](WidebandDeconvolution) and create multiple model images, e.g. with:
+the model is not extrapolated to the right frequency, so one will end up with the flux levels of the 100 MHz model. 
+
+It will also not look at the frequency of the FITS files when you make use the [wide-band modes](WidebandDeconvolution) and create multiple model images. E.g. when you image with:

     :::bash
     wsclean -channelsout 4 [...] ms-at-100MHz.ms ms-at-110MHz.ms ms-at-120MHz.ms ms-at-130MHz.ms

-Predicting those four models back at a higher frequency will not extrapolate those, but will rather predict the first model into the first MS, etc. Prediction performs the complementary action of imaging.
+and predict those four models into only one MS with:
+
+    :::bash
+    wsclean -channelsout 4 -predict [...] ms-at-100MHz.ms
+
+this will split the channels in the 100MHz MS into four groups and predict the four channel images into these groups. So you will get different results in this MS than you would get with:
+
+    :::bash
+    wsclean -channelsout 4 -predict [...] ms-at-100MHz.ms ms-at-110MHz.ms ms-at-120MHz.ms ms-at-130MHz.ms

 ###Some MWA specifics
 Applying the beam to an MWA image is a bit tricky, because the feeds are not orthogonal for anything but zenith. You can use the 'pbcorrect' tool (in my MWA repository) to apply a beam to an image. The basic syntax is: