Closed aagnello closed 10 years ago
I would not be above running a simple peak finder to initialize both the positions and magnitudes of the nebula components. In fact I'd say this was fairly essential, given what we've seen so far in the surprisingly difficult H1413. @davidwhogg shall we take this on tomorrow?
On Tue, Apr 22, 2014 at 3:37 PM, Adriano Agnello notifications@github.comwrote:
Related to #20 https://github.com/davidwhogg/LensTractor/issues/20 and
26 https://github.com/davidwhogg/LensTractor/issues/26
The total flux can be guessed from the photocal. But what about the best initial guess for fluxes of galaxy and pt.sources? Hint: the model, with positions initialised somehow, predicts a composite SB at each point, represented by an array of the kind: model = fgal*SBgal + Sum[fpt[i]
SBpt[i], i in range(sources)] The fgal and fpt normalisations are linear parameters and ought to be solved for. Then we take max(fgal,0) and max (fpt,0) as the guesses to be fed to the driver. However, how do we exploit the flux information from photocal? Another piece in the likelihood? *Further(!) development: use a peak finder to position the sources around the galaxy?
— Reply to this email directly or view it on GitHubhttps://github.com/davidwhogg/LensTractor/issues/27 .
That’d be nice, but before that it wouldn’t be bad if someone figured out why -l -s crashes completely, at least when applied to the sdss examples… Judging from the error messages, it seems that it doesn’t store the parameter values and likelihood as it should. I’ll pull on the laptop and see what happens on the ps1 example, but an expert look may help way more than my conjectures. Best Adri
On Apr 22, 2014, at 6:41 PM, Phil Marshall notifications@github.com wrote:
I would not be above running a simple peak finder to initialize both the positions and magnitudes of the nebula components. In fact I'd say this was fairly essential, given what we've seen so far in the surprisingly difficult H1413. @davidwhogg shall we take this on tomorrow?
On Tue, Apr 22, 2014 at 3:37 PM, Adriano Agnello notifications@github.comwrote:
Related to #20 https://github.com/davidwhogg/LensTractor/issues/20 and
26 https://github.com/davidwhogg/LensTractor/issues/26
The total flux can be guessed from the photocal. But what about the best initial guess for fluxes of galaxy and pt.sources? Hint: the model, with positions initialised somehow, predicts a composite SB at each point, represented by an array of the kind: model = fgal*SBgal + Sum[fpt[i]
SBpt[i], i in range(sources)] The fgal and fpt normalisations are linear parameters and ought to be solved for. Then we take max(fgal,0) and max (fpt,0) as the guesses to be fed to the driver. However, how do we exploit the flux information from photocal? Another piece in the likelihood? *Further(!) development: use a peak finder to position the sources around the galaxy?
— Reply to this email directly or view it on GitHubhttps://github.com/davidwhogg/LensTractor/issues/27 .
— Reply to this email directly or view it on GitHub.
We did not get to the magnitude initialisation issue today, other than to recognise it and flag it up in the code.
On Tue, Apr 22, 2014 at 8:40 PM, Adriano Agnello notifications@github.comwrote:
That’d be nice, but before that it wouldn’t be bad if someone figured out why -l -s crashes completely, at least when applied to the sdss examples… Judging from the error messages, it seems that it doesn’t store the parameter values and likelihood as it should. I’ll pull on the laptop and see what happens on the ps1 example, but an expert look may help way more than my conjectures. Best Adri
On Apr 22, 2014, at 6:41 PM, Phil Marshall notifications@github.com wrote:
I would not be above running a simple peak finder to initialize both the positions and magnitudes of the nebula components. In fact I'd say this was fairly essential, given what we've seen so far in the surprisingly difficult H1413. @davidwhogg shall we take this on tomorrow?
On Tue, Apr 22, 2014 at 3:37 PM, Adriano Agnello notifications@github.comwrote:
Related to #20 https://github.com/davidwhogg/LensTractor/issues/20 and
26 https://github.com/davidwhogg/LensTractor/issues/26
The total flux can be guessed from the photocal. But what about the best initial guess for fluxes of galaxy and pt.sources? Hint: the model, with positions initialised somehow, predicts a composite SB at each point, represented by an array of the kind: model = fgal*SBgal + Sum[fpt[i]
SBpt[i], i in range(sources)] The fgal and fpt normalisations are linear parameters and ought to be solved for. Then we take max(fgal,0) and max (fpt,0) as the guesses to be fed to the driver. However, how do we exploit the flux information from photocal? Another piece in the likelihood? *Further(!) development: use a peak finder to position the sources around the galaxy?
— Reply to this email directly or view it on GitHub< https://github.com/davidwhogg/LensTractor/issues/27> .
— Reply to this email directly or view it on GitHub.
— Reply to this email directly or view it on GitHubhttps://github.com/davidwhogg/LensTractor/issues/27#issuecomment-41121588 .
Magnitude initialisation is now refactored. Closing this until this becomes an issue again (I don't think it will for a while!)
Related to #20 and #26 The total flux can be guessed from the photocal. But what about the best initial guess for fluxes of galaxy and pt.sources? Hint: the model, with positions initialised somehow, predicts a composite SB at each point, represented by an array of the kind: model = fgal_SBgal + Sum[fpt[i]_SBpt[i], i in range(sources)] The fgal and fpt normalisations are linear parameters and ought to be solved for. Then we take max(fgal,0) and max (fpt,0) as the guesses to be fed to the driver. However, how do we exploit the flux information from photocal? Another piece in the likelihood? Further(!) development: use a peak finder to position the sources around the galaxy?