Skip to content

Commit 3e62f8e

Browse files
committed
Added section on differences with FAST
1 parent 91f9a71 commit 3e62f8e

File tree

1 file changed

+28
-0
lines changed

1 file changed

+28
-0
lines changed

README.md

Lines changed: 28 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -19,6 +19,11 @@
1919
- [Why is it faster?](#why-is-it-faster)
2020
- [Why does it use so little memory?](#why-does-it-use-so-little-memory)
2121
- [There must be a price to pay... did the code become more complex in C++?](#there-must-be-a-price-to-pay-did-the-code-become-more-complex-in-c)
22+
- [Differences with FAST](#differences-with-fast)
23+
- [Photometry](#photometry)
24+
- [Spectra](#spectra)
25+
- [Photo-z](#photo-z)
26+
- [Monte Carlo uncertainties](#monte-carlo-uncertainties)
2227
- [Additional features](#additional-features)
2328
- [Controlling output to the terminal](#controlling-output-to-the-terminal)
2429
- [Multithreading](#multithreading)
@@ -176,6 +181,29 @@ FAST++ is about 104 kB (including comments and empty lines), while FAST-IDL weig
176181
At similar code size, deciding if a code is more complex than another is a subjective question, especially if the languages are not the same. While C++ has suffered from a reputation of being an obscure language for many years, the situation has vastly improved with the 2011 version of the C++ standard, on which FAST++ relies greatly. These changes made C++ much more expressive and clear, and as always without compromising on performances. It is therefore my opinion that the code of FAST++ is *not* complex nor obscure, at least not more than that of FAST-IDL, but this is the opinion of one who has always programmed in C++.
177182

178183

184+
# Differences with FAST
185+
186+
While the implementation of FAST++ was designed to resemble FAST-IDL as much as possible, some aspects of FAST-IDL were not preserved. These are listed here.
187+
188+
## Photometry
189+
* The default "zero point" of the fluxes is 23.9 instead of 25. This means that your input fluxes are expected to be given in micro Janskys by default. You can of course specify your own "zero point" if this is not the case, as in FAST-IDL.
190+
* If a galaxy has no measured flux in a band (i.e., we have no information about it), the _uncertainty_ must be set to a negative value or "NaN". Note that there is a difference between "no measured flux" and "non-detection": the latter provides information that the fit can use. In FAST-IDL the _flux_ had to be set to -99, which was problematic because -99 could be a valid flux measurement (i.e., a non-detection).
191+
* The input photometry catalog can contain columns starting with "F" or "E" which are not flux columns, provided the "F" or "E" is not followed by numbers. In FAST-IDL a catalog containing such columns was rejected.
192+
193+
## Spectra
194+
* The template error function is now also applied to the spectra, as for the photometry. In FAST-IDL this was not the case.
195+
* Galaxies with a spectrum can still use the input photo-z. In FAST-IDL their photo-z was ignored.
196+
* An alternative format for the spectrum file is allowed (see [Better treatment of spectra](#better-treatment-of-spectra) below), but the old format is still supported.
197+
* The ```bin``` column is used in FAST++ to combine multiple spectral elements into one before the fit, using inverse variance weighting, which can speed up computations if the spectral resolution is very high. In FAST-IDL only the first element in a bin was used.
198+
199+
## Photo-z
200+
* If an input photo-z is negative, FAST++ will simply ignore the photo-z and let the redshift vary freely. In FAST-IDL a galaxy with a negative photo-z was ignored completely.
201+
* The way photometric redshifts are used in the fit is quite different by default, but it can be made to reproduce the original behavior of FAST-IDL. See [Photometric redshifts from EAzY](#photometric-redshifts-from-eazy) below.
202+
203+
## Monte Carlo uncertainties
204+
* The way FAST++ computes uncertainties using the Monte Carlo simulations is different from FAST-IDL, and this can lead to small differences in the resulting confidence intervals. This should not be significant.
205+
206+
179207
# Additional features
180208

181209
In addition to the base feature set provided by FAST-IDL, FAST++ has a number of additional feature, some of which you can enable in the parameter file. These features are all listed below.

0 commit comments

Comments
 (0)