CHANGELOG.md 37.5 KB
Newer Older
1 2
# Change Log

3 4
## Unreleased
### Added
5
- It is now possible to optionally indicate the distance in Mpc in the input file. If present it will be used in lieu of the distance computed from the redshift. This is especially useful in the nearby universe where the redshift is a very poor indicator of the actual distance. (Médéric Boquien)
6
- It is now possible to fit any physical property indicated by the code (e.g. equivalent width, dust luminosity, etc.). For this the physical property needs to be given in the input file and the properties to be fitted must be given in the properties filed in pcigale.ini. (Héctor Salas & Médéric Boquien)
7

8
### Changed
9 10
- The `sfhdelayed` module has been extended to optionally include an exponential burst to model the latest episode of star formation. (Médéric Boquien & Barbara Lo Faro)

11
### Fixed
12
- The histogram bin width was not computed optimally when some models were invalid. (David Corre & Médéric Boquien)
13 14
- Missing import in the `m2005` module. (Médéric Boquien, reported by Dominika Wylezalek)

15
### Optimised
16
- The estimation of the physical properties is made a bit faster when all the models are valid. (Médéric Boquien)
17

18
## 0.12.1 (2018-02-27)
19
### Fixed
20
- The best fit could not be computed in photo-z mode because the redshift was negative. (Médéric Boquien)
21
- The bayesian estimates could not be computed when some models were older than the age of the universe. (Médéric Boquien)
22
- The usage of `dustatt\_cazleit` causes some confusion regarding the reddening of the stars and of the gas. We have clarified that they are both attenuated with the same law and switched the differential reddening to 1 by default. (Médéric Boquien & Véronique Buat)
23
- When some models were invalid, it was not possible to plot the PDF. (Médéric Boquien & Denis Burgarella)
24

25

26
## 0.12.0 (2018-02-19)
27
### Added
28
- Provide the possibility not to store a given module in cache. This can be useful on computers with a limited amount of memory. The downside is that when not caching the model generation will be slower. (Médéric Boquien)
29
- An option `redshift\_decimals` is now provided in `pdf\_analysis` to indicate the number of decimals to round the observed redshifts to compute the grid of models. By default the model redshifts are rounded to two decimals but this can be insufficient at low z and/or when using narrow-band filters for instance. This only applies to the grid. The physical properties are still computed for the redshift at full precision. (Médéric Boquien)
30
- Bands with negative fluxes are now considered valid and are fitted as any other band. (Médéric Boquien)
31
- Allow the models to be computed by blocks in `savefluxes`. This can be useful when computing a very large grid and/or to split the results file into various smaller files as large files can be difficult to handle. The number of blocks is set with the `blocks` parameters in the pcigale.ini. (Médéric Boquien)
32
- Allow the observations to be analysed by blocks of models in `pdf\_analysis`. This is useful when computing a very large grid of models that would not fit in memory. The number of blocks is set with the `blocks` parameters in the pcigale.ini. (Médéric Boquien)
33
- The integrated stellar luminosity is now provided as `stellar.lum`. (Médéric Boquien)
34
- The high resolution BC03 models have been added. They can be activated when building the database by adding `--bc03res=hr` to the build command. In that case the low resolution models are not built. (Médéric Boquien)
35
- Dust templates generated with THEMIS (Jones et al. 2017) have been contributed by the DustPedia team (Davies et al. 2017). Special acknowledgement to Angelos Nersesian and Frédéric Galliano for creating the dust templates and writing the code. (Dustpedia team)
36
- The Herschel SPIRE filters for extended sources have been added. (Médéric Boquien)
37

38
### Changed
39
- Make the timestamp more readable when moving the out/ directory. (Médéric Boquien)
40 41
- To accommodate the analysis of the observations by blocks, all models are now included in the estimation of the physical properties and no cut in chi² is done anymore. (Médéric Boquien)
- To accommodate the analysis of the observations by blocks, the `save_pdf` option has been eliminated. To plot PDF one needs to set `save_chi2` to True and then run `pcigale-plots pdf`. (Médéric Boquien)
42
- In order to capture rapid evolutionary phases, we assume that in a given period of 1 Myr, 10 small episodes of star formation occurred every 0.1 Myr, rather than one episode every 1 Myr.
43
- When computing the attenuation curve, the bump is now added so that its relative strength does not depend on δ. (Médéric Boquien, issue reported by Samir Salim)
44
- Fν was computed by calculating Fλ and then converting to Fν, which led to typical differences in fluxes of typically 1-2% and a bit more for a handful of pathological filters. Now Fν is computed directly and a bit faster. (Médéric Boquien, issue reported by Yannick Roehlly and Wouter Dobbels)
45

46
### Fixed
47
- Corrected a typo that prevented `restframe\_parameters` from being listed among the available modules. (Médéric Boquien)
48
- The filters in the residual plot of `pcigale-plots sed` are now drawn in order of increasing wavelength so that the line joining all the filters does not make loops. (Médéric Boquien)
49
- In the absence of a nebular component `restframe\_parameters` would crash when attempting to compute the equivalent widths of the lines listed in `EW_lines`. Now they are simply ignored. (Médéric Boquien)
50
- The luminosity spectrum of the best fit was saved assuming the distance corresponding to the redshift rounded to two decimals. This was an issue in particular at very low redshift as a difference of 0.005 in redshift can translate to a large difference on the luminosity distance. Now the exact luminosity distance of the object is used to compute the spectrum luminosity. (Médéric Boquien, reported by Jorge Melnick)
51
- When using the `parameters\_file` option, the indices of the models now correspond to the line number of the input file. (Médéric Boquien)
52
- When using the `parameters\_file` option, the list of modules is read from `sed\_modules` rather than being inferred from the input file. (Médéric Boquien)
53
- The computation of the upper limits would only work for the first few models, reverting back to regular fits for the others. (Médéric Boquien)
54
- A more explicit message is now given when the flux table cannot be read properly. (Médéric Boquien)
55
- Make sure that we do not try to fit data that have an error bar of 0 mJy. (Médéric Boquien)
56
- An erroneous warning was displayed when using the `restframe\_parameters` module. (Médéric Boquien)
57
- The formula from Sawicki et al. (2012) used to compute the χ² in the presence of upper limits was not correct. This led the χ² to depend directly on the absolute value of the upper limit. The formula has been rederived and corrected. (Médéric Boquien & Denis Burgarella)
58
- For some reason the wavelengths of the SCUBA 450 μm filter were a factor 10 too small. (Médéric Boquien)
59
- Ensure that the computation of the continuum level is correct when determining the equivalent width, in particular when the line width is very narrow. (Médéric Boquien)
60
- When using different line widths during a single run, ensure that the fluxes and other quantities are always computed correctly. (Médéric Boquien, special thanks to Genoveva Micheva)
61
- Compute Dn4000 more rigorously by integrating properly over Fν. (Médéric Boquien)
62

63
### Optimised
64
- The cache architecture has been simplified, making it somewhat faster. It speeds up the model generation by ~1%. (Médéric Boquien)
65

66
## 0.11.0 (2017-02-10)
67
### Added
68
- The stellar mass-weighted age is now provided. This is a much more usual measure of the age than the age of the oldest star. This is accessible with the `stellar.age_m_star` keyword in the `bc03` module with with the `stellar.age_mass` keyword in the `m2005` module. (Médéric Boquien)
69 70
- The nebular models have been expanded from log U=-3 to log U=-4. (Médéric Boquien & Akio Inoue)
- The nebular models are now sampled in steps of 0.1 dex in log U rather than 1.0 dex steps. (Médéric Boquien & Akio Inoue)
Médéric Boquien's avatar
Médéric Boquien committed
71
- A new set of filters from GAZPAR has been added. The pattern of the filter name is "telescope.instrument.filter", e.g. "hst.wfc3.F160W". If the telescope has one instrument, it is skipped, e.g. "galex.FUV". For now the original set of filters is still provided. (Médéric Boquien & Olivier Ilbert)
72
- A brand new module `restframe\_parameters` has been added to compute rest frame parameters: UV slope β (Calzetti et al. 1994), Dn4000 (Balogh et al. 1999), IRX, emission lines equivalent widths at any wavelength, luminosity in any filter, colour in any pair of filters. This module has to be inserted right before the redshifting module. (Médéric Boquien)
73

74
### Changed
75
- We do not output the break strength from the `bc03` module anymore as these were not computed properly. (Médéric Boquien)
76
- The new `restframe\_parameters` module replaces the unofficial `param` module, which has now been trimmed to only compute fluxes in the observed frame as the rest of its functionalities have been transferred to the much more efficient `restframe\_parameters` module. To reflect this, it has been renamed `fluxes`. (Médéric Boquien)
77
- We now make use of new features available in Python 3.5. Previous versions are henceforth unsupported. However Python 3.6 or later is recommended for better performance. (Médéric Boquien)
78

79
### Fixed
80
- When the pcigale.ini file was missing, pcigale would crash and display a fairly cryptic backtrace. Now it explicitly states that the file could not be found. (Médéric Boquien)
81
- The nebular emission now takes into account deviations from the 10000K case B assumption. In practice this yields fluxes about 10% fainter. (Médéric Boquien & Akio Inoue)
82
- Some filters were incorrectly assumed to be defined in units of energy when they were actually defined in units of photons, yielding slightly incorrect fluxes. Now all the filters are converted into units of energy when imported. (Médéric Boquien)
83
- Remove the PSEUDO_D4000 filter which was incorrect. (Médéric Boquien)
84
- Indicate the correct transmission type for the PACS green and red filters. (Médéric Boquien)
85
- IRAS filters are defined in energy rather than photons. (Médéric Boquien)
86
- Remove a Numpy warning in the computation of the IGM absorption (Médéric Boquien)
87
- The Herschel passbands only included the filter. Now they include the full throughput of the instrument. Flux differences are expected to be no more than 1%. (Médéric Boquien)
88
- The `dustatt_powerlaw` module indicated that the Milky Way bump has an amplitude of 3. This is only valid for the `dustatt_calzleit` module. As `dustatt_powerlaw` is normalised to A(V) rather than E(B-V) for `dustatt_calzleit`, the bump is a factor Rv larger. A more reasonable value is now given. (Médéric Boquien)
89
- The correction of the χ² for the upper limits now properly takes into account earlier changes intended to reduce memory usage and speed up the analysis. This prevents a crash. (Médéric Boquien)
90
- Fix a crash in the computation of the rectangular periodic SFH with Numpy 1.12. (Médéric Boquien)
91

92
### Optimised
93
- By default the MKL library created many threads for each for the parallel processes. Not only was this not necessary as a high-level parallelisation already exists, but it generated a strong oversubscription on the CPU and on the RAM. The slowdown was over a factor of ~2 in some cases. Now we mandate KML to use only 1 thread fo each process. (Médéric Boquien & Yannick Roehlly)
94
- The generic numpy interpolation function was used. As we are in a well-controlled environment, this generated unnecessary verifications on the type and shape of the arrays. The compiled numpy interpolation function is now used, bypassing those checks. This generates a gain of 5-10% in computing speed for the generation of the models. (Médéric Boquien)
95
- The interpolation of the spectra of the different physical components on a new wavelength grid was not optimal as the interpolation was done separately for each component. Now a specific function has been implemented caching repetitive computations and vectorising the interpolation to compute it for all the components in a single step. This generates a gain of 10% in computing speed for the generation of the models. (Médéric Boquien)
96
- An optimisation in the sorting of a 2D array led to a gain of 10% in computing speed for the generation of the models. (Médéric Boquien)
97 98

## 0.10.0 (2016-09-15)
99
### Added
100 101
- Enable the possibility of having different normalisation factors for the star formation history. (Médéric Boquien)

102
### Changed
103
- Various descriptions have been improved and clarified. (Médéric Boquien)
104
- The `output\_file` and `output\_format` parameters have been removed from the `savefluxes` module. They served little purpose and made the code more complex. The same strategy as for the `pdf\_analysis` modules is now adopted, saving the output both as FITS and ASCII tables. Old configuration file still work, with these two parameters simply ignored. (Médéric Boquien)
105

106
### Fixed
107
- With the new sanity check of the input parameters, cigale did not handle the fact that the redshift could be given in the parameters file. Now this is handled properly. (Médéric Boquien)
108
- When giving the list of parameters through a file, cigale did not compute properly what parameter changed between to successive models. (Médéric Boquien)
109
- Using the m2005 module led to a crash. This is now fixed. (Yannick Roehlly)
110
- When computing models on a grid, the order could change from one run to the next, which is an issue when comparing the outputs of `savefluxes` for instance. Now models are always computed in the same order. The last parameter of the last module being in innermost loop and the first parameter of the first module being the outermost loop. (Médéric Boquien)
111

112
### Optimised
113
- A significant fraction of the total run time is spent computing integrals (e.g. fluxes in passbands). We can make the integration faster by rewriting the trapezoidal rule in terms of np.dot(). This allows to offload the computation to optimised libraries. The end result is that the integration is twice as fast, with a gain of ~10-15% on the total run time. (Médéric Boquien)
114
- The conversion from luminosity to flux is now a bit faster. (Médéric Boquien)
115
- The order the models are stored in memory has been changed to make the computation of the χ² faster. (Médéric Boquien)
116

117
## 0.9.0 (2016-04-04)
118
### Added
119 120
- When using the `savefluxes` module, all the output parameters were saved. This is not efficient when the user is only interested in some of the output parameters but not all. We introduce the "variables" configuration parameter for `savefluxes` to list the output parameters the user wants to save. If the list is left empty, all parameters are saved, preserving the current behaviour. This should increase the speed substantially when saving memory. (Médéric Boquien)
- Similarly to the `savefluxes` module, in the `pdf_analysis` module if the list of physical properties is left empty, all physical parameters are now analysed. (Médéric Boquien)
121
- It is now possible to pass the parameters of the models to be computed from a file rather than having to indicate them in pcigale.ini. This means that the models do not necessarily need to be computed on a systematic grid of parameters. The name of the file is passed as an argument to the parameters\_file keyword in pcigale.ini. If this is done, the creation\_modules argument is ignored. Finally, the file must be formatted as following: each row is a different model and each column a different parameter. They must follow the naming scheme: module\_name.parameter\_name, that is "bc03.imf" for instance. (Médéric Boquien)
122
- Addition of the `schreiber2016` SED creation module implementing the Schreiber et al. (2016) dust models. (Laure Ciesla)
123
- The physical parameters provided in pcigale.ini were not checked at startup against what the modules could accept. This could lead to a runtime crash if an unexpected value was passed to the module. Now the parameters are checked at startup. If an issue is found, it is indicated and the user is asked to fix it before launching cigale again. The validation file is build at the same time as pcigale.ini. (Médéric Boquien)
124
- There is a new attenuation module (`dustatt_2powerlaws`) based on a double power law: a birth cloud power law applied only to the young star population and an ISM power law applied to both the young and the old star population. (Yannick Roehlly & Véronique Buat)
125

126
### Changed
127
- The estimates of the physical parameters from the analysis of the PDF and from the best fit were recorded in separate files. This can be bothersome when trying to compare quantities from different files. Rather, we generate a single file containing all quantities. The ones estimated from the analysis of the PDF are prefixed with "bayes" and the ones from the best fit with "best". (Médéric Boquien)
128
- To homogenize input and output files, the "observation\_id" has been changed to "id" in the output files. (Médéric Boquien)
129
- The output files providing estimates of the physical properties are now generated both in the form of text and FITS files. (Médéric Boquien)
130
- When using the `dustatt_calzleit module`, choosing ẟ≠0 leads to an effective E(B-V) different from the one set by the user. Now the E(B-V) will always correspond to the one specified by the user. This means that at fixed E(B-V), A(V) depends on ẟ. (Médéric Boquien)
131
- The pcigale-mock tool has been merged into pcigale-plots; the mock plots can be obtained with the "mock" command. (Médéric Boquien)
132 133 134
- The `sfhdelayed` module is now initialised with _init_code() to be consistent with the way things are done in other modules. This should give a slight speedup under some sircumstances too. (Médéric Boquien)
- In `sfhfromfile`, the specification of the time grid was vague and therefore could lead to incorrect results if it was not properly formatted by the end user. The description has been clarified and we now check that the time starts from 0 and that the time step is always 1 Myr. If it is not the case we raise an exception. (Médéric Boquien)
- When the redshift is not indicated in pcigale.ini, the analysis module fills the list of redshifts from the redshifts indicated in the input flux file. This is inefficient as analysis modules should not have to modify the configuration. Now this is done when interpreting pcigale.ini before calling the relevant analysis module. As a side effect, "pigale check" now returns the total number of models that cigale will compute rather than the number of models per redshift bin. (Médéric Boquien)
135
- The optionally saved spectra in the `pdf_analysis` and `savefluxes` modules were saved in the VO-table format. The most important downside is that it is very slow to write to, which proves to be a major bottleneck in the computing speed. To overcome this issue, we rather save the spectra using the FITS formation. Instead of having one file containing the spectra (including the various components) and the SFH in a single file, now we have one file for the spectra and one file for the SFH.
136

137
### Fixed
138
- To estimate parameters in log, pcigale determines which variables end with the "\_log" string and removed it to compute the models. However in some circumstances, it was overzealous. This has been fixed. (Médéric Boquien)
139
- When estimating a parameter in log, these were not scaled appropriately and taken in log when saving the related χ² and PDF. (Médéric Boquien)
140
- In the presence of upper limits, correct the scaling factor of the models to the observations before computing the χ², not after. (Médéric Boquien)
141
- When called without arguments, pcigale-plots would crash and display the backtrace. Now it displays the a short help showing how to use it. (Médéric Boquien)
142
- For `sfh2exp`, when setting the scale of the SFH with sfr0, the normalisation was incorrect by a factor exp(-1/tau\_main). (Médéric Boquien)
143
- The mass-dependent physical properties are computed assuming the redshift of the model. However because we round the observed redshifts to two decimals, there can be a difference of 0.005 in redshift between the models and the actual observation if CIGALE computes the list of redshifts itself. At low redshift, this can cause a discrepancy in the mass-dependent physical properties: ~0.35 dex at z=0.010 vs 0.015 for instance. Therefore we now evaluate these physical quantities at the observed redshift at full precision. (Médéric Boquien, issue reported by Samir Salim)
144 145
- In the `sfhfromfile` module, an extraneous offset in the column index made that it took the previous column as the SFR rather than the selected column. (Médéric Boquien)
- In `sfhfromfile`, if the SFR is made of integers cigale crashed. Now we systematically convert it to float. (Médéric Boquien)
146
- The order of the parameters for the analysis modules would change each time a new pcigale.ini was generated. Now the order is fixed. (Médéric Boquien)
147
- In the output the sfh.age parameter would correspond to the input value minus 1. Now both values are consistent with one another. (Laure Ciesla & Médéric Boquien)
148
- In rare circumstances requiring a specific distribution of redshifts the integration of the spectrum in some filters was not done correctly, inducing relative errors of ~10¯⁵-10¯⁶. (Médéric Boquien)
149
- The absorption of the Lyman continuum from old stars tended to be overestimated leading to some “negative fluxes” for the Lyman continuum. (Médéric Boquien)
150

151
### Optimised
152
- Prior to version 0.7.0, we needed to maintain the list of redshifts for all the computed models. Past 0.7.0 we just infer the redshift from a list unique redshifts. This means that we can now discard the list of redshifts for all the models and only keep the list of unique redshifts. This saves ~8 MB of memory for every 10⁶ models. the models should be computed slightly faster but it is in the measurement noise. (Médéric Boquien)
153
- The `sfhfromfile` module is now fully initialised when it is instantiated rather than doing so when processing the SED. This should be especially sensitive when processing different SED. (Médéric Boquien)
154
- We do not store the time grid in the SED anymore given that we assume it starts at 0 Myr with steps of 1 Myr, we can easily reconstruct to save it if needed. It should save a little bit of memory and it should go a little bit faster. (Médéric Boquien)
155
- To compute the stellar spectrum of the young component, do not pass the full SFH with the old part set to 0. Rather, only pass the corresponding part of the SFH. This nearly doubles the computing speed of the stellar spectrum (Médéric Boquien)
156 157 158
- Computers are much better at multiplying than at dividing. Therefore to correct the spectral emission when redshifting we multiply by 1/(1+z) rather than dividing by 1+z. (Médéric Boquien)

---
159

160 161
## 0.8.1 (2015-12-07)
### Fixed
162
-  To estimate parameters in log, pcigale determines which variables end with the "\_log" string and removed it to compute the models. However in some circumstances, it was overzealous. This has been fixed. (Médéric Boquien)
163 164

## 0.8.0 (2015-12-01)
165
### Added
166
- The evaluation of the parameters is always done linearly. This can be a problem when estimating the SFR or the stellar mass for instance as it is usual to estimate their log rather. Because the log is non-linear, the likelihood-weighted mean of the log is not the log of the likelihood-weighted mean. Therefore the estimation of the log of these parameters has to be done during the analysis step. This is now possible. The variables to be analysed in log just need to be indicated with the suffix "_log", for instance "stellar.m\_star\_log". (Médéric Boquien, idea suggested by Samir Salim)
167

168
### Fixed
169 170
- Running the scripts in parallel trigger a deadlock on OS X with python 3.5. A workaround has been implemented. (Médéric Boquien)
- When no dust emission module is used, pcigale genconf complains that no dust attenuation module is used. Correctly specify dust emission and not attenuation. (Médéric Boquien and Laure Ciesla)
171
- Allowing more flexibility to read ASCII files broke the handling of FITS files. It is now fixed. (Yannick Roehlly)
172

173 174 175
### Changed
- The attenuation.ebvs\_main and attenuation.ebvs\_old parameters are no longer present as they were duplicates of attenuation.E\_BVs.stellar.old and attenuation.E\_BVs.stellar.young (that are still available).

176 177
---

178
## 0.7.0 (2015-11-19)
179
### Added
180
- The pcigale-mock utility has been added to generate plots comparing the exact and pcigale-estimated parameters. This requires pcigale to be run beforehand with the `pdf_analysis` module and the mock\_flag option set to True. (Denis Burgarella and Médéric Boquien)
181 182 183
- The pcigale-filter utility has been added to easily list, plot, add, and remove filters without having the rebuild the database entirely. (Médéric Boquien)
- It is now possible to analyse the flux in a band as a regular parameter. It can be useful for flux predictions. (Yannick Roehlly)
- The redshift can be a now used as a free parameter, enabling pcigale to estimate the photometric redshift. (Médéric Boquien)
Médéric Boquien's avatar
Médéric Boquien committed
184
- When running "pcigale genconf", the list of modules is automatically checked against the list of official modules. If modules are missing, information is printed on the screen indicating the level of severity (information, warning, or error) and the list of modules that can be used. (Médéric Boquien)
185 186

### Changed
187
- The galaxy\_mass parameter was very ambiguous. In reality it corresponds to the integral of the SFH. Consequently it has been renamed sfh.integrated. (Médéric Boquien)
188
- In the Calzetti attenuation module, add a warning saying that is the power law slope is different than 0, E(B-V) will no longer be the real one. (Yannick Roehlly)
189
- Add "B\_B90" to the list of computed attenuation so that users can calculate the effective E(B-V). (Yannick Roehlly)
190 191
- Computing the parameters and their uncertainties through the histogram of the PDF is slow and can introduce biases in some cases. Rather, now the estimated values of the parameters and the corresponding uncertainties are simply computed from the weighted mean and standard deviation of the models that are at least 0.1% as likely as the best model to reproduce the observations. The differences in the estimates are very small except when very few models are used. (Médéric Boquien)
- Magic values to indicate invalid values (e.g. values lower than -99) are difficult to handle safely. They have been replaced with NaN wherever appropriate. The logic of the input flux file stays the same for the time being but the magic values are converted internally after reading it. Users are advised to replace magic values with NaN. The output files now use NaN instead of magic number to indicate invalid values. (Médéric Boquien)
192
- Rename the the AGN faction added by dale2014 module from agn.fracAGN to agn.fracAGN\_dale2014 to avoid conflict with fritz2006 module. (Yannick Roehlly)
193
- Remove the radio component from the dale2014 model so that it can be used with the more flexible radio module, courtesy Daniel Dale. (Laure Ciesla and Médéric Boquien)
194 195 196

### Fixed
- The SFH is modelled using discrete star formation episodes every 1 Myr. This means that as the SFH is not really continuous (the input single stellar population do not allow us to compute that properly), we should not integrate SFH(t), but simply sum SFH(t) as t is discrete. In most cases the difference is very small. The only case where it makes a difference is for a very rapidly varying SFH, for instance taking τ=1 Myr. (Médéric Boquien)
197
- Ensure that the flux can be computed even if the redshifting module has not been applied. By default in that case we assume a distance of 10 parsecs. While in practice it should never happen as the redshifting module is mandatory, this can be more important when using pcigale as a library. (Médéric Boquien and Yannick Roehlly)
198 199 200 201 202 203 204
- When called without arguments, pcigale would crash. Now it displays a brief message to remind how it should be invoked. (Médéric Boquien)
- Raise an exception instead of crash when an unknown IMF is requested. (Médéric Boquien)
- When the error column for a flux was present in the observation table but not in the used column list (when the user prefers to use a default error instead of the real one), the error on the flux was set to 0. (Yannick Roehlly)
- For some reason a point in the GALEX FUV filter has a negative transmission. That should not happen. After comparison with the filter on the GALEX website it has been set to 0. (Médéric Boquien)
- Shorten the left and right 0 parts of the pseudo D4000 filter so that it can be applied on smaller spectra. (Yannick Roehlly)
- To compute the reduced χ², we need to divide by the number of bands-1 (and not the number of bands). We do that because we consider that the models depend on one meta-parameter. (Médéric Boquien)
- The test to determine when to take into account upper limits did not work according the specifications. Now upper limits are always taken into account when they should. (Médéric Boquien)
Médéric Boquien's avatar
Médéric Boquien committed
205
- The nebular emission could be counted in excess in the dust luminosity as both lines and the Lyman continuum could be attenuated. Now we do not extend the attenuation under 91 nm. Also, a new component as been added, taking specifically the Lyman continuum absorption by gas, allowing to conserve the information about the intrinsic stellar Lyman continuum if need be. (Yannick Roehlly and Médéric Boquien)
Médéric Boquien's avatar
Médéric Boquien committed
206
- The Flambda table in the VO-table export does not reflect the fact that it stores luminosity densities. Accordingly, it has been renamed Llambda. (Yannick Roehlly and Médéric Boquien)
207
- When the flux file contains a mix of spaces and tabulations as column separators, pcigale discards the header and takes the first data line as the header. Now pcigale properly handles such a combination. Bug reported by Paola Santini. (Médéric Boquien)
208 209 210 211 212 213 214 215 216

### Optimised
- Major speedup to build the database by inserting multiple models in the database at once rather one model at a time. On an SSD, the total run time of "python setup.py build" goes from 5m20s to 2m42s. The speedup should be even more spectacular on a rotating hard drive. (Médéric Boquien)
- Memory usage reduction using in-place operations (e.g. a/=2 rather than a=a/2, saving the creation of a temporary array the size of a) where possible. (Médéric Boquien)
- The creation and handling of mock catalogues has been streamlined. (Médéric Boquien)
- Slight speedup using np.full() where possible to create an array with all elements set to the same value. (Médéric Boquien)
- Computing the scaling factors and the χ² in one step over the entire grid of models is very memory-intensive, leading to out-of-memory issues when using multiple cores. Rather, let's compute them band by band, as this avoids the creation of large temporary arrays, while keeping the computation fast. (Médéric Boquien).
- Each core copied the subset of models corresponding to the redshift of the object to be analysed. This is a problem as it can strongly increase memory usage with the number of cores, especially when there are many models and just one redshift. Rather than making a copy, we use a view, which not only saves a considerable amount of memory but is also faster as there is no need to allocate new, large arrays. This is made possible as models are regularly ordered with redshift. (Médéric Boquien)
- Various minor optimisations. (Médéric Boquien)
217

218
---
219

220
## 0.6.0 (2015-09-07)
221 222 223 224 225 226 227 228 229 230
### Added
- New module to compute a star formation history as described in Buat et al. 2008. (Yannick Roehlly)
- New module to compute a periodic SFH. Each star formation episode can be exponential, "delayed", or rectangular. (Médéric Boquien and Denis Burgarella)
- New module performing a quench on the star formation history. (Yannick Roehlly)
- New module to compute the emission of a modified black body. (Denis Burgarella)
- New module to compute the physical properties measured on the emission spectrum (e.g. spectral indices, ultraviolet slope, etc.). (Denis Burgarella)
- New pseudo filters to compute line fluxes and spectral indices. (Yannick Roehlly)
- The dust masses are now computed for the draine2007 and draine2014 modules. (Médéric Boquien)

### Changed
231
- The nebular\_lines\_width parameter of the nebular module is now called lines\_width as the nebular prefix was redundant. (Médéric Boquien)
232 233 234 235 236
- Prefix the ouput variables of SFH-related modules with "sfh" to facilitate their identification in the output files. (Médéric Boquien)
- Prefix the output variables of the fritz2006 AGN module with "agn" to facilitate their identification in the output files. (Médéric Boquien)
- Prefix the redshift with "universe". (Médéric Boquien)
- With pcigale-plot, draw the spectra only to λ=50 cm as the models do not extend much further and there is very rarely any observation beyond λ=21 cm. (Médéric Boquien)
- As pcigale is getting much faster, display the number of computed models every 250 models rather than every 100 models. (Médéric Boquien)
237
- Give default values for the dl2014, sfh\_buat, and sfhdelayed modules to allow for quick test runs. (Médéric Boquien)
238 239 240 241 242 243 244 245 246
- Now pcigale-plots plots errors on upper limits. (Denis Burgarella)

### Fixed
- When plotting, round the redshift to two decimals to match the redshift of the model. (Médéric Boquien)
- Ensure that all the input parameters of the nebular module are also output so it is possible to analyse them. (Médéric Boquien)
- Properly take the Lyman continuum photons absorbed by dust into account to compute the dust emission. (Médéric Boquien)
- Improve the readability of the pcigale-plots generated spectra by drawing the observed fluxes on top of other lines. (Médéric Boquien)
- The nebular components are now plotted with pcigale-plots. (Médéric Boquien)
- When a filter that was not in the database was called, pcigale would crash ungracefully as the exception invoked does not exist in Python 3.x. Now use a Python 3.x exception. (Médéric Boquien)
247
- The dustatt\_powerlaw module could not identify which physical components to attenuate when the nebular module was called. (Médéric Boquien)
248 249 250 251 252 253 254 255 256 257 258 259 260 261 262
- The displayed counter for the number of objects already analysed could be slightly offset from the number of models actually computed. (Médéric Boquien)
- Change the method to compute the χ² in the presence of upper limits as the previous method did not always converge. (Denis Burgarella and Médéric Boquien)

### Optimised
- When plotting, do not recompute the luminosity distance, which is very slow, but rather get the one computed during the analysis and that is given in the output files. (Médéric Boquien)
- Adding new physical components with a wavelength sampling different than that of the pre-existing grid is slow as a common grid has to be computed and all components interpolated over it. The nebular lines are especially bad for that, owing to the fact that each line is sampled with 19 points. This is excessive as sampling over 9 points barely changes the fluxes while speeding up the computation of the models by ~20%. (Médéric Boquien)
- Rather than resampling the filter transmission on the wavelength grid every time the flux is computed in a given filter, put the resampled filter into a cache. (Médéric Boquien)
- Rather than recomputing every time the merged wavelength grid from two different wavelength grids (for instance when adding a new physical component or when integrating the spectrum in a filter), put the results in a cache. (Médéric Boquien)
- Before adding a new component to a SED, we first copy the original SED without that component from the cache. This copy can be very slow when done automatically by python. We rather do this copy manually, which is much faster. (Médéric Boquien)
- When adding a new physical component with a different wavelength sampling, rather than reinterpolating all the components over the new grid, compute the interpolation only for new wavelengths. (Médéric Boquien)
- Various minor optimisations. (Médéric Boquien)
- When computing the flux in filters, np.trapz() becomes a bottleneck of the code. A large part of the time is actually spent on safeguards and on operations for nD arrays. However here we only have 1D arrays and some variables can be cached, which allows some optimisations to compute fluxes faster. (Médéric Boquien)
- The output parameters of a model were stored in an ordered dictionary. While convenient to keep the order of insertion it is very slow as it is implemented in pure Python for versions up to 3.4. Rather we use a regular dictionary and we reorder the parameters alphabetically. (Médéric Boquien)
- To store the SED in memory and retrieve them later, we index them with the list of parameters used to compute them. We serialise those using JSON. However JSON is slow. As these data are purely internal, rather use marshal, which is much faster than JSON. (Médéric Boquien)

263
---
264

265
## 0.5.1 (2015-04-28)
266
### Changed
267
- Set the default dale2014 AGN fraction to 0 to avoid the accidental inclusion of AGN. (Denis Burgarella)
268
- Modify the name of the averaged SFR: two averaged SFRs over 10 (sfh.sfr10Myrs) and 100Myrs (sfh.sfr100Myrs). (Denis Burgarella)
269
- Improve the documentation of the `savefluxes` module. (Denis Burgarella)
270 271 272

### Fixed
- Correction of the x-axis limits. (Denis Burgarella)
273
- Fix the detection of the presence of the agn.fritz2006\_therm in pcigale-plots. (Denis Burgarella)
274 275 276
- Correct the wavelength in the SCUBA 450 μm filter. (Denis Burgarella)
- Install the ancillary data required to make plots. (Yannick Roehlly)

277
## 0.5.0 (2015-04-02)
278

279 280
---

281
## 0.4.0 (2014-10-09)
282

283 284
---

285
## 0.3.0 (2014-07-06)
286

287 288
---

289
## 0.2.0 (2014-06-10)
290

291 292
---

293
## 0.1.0 (2014-05-26)