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.

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. At the same time rather than computing an effective wavelength (which was more a mean wavelength), we compute the pivot wavelength, which is more appropriate when converting between Fν and Fλ.

When computing the attenuation curve, the bump is now added so that its relative strength does not depend on δ.

For more reliability we now save the chi² by blocks. Otherwise it could lead to some crashes when there was more than 1 redshift.

Ensure that we do not try to fit bands that have an error bar of 0 mJy. We now simply discard such bands.

Change the way the Maraston models are built in order to capture the rapid evolutionary phases, similarly to the changes done for the BC03 models.

In order to capture short evolutionary phases we assume that over 1 Myr star formation occurs in 10 small episodes every 0.1 Myr, rather than 1 large episode every 1 Myr and nothing in-between.

Add the option to build the database with the high resolution BC03 models rather than the low resolution ones.

The computation of the fit taking into the upper limits would only work for the first few models because the number of bands was taken as the number of models.

The standardisation of error bars must be done on the final estimates, not on the estimates for a single block.

Use a much more readable formula to compute the uncertainties. We use the fact that Var(X) = E(Var(X)) + Var(E(X)). At the same time, broadcast the weights array to the same shape as the other arrays in order to simplify the syntax for the computation.

Do not correct intensive physical properties for the difference in redshift between the observation and the grid.

Read the list of sed modules from pcigale.ini. This is more flexible than inferring this list from the input parameters file as some modules can be missing (for instance when using the default values for all the parameters of a given module).

Compute the best fit at the at the exact distance of the analysed object and not at the distance of the rounded redshift. This eliminates the need for a correction factor.

Give the redshift of the model rather than the list of redshifts to correct for the redshift difference between the model and the observation.

Add a results manager to store the results of the analysis (bayesian and best fit estimates, best chi², etc.). Several such managers can be merged to combine appropriately the analysis of the observations in different blocks. This is the base class. The implementation will come in subsequent patches.

More correct block size computation to ensure that all the models are computed and that each block contains all the redshifts for a given model.

Save the model id rather than the index within the block as this would give an incorrect id when using more than one block.

Switch savefluxes to using a models manager and make all the necessary adjustments with the workers.