Data splits in destriper #309
Conversation
I propagated the time and det splits to the `_compute_binned_map` function (and all the liked ones)
|
In this last commit, both binned and destriped maps out of the Currently, only the single-split pipeline is in place, thus the |
|
The loop over every split is implemented. It would be useful to give the possibility to "recycle" the baselines between splits. |
…n loading This may be modified by adding the splits to the actual .fits file
Now, the errors on baselines are not looped over detectors when saved, since they do not depend on them (they are summed over deterctors). The same is done for loading a `DestriperResult` from file. See also #310 for more details.
Fix bug happening when no destriping is perfomed, but report is requested
When a list of splits is passed, we assume the computation of the full split first. This allows to recycle its baselines
|
I added the recycling of baselines from the full splits. Apart from propagating the relevant attributes and changes, the main modification is that if baselines are passed to 'make_destriped_map' it does not iterate to find a new solution and instead uses the provided one. Also, I "stored" the info about the solution being recycled in the 'converged' attribute of the 'DestriperResult' object. Now, that can also be a string (not only a bool as before). This string says to the user that the baselines have been recycled from the full split and also reports the convergence of the original run. This might not be the best solution, let me know what you think eventually. A "side-effect" of this modification is that now the user can pass a custom baseline solution obtained externally. |
|
I added a notebook doing some examples of data splits, for Binner and Destriper. If you can, have a look at its flow and let me know if it is OK. After this, I think we can proceed with merging. |
|
Hi @ziotom78, I think this PR is mergeable if you agree. Some tests are failing because apparently TOAST is breaking on macos, but I doubt that depends on this PR... I discussed the interplay of this PR with #319 with @paganol and the best thing to do seems to be: we merge this into the master and then we merge again the master in #319 before starting working on the destriper. What do you think? |
Great! I agree, let's do this. |
This PR is meant to implement the same data splits of the
binner(see #291), but in thedestriper.make_binned_map_splits