W1 option take-over into W2

[Juergen Reuter]

obsolete:

1. use_efficiency (efficiency fluctuates)
2. time_limit_adaptation (dangerous and machine dependent)
3. generate_phase_space (old development flag)
4. write_phase_space_channels_file
5. extra_off_shell_lines
6. splitting_depth (only single logs, not important)
7. single/double_off_shell_branchings/decays
8. massive_fsr
9. single/double_off_shell_branchings/decays
10. write_weights/write_weights_file
11. guess_color_flow
12. pythia_processes
13. time_limit (only for hard nosed batch systems)
14. warn_empty_channel (only for pathological cases)
15. show_pythia_initialization
16. show_pythia_statistics
17. screen_events, screen_histograms
18. screen_diagnostics
19. plot_grids_channels, plot_grids_logscale (impractical for large number of channels?)
20. keep best grid as fall back option like W1

[kilian]

OK, let's start working on this ...

[kilian]

accuracy_goal: done in r3658.

There are now three goals: accuracy_goal, error_goal, relative_error_goal. If all goals that are set have been reached in the adaptation pass, the rest of the adaptation pass is skipped and WHIZARD proceeds with the final integration pass.

The nontrivial part was to make this work correctly with iterations read from file. They are reused where possible, even if the goals are changed in a second run.

[Juergen Reuter]

Those are actually defects as features that have been damaged or crippled.

[Juergen Reuter]

One of the last remaining things from W1.

[Juergen Reuter]

Actually, I just had needed option 2 of the remaining tasks :(

[Juergen Reuter]

Well, I need them now(ish)

[Juergen Reuter]

After the upgrade from WHIZARD 2.1.x to 2.2.0 we have to recheck again that all these options are in :(

[Juergen Reuter]

W1 -> W2.1.x -> W2.2.x .... :(

[kilian]

Since this is already assigned, I might just do it now ...

[Juergen Reuter]

Cool. And please don't forget to re-check those features marked as implemtented here again.

[kilian]

r4981 covers

1. safety_factor, flag events w/weight > 1 (plot_excess)

Event weight excess in rejection (unweighting) is now always recorded and displayed at the end of the simulation pass. A safety_factor may be given to the simulate command to reduce the acceptance probability and thus reduce excess. Event excess can be separately plotted using record_excess. This offers more flexibility than the W1 version, where it was always plotted on top of the unweighted events. (Assuming that the drawing options will be improved in future versions, anyway.)

[kilian]

Partially done:

1. events_per_file, including starting number

Implemented for the eio_ascii set of formats. The global variables are sample_split_n_evt and {{sample_split_index}}}. Not yet implemented for other formats, but this should be straightforward. the relevant TBP is called eio_split_out.

Also done:

2. signal handling: save/restore vamp grid CONTENTS

Signals are blocked by default. At important points in code, the signal status is queried by signal_is_pending and, if necessary, the current routine returns. This works like a poor-man's exception mechanism. The actual termination points are after each SINDARIN command. (In the cascades module, I kept termination points as they were before.)

This way of treating signals allows to do some cleanup, such as properly closing event files, before the program aborts. In the previous implementation, there was only the possibility of immediate abort at various points.

The issue of snapshooting VAMP is nontrivial. I deferred this to #572.

There are no tests for signal handling; this is difficult. Practice should show whether I missed something.

[kilian]

event file splitting: covered LHEF in r4989.

Not yet covered: stdhep and HepMC.

[kilian]

1. show_* for debug output format

done in r4998. The user variables are ?debug_process for displaying the hard process instance, ?debug_decay for displaying the decay chain, and ?debug_verbose. A more fine-grained control seems unnecessary (for me), but may be added later.

[Juergen Reuter]

I rechecked the fragmentation/fragmentation_method. THis was mainly an opion for 1. no hadronization, 2. jetset with guessed color flows (which we don't need), PYTHIA hadronization (steered with the interface), 4. user defined hadronization (who wants this?) So this is indeed obsolete from our perspective now.

[kilian]

So, I remove fragmentation from the list. The issue behind it is covered in ticket #44.

[kilian]

1. plot_width, plot_height

This was already there, but called graph_width_mm and graph_height_mm. I expanded a unit test (commands_23) where this is actually checked.b

[kilian]

1. accuracy_goal

Implemented in r5003. (Also error_goal and relative_error_goal.)

[kilian]

1. keep best grid as fall back option like W1

After actually writing code for this, I came to the conclusion that it doesn't make sense.

If it matters, this is an indication that something's wrong in the integration. If it doesn't matter, we don't care. In any case, the last grid will be updated before being used in the final integration pass. And in this integration pass, the grids are not updated anymore in the default setup. This was different previously, and there was also the bug that rendered the reported error too small.

So, using the 'best grid' gives a false sense of security. Furthermore, it takes disk space, which might matter for some use cases. I'll push this to the obsolete list.

[kilian]

1. allow to ignore checksum test

r5006: implemented. The ignore flags are ?check_phs_file, ?check_grid_file, ?check_event_file. For PHS and events, the flags were already there.

[kilian]

1. events_per_file, including starting number, for some formats

r5009: also implemented for HepMC and STDHEP. I include checks for both, but I could verify only the HepMC test. The tests just check the screen messages that should indicate opening and closing of multiple files, assuming that the commands are actually executed. They don't check file contents.

Note also that, for all formats, the information about the expected number of events is not correct (I think they all insert zero). This doesn't change for split files.

With this addition in place, all issues in this tickets have been resolved or declared as obsolete. Fingers crossed, I close the ticket.