Standard

Record decoder variant that decodes node features, targets, graph connectivity, and auxiliary columns from blocks written by the data pipeline in the standard columnar schema: each key is one column of the block, ragged columns carry per-record row offsets. It is the counterpart to the standard attribute decoder and the default choice for datasets produced by IceGraph.

It is also the decoder for IceCube simulation, and is the only part of the engine that knows about simweights: the weights role is decoded by building the generation surface and weighting each record against a flux.

Weighting

The i3-simulation processor does not store a weight. It stores the per-event generation quantities as a column and the shard’s simulation metadata as a local attribute, so the mixture is chosen when the data is read rather than when it is written.

On load this decoder:

• Builds the surface of each simulation set with simweights, treating the set’s loaded shards as one.

  • Counts the input files each shard records under nfiles, including files without events.

• Sums the surfaces of the sets of each type.

• Weights each block’s generation columns against those sums.

• Returns the final weights, one per record.

Shards without a simulation are inferred to be real data and get no weights.

The weights role reads the generation column through the decode service’s keymap.weights, which defaults to generation.

Mixed simulation

A source may mix simulation types, for instance CORSIKA files alongside NuGen ones. The sets of each type are summed into their own surface, each surface is weighted against its own flux, and each record is routed to one of them. Configure one flux per type.

Routing is per record because a block spans shards. The i3-simulation processor writes a sim code naming the set into the generation columns of every event and declares, in the same shard’s attributes, what type that code stands for and which column carries it, so the decoder learns the mapping from the data rather than from a convention shared with the writer.

Note

Real data and simulation cannot be loaded together.

Configuration

Selected as name: standard.

Option	Description	Type	Default
flux	Flux model each simulation type is weighted against, keyed by the simulation that generated it (corsika, nugen). One entry is needed per type present in the data.	mapping	{}

Each entry selects a model from one of the two packages that ship them:

Option	Description	Type	Default
source	Package the model comes from: simweights for the cosmic ray fluxes, nuflux for the atmospheric neutrino fluxes.	str	required
name	Model name as that package spells it, e.g. GaisserH4a or honda2006.	str	required
kwargs	Constructor arguments for a simweights model, property assignments for a nuflux one.	mapping	{}

services:
  decode:
    records:
      name: standard
      kwargs:
        flux:
          corsika:
            source: simweights
            name: GaisserH4a
            kwargs: { }

A source mixing CORSIKA and NuGen files (for example) configures both, and each is weighted against its own model:

flux:
  corsika:
    source: simweights
    name: GaisserH4a
    kwargs: { }

  nugen:
    source: nuflux
    name: honda2006
    kwargs: { }

Note

nuflux is not a pip dependency; it comes from the IceTray environment on CVMFS, and is imported only when a nuflux model is configured.