File formats

TL;DR

The canonical network file is a sparse directed edge table with integer occupations under the column schema source target occ_num. Zero occupations are dropped; negative or fractional values are rejected at the boundary.

Canonical edge-table schema

Column	Type	Rule
source	non-negative integer	source node id
target	non-negative integer	target node id
occ_num	non-negative integer	occupation number; zero rows are ignored

Canonical CSV:

source,target,occ_num
0,1,12
0,2,3
2,1,7

Readers may accept weight as an input alias where the current I/O code explicitly supports it; writers always emit occ_num. weight is never presented as the canonical writer schema.

Supported edge formats

Extension	Format	Read	Write
.csv	CSV	yes	yes
.tsv, .tab	TSV	yes	yes
.parquet, .pq	Apache Parquet	yes	yes
.arrow, .ipc, .feather	Arrow IPC	yes	yes
.graphml	GraphML	yes	no
.mtx, .mm	Matrix Market	yes	no
.net, .paj	Pajek	yes	no

Probability/rate table schema

Custom sparse probabilities use:

Column	Meaning
source	source node id
target	target node id
probability	value in [0, 1]

Some custom Poisson filter paths use the same sparse pair idea with a rate column at the CLI boundary.

Python I/O

from menobis.data import read_edges, write_edges

edges = read_edges("network.csv")
write_edges(edges, "network.parquet")

Round-trip guarantee: writing then reading back preserves source, target, and positive occ_num entries.

CLI scope

The Python API is the authoritative full model interface; CLI command strings expose a convenience subset and may retain command names that do not mirror the model ontology exactly (see CLI overview). Use the Python I/O functions for format conversion.