Skip to content
inMotion

Dataset reference

Generated from inmotion.datasets; regenerate the measured fields with inmotion datasets manifest --write. inmotion datasets verify re-checks every file against its recorded checksum.

The datasets are tracked in git, so their history is the version history. This table adds what git cannot express: checksums, class balance, lineage, and whether the labels can be trusted.

Canonical training set: ds-augmented-v3

id file rows devices classes status label audit
ds-pure data/processed/dataset_only_pure.csv 160 2 AA, AB, BA, BB superseded known_incorrect
ds-noise data/processed/dataset_only_noise.csv 1,196 6 AA, AB, BA, BB superseded known_incorrect
ds-augmented-v1 data/processed/dataset_augmented.csv 10,113 8 AA, AB, BA, BB superseded known_incorrect
ds-icaisf data/processed/dataset.icaisf.csv 2,251 8 AA, AB, BA, BB current unverified
ds-base-v2 data/processed/dataset.csv 3,511 13 AA, AB, BA, BB current unverified
ds-augmented-v2 data/processed/dataset_augmented2.csv 20,275 13 AA, AB, BA, BB superseded unverified
ds-augmented-v3 data/processed/dataset_augmented3.csv 35,839 13 AA, AB, BA, BB current unverified

Details

Section titled “Details”

ds-pure

Section titled “ds-pure”
  • file: data/processed/dataset_only_pure.csv
  • rows: 160 (13,962 bytes)
  • sha256: 5649b6e3fcda5b3225bd8b9c1b8febb20a137d5043dd863319ab299f3d663a23
  • columns: mac, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, label, noise_label
  • class counts: AA=40, AB=40, BA=40, BB=40
  • devices: 2
  • has synthetic column: False
  • first commit: c221d7f1a3ec (2026-02-04)
  • role: subset; status: superseded
  • label audit: known_incorrect
  • note: Early collection. The project owner reports this lineage carries incorrect class labels; kept for provenance only. Do not train on or report from it without re-auditing the labels.
  • lineage: description=isolated single-device collection, devices=2, successor=data/processed/dataset_only_noise.csv

ds-noise

Section titled “ds-noise”
  • file: data/processed/dataset_only_noise.csv
  • rows: 1,196 (102,899 bytes)
  • sha256: 97c33bb3e07aa1777a2aeedf7346478a7bc765b3bed1860fb21e8fdee7495833
  • columns: mac, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, label, noise_label
  • class counts: AA=298, AB=300, BA=298, BB=300
  • devices: 6
  • has synthetic column: False
  • first commit: c221d7f1a3ec (2026-02-04)
  • role: subset; status: superseded
  • label audit: known_incorrect
  • note: Early collection. The project owner reports this lineage carries incorrect class labels; kept for provenance only. Do not train on or report from it without re-auditing the labels.
  • lineage: description=noisy multi-device collection, devices=6, successor=8-device pool

ds-augmented-v1

Section titled “ds-augmented-v1”
  • file: data/processed/dataset_augmented.csv
  • rows: 10,113 (951,333 bytes)
  • sha256: 507973bcd63245b4dd62e3b052a83ac29394e5e06ff9276b7b97d60b399435fa
  • columns: mac, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, label, noise, concurrent_noise_path, synthetic
  • class counts: AA=2,574, AB=2,490, BA=2,478, BB=2,571
  • devices: 8
  • has synthetic column: True
  • first commit: cd4cb67abe85 (2026-04-25)
  • role: augmented; status: superseded
  • label audit: known_incorrect
  • note: Early collection. The project owner reports this lineage carries incorrect class labels; kept for provenance only. Do not train on or report from it without re-auditing the labels.
  • lineage: description=augmentation config 1, devices=8, parent=8-device pool, successor=data/processed/dataset_augmented2.csv

ds-icaisf

Section titled “ds-icaisf”
  • file: data/processed/dataset.icaisf.csv
  • rows: 2,251 (200,238 bytes)
  • sha256: 05f11fb4cba9176d4361302630c02e7c7f09e352ab997b616cb1fa33265a74c9
  • columns: mac, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, label, noise, concurrent_noise_path
  • class counts: AA=578, AB=550, BA=546, BB=577
  • devices: 8
  • has synthetic column: False
  • first commit: 0f5c0b1c4438 (2026-08-24)
  • role: variant; status: current
  • label audit: unverified
  • note: ICAISF variant (4 classes AA/AB/BA/BB), 8-device pool. This is the dataset the ISAC paper’s ensemble was scored on, so its labels need confirming: if this lineage is the older one, the published numbers are affected.
  • lineage: description=ICAISF paper variant, devices=8

ds-base-v2

Section titled “ds-base-v2”
  • file: data/processed/dataset.csv
  • rows: 3,511 (312,378 bytes)
  • sha256: 39dc03701e7882ac850262500b287e36afab5d7a9657edc4618604b0052dccc1
  • columns: mac, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, label, noise, concurrent_noise_path
  • class counts: AA=878, AB=880, BA=876, BB=877
  • devices: 13
  • has synthetic column: False
  • first commit: 70d22b298f90 (2025-10-21)
  • role: base; status: current
  • label audit: unverified
  • note: 13-device pool, collected after the earlier lineage. Source for data/processed/dataset_augmented2.csv and data/processed/dataset_augmented3.csv.
  • lineage: description=current base dataset, devices=13, successors=[‘data/processed/dataset_augmented2.csv’, ‘data/processed/dataset_augmented3.csv’]

ds-augmented-v2

Section titled “ds-augmented-v2”
  • file: data/processed/dataset_augmented2.csv
  • rows: 20,275 (1,908,203 bytes)
  • sha256: 11fed6d02bd8de70f3c573ece5943097ed5acf48efe78fcd44b0a8f1b11191b7
  • columns: mac, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, label, noise, concurrent_noise_path, synthetic
  • class counts: AA=5,220, AB=4,930, BA=4,910, BB=5,215
  • devices: 13
  • has synthetic column: True
  • first commit: d465854f25a3 (2026-08-05)
  • role: augmented; status: superseded
  • label audit: unverified
  • note: Augmentation config 2 over the 13-device pool. Legitimate, but superseded by config 3.
  • lineage: description=augmentation config 2, devices=13, parent=data/processed/dataset.csv, successor=data/processed/dataset_augmented3.csv

ds-augmented-v3

Section titled “ds-augmented-v3”
  • file: data/processed/dataset_augmented3.csv
  • rows: 35,839 (3,370,206 bytes)
  • sha256: 82a3c637a70eaf92c431f244b9d51188a80e309ffa2cdad5375f1725752cb0db
  • columns: mac, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, label, noise, concurrent_noise_path, synthetic
  • class counts: AA=8,962, AB=8,980, BA=8,944, BB=8,953
  • devices: 13
  • has synthetic column: True
  • first commit: d465854f25a3 (2026-08-05)
  • role: augmented; status: current
  • label audit: unverified
  • note: Canonical training set: the newest augmentation config, built from data/processed/dataset.csv. This is what the trained JEPA/SIGReg checkpoints were trained on, so it is the reference for reproducing their scaling.
  • lineage: description=augmentation config 3 (latest), devices=13, parent=data/processed/dataset.csv, successor=None

Schema

Section titled “Schema”
column meaning
mac device identifier. Retained for cross-device analysis and for
leave-one-device-out evaluation; not a model feature.
1..10 RSSI in dBm, one reading per second for 10 seconds. The only
raw model features.
label route taken: AA, AB, BA, BB. The prediction target.
noise / noise_label whether other devices transmitted concurrently.
concurrent_noise_path which route the interfering device was taking, when
known. Used only by the interference analysis, never as a model input.
synthetic padding produced by inmotion data augment (v2/v3 only).

Tiers

Section titled “Tiers”
tier contents produced by
data/raw/ captures off the access point (Wavecom .txt) field collection
data/interim/ one CSV per collection session inmotion data export
data/processed/ merged and augmented training sets
inmotion data merge, inmotion data augment

inmotion.data.paths.resolve searches processed, interim and raw in that order, so a bare filename still resolves.