JANUS/README.md

# JANUS

**JANUS** (Joint Anomaly via Normalizing-flows of Unified States) — flow-matching unsupervised network anomaly detection over packet sequences.

JANUS is a packet-causal Transformer with **two output heads on a shared backbone**:

- **Continuous Flow Matching head** over the (size, IAT, win) packet channels.
- **Discrete Flow Matching head** over the 6 binary protocol-flag / direction channels.

Trained jointly on benign traffic only (no attack labels at any stage). The deployable scalar score is a **Mahalanobis-OAS distance** over a 10-d per-flow score vector emitted by the trained model, with the aggregator fit on benign val only — entirely unsupervised end-to-end.

JANUS is the first NIDS method to use Flow Matching as the training paradigm in mixed continuous–discrete state spaces over packet sequences.

## Headline results

3-seed mean ± std AUROC. Selection-bias-free Mahalanobis-OAS aggregator on the 10-d JANUS score vector, fit on benign val only.

### Within-dataset comparison (AUROC %, mean ± std)

| Method | Venue | CIC-IDS2017 | CIC-DDoS2019 | CIC-IoT2023 | ISCXTor2016 |
|---|---|---:|---:|---:|---:|
| Isolation Forest | classical | 55.27 ± 0.4 † | — | — | — |
| OCSVM | classical | 59.59 ± 0.6 † | — | — | — |
| AnoFormer | ICLR'22 | 63.37 ± 0.7 † | — | — | — |
| GANomaly | BMVC'18 | 82.75 ± 5.6 † | — | — | — |
| RD4AD | CVPR'22 | 83.78 ± 0.8 † | — | — | — |
| TSLANet | ICML'24 | 84.45 ± 1.7 † | — | — | — |
| ARCADE | — | 84.85 ± 2.0 † | — | — | — |
| MFAD | — | 86.02 ± 0.8 † | — | — | — |
| STFPM | BMVC'21 | 86.29 ± 1.7 † | — | — | — |
| MMR | — | 89.26 ± 1.2 † | — | — | — |
| Shafir NF + Shapley | arXiv'26 | 93.03 ‡ | 93.00 ‡ | 72.24 ± 6.08 ★ | 87.31 ‡ |
| ConMD | TIFS'26 | 94.43 ± 0.1 † | — | — | — |
| **JANUS (ours)** | — | **98.26 ± 0.35** | **99.18 ± 0.05** | **95.90 ± 0.22** | **99.09 ± 0.13** |

† Numbers from ConMD (TIFS'26) Table I; protocol = train 10 K benign / test 5 K + 5 K balanced; 5-seed mean ± std.
‡ Numbers from Shafir et al. (arXiv'26) headline tables; protocol = train 10 K benign / SHAP-selected feature subsets per dataset (single NF).
★ Reproduced by us (3-seed mean ± std, 2-NF ensemble, CSV pipeline, paper-specified 5-feat SHAP subset). Shafir's paper does not publish an AUROC for CIC-IoT2023 — only F1 = 99.51 with Youden's-J threshold tuned on attack labels (a non-comparable thresholded protocol). For threshold-free head-to-head AUROC on this dataset we cite our reproduction.

JANUS sets new SOTA on **4/4 within-dataset benchmarks** under matched AUROC protocol — CIC-IDS2017 **+3.83**, CIC-DDoS2019 **+6.18**, CIC-IoT2023 **+23.66** (vs reproduced Shafir), ISCXTor2016 **+11.78** — all margins outside seed std. JANUS is fully unsupervised (benign-only training, no attack labels at any stage) and uses the Mahalanobis-OAS aggregator over its 10-d raw score vector with parameters fit on benign val only. Thresholded F1 metrics for JANUS across all four datasets are in `RESULTS.md` Section D and `artifacts/route_comparison/THRESHOLDED.md`.

### 3×3 cross-dataset transfer matrix

Source (rows) trained on 10K benign of source dataset; target (columns) tested on full target benign + **all** target attacks. Aggregator fit on target benign val only — no attack labels at any stage. Diagonal italic = within-dataset.

| Source ↓ / Target → | CICIDS17 | CICDDoS19 | CICIoT23 |
|---|---|---|---|
| **CICIDS17** | _0.9826 ± 0.0035_ | **0.9690 ± 0.0047** | 0.8698 ± 0.0031 |
| **CICDDoS19** | 0.9413 ± 0.0212 | _0.9918 ± 0.0005_ | 0.8767 ± 0.0068 |
| **CICIoT23** | 0.9394 ± 0.0063 | 0.9030 ± 0.0075 | _0.9590 ± 0.0022_ |

Forward CICIDS17→CICDDoS19 (0.969) beats Shafir 0.89 by **+0.08**; reverse CICDDoS19→CICIDS17 (0.941) approximately matches Shafir 0.93. CICIoT23 is hardest both as source and target — its IoT-protocol diversity makes the "benign of source ≈ benign of target" assumption brittle. Full table at `artifacts/route_comparison/CROSS_MATRIX_3x3.md`.

### Ablations (architecture & aggregator)

Two orthogonal ablation axes, each evaluated **within-dataset** (4 datasets × 3 seeds) **and** **cross-dataset** (3×3 transfer × 3 seeds):

- **Group A** — 7 alternative aggregators on the same JANUS-full sub-score vector (post-processing only; no retraining).
- **Group B** — 5 architecture variants, each retrained 4 datasets × 3 seeds = 60 runs + 90 cross-evals.

Every load-bearing JANUS design choice has the **same shape of ablation curve**: small in-distribution cost, large cross-dataset gain.

| Component (removed in ablation) | Variant | Within Δ | Cross-mean Δ | Cross-worst Δ |
|---|---|---:|---:|---:|
| FLOW token (global context) | B1 | **−0.94** | −6.70 | −19.97 |
| Packet sequence | B2 | +0.15 | **−23.82** | **−36.27** |
| Cont/disc head split (drop disc head) | B3 | +0.44 | **−13.14** | **−25.03** |
| CFM head (drop continuous side) | B4 | **−2.37** | −2.03 | −2.86 |
| Joint training of two heads | B5 | +0.20 | **−18.93** | **−27.54** |
| OAS Mahalanobis aggregator | A1 vs A5 | +0.37 | **−15.88** | **−27.38** |

Three ablations (B3 / B5 / A-aggregator) **marginally beat JANUS-full at within-dataset evaluation** but collapse on at least one cross-dataset transfer direction. The disc head, joint training, and OAS aggregator are deliberate trades: their value is exclusively in cross-dataset robustness.

Full headline summary: `artifacts/ablation/ABLATION_SUMMARY.md`. Per-variant 3×3 cross matrices: `artifacts/ablation/ABLATION_CROSS_B_full.md` and `artifacts/ablation/ABLATION_TABLE_CROSS_full.md`.

## Layout

```
common/                    Data contract — single source of truth for the
                           9-d packet schema, 20-d packet-derived flow schema,
                           label normalization, and packet preprocessing.
Mixed_CFM/                 The JANUS model. Mixed continuous–discrete CFM
                           with two output heads on a shared causal Transformer.
                             configs/   Per-(dataset × seed) training configs.
                             model.py   MixedTokenCFM + MixedVelocity.
                             train.py / eval_phase1.py / eval_cross.py
Unified_CFM/               Legacy unified token CFM. Mixed_CFM imports its
                           AdaLNBlock + sinusoidal time embedding for backbone
                           reuse. Kept as internal ablation reference.
scripts/                   Workspace-level pcap → artifact pipeline,
                           CSV adapters, cross-package eval tooling.
  download/                UNB/CIC dataset downloaders.
  baselines/               Third-party baseline runners (Kitsune, Shafir-NF,
                           Anomaly-Transformer).
  aggregate/               Mahalanobis-OAS score-router + cross-matrix
                           orchestration. aggregate_score_router.py is the
                           deployable score path; run_cross_3x3.sh +
                           cross_3x3_table.py produce the cross matrix.
                           aggregate_ablation.py / aggregate_ablation_cross.py /
                           aggregate_ablation_cross_B.py produce the ablation
                           tables in artifacts/ablation/.
  ablation/                B-group ablation training/eval drivers
                           (generate_configs.py, run_groupB.sh,
                           run_cross_groupB.sh).
tests/                     Data-contract unit tests.
```

The following directories are **gitignored** (live on the dev box, not in the repo):

```
artifacts/                 All run outputs (checkpoints, eval JSONs, score
                           npzs, figures). Per-(dataset × seed) model dirs at
                           artifacts/route_comparison/janus_<ds>_seed<N>/.
datasets/                  Raw + processed datasets (~1 TB).
baselines/                 Third-party baseline forks (Kitsune-py,
                           Anomaly-Transformer, ConMD, ganomaly, TIPSO-GAN, ...).
paper/                     Paper sources & external PDFs (Shafir 2026, Lipman
                           2210.02747, etc.).
.venv/                     uv-managed Python 3.14 virtual env.
```

## Data contract

Every processed dataset under `datasets/<name>/processed/` ships an aligned triple, all with the same row order (`flow_id = arange(N)`):

```
packets.npz            packet_tokens [N, T_full, 9], packet_lengths [N], flow_id [N]
                       (or full_store/ — sharded PacketShardStore — for large datasets)
flows.parquet          flow_id + label + 5-tuple metadata (src_ip, dst_ip, ports, protocol)
flow_features.parquet  flow_id + label + 20 canonical packet-derived features
```

The 9-d packet schema and 20-d flow schema are FIXED in `common/data_contract.py`. Flow features are computed by `compute_flow_features_from_packets(packet_tokens, lens)` so row alignment is guaranteed.

## Quick start

```bash
# Train JANUS on CICIDS2017 (3 seeds available: 42, 43, 44)
cd Mixed_CFM
uv run --no-sync python train.py --config configs/cicids2017_seed42.yaml

# Phase-1 evaluation: per-attack-class AUROC + 10-d score export
uv run --no-sync python eval_phase1.py \
    --model-dir <model_dir> --out-dir <eval_dir>

# Single cross-dataset eval
uv run --no-sync python eval_cross.py \
    --model-dir <src_model_dir> \
    --target-store datasets/<tgt>/processed/full_store \
    --target-flows datasets/<tgt>/processed/flows.parquet \
    --target-flow-features datasets/<tgt>/processed/flow_features.parquet \
    --benign-label normal --n-benign 10000 --n-attack 1000000 \
    --out <result.json>

# 3×3 cross matrix (6 off-diagonal directions × 3 seeds, 2-GPU parallel)
bash ../scripts/aggregate/run_cross_3x3.sh
uv run --no-sync python ../scripts/aggregate/cross_3x3_table.py
```

JANUS hyper-parameters (locked in `Mixed_CFM/configs/<dataset>_seed*.yaml`):

```yaml
T: 64                  # max packet sequence length
d_model: 128
n_layers: 4
n_heads: 4
sigma: 0.1             # within-dataset; cross uses 0.6
lambda_disc: 1.0
use_ot: true           # OT-CFM (Sinkhorn coupling on benign batch)
reference_mode: causal_packets    # Route A: packet-causal attention
```

## Producing the deployable scalar score

`eval_phase1.py` exports a 10-d per-flow score vector to `phase1_scores.npz`:

```
3 continuous-side scores  : terminal_norm, terminal_flow, terminal_packet
7 discrete-side scores    : disc_nll_total + disc_nll_ch{2,3,4,5,6,7}
                            (direction + 5 TCP flags)
```

The deployable scalar is the Mahalanobis-OAS distance:

```
d²(s) = (s − μ)ᵀ Σ⁻¹ (s − μ),    where (μ, Σ) come from sklearn.covariance.OAS
                                 fit on benign val ONLY (no attack labels).
```

Reference implementation: `scripts/aggregate/aggregate_score_router.py`. It reads `artifacts/route_comparison/janus_<ds>_seed*/phase1_scores.npz` and `artifacts/route_comparison/cross/janus_seed*_<src>_to_<tgt>.npz`, then writes `artifacts/route_comparison/SCORE_ROUTER.md` (within-dataset rows) and `artifacts/route_comparison/CROSS_MATRIX_3x3.md` (cross matrix, via `cross_3x3_table.py`).

## Tests

```bash
uv run --no-sync python -m pytest tests/ Mixed_CFM/tests/ Unified_CFM/tests/
```

## Adding a new dataset

Write one driver at `scripts/extract_<name>.py` that calls `extract_lib.extract_dataset(...)` (see `scripts/extract_cicids2017.py` as the reference template). The driver hardcodes CSV column names, timestamp formats, benign aliases, and drop patterns as module constants, then feeds `extract_lib` a per-day `(canonical_key → [(row_idx, ts_epoch)])` mapping and a per-day pcap file map. The extract pipeline writes all three artifacts (packets.npz, flows.parquet, flow_features.parquet) row-aligned by `flow_id = arange(N)`.

To upgrade an existing artifact pair that lacks `flow_features.parquet`, run `scripts/generate_flow_features.py --packets-npz ... --flows-parquet ... --out ...` (or `--source-store` for sharded stores).

Common gotcha: if CSV timestamps and pcap epochs are in different time zones, `extract_lib` prints a diagnostic with the recommended `--time-offset`; rerun with that value.

## Authoritative documents

- `RESULTS.md` — full headline tables, per-attack analysis, JANUS configuration, thresholded operating-point metrics, what the experiments proved / disproved.
- `artifacts/ablation/ABLATION_SUMMARY.md` — paper-facing ablation summary (Group A aggregator + Group B architecture, both within and cross views).
- `Mixed_CFM/model.py` and `common/data_contract.py` — model + data-contract source of truth.

## Python environment

- `requires-python = ">=3.14"`; PyTorch pinned to the `pytorch-cu128` index, plus `mamba-ssm`, `causal-conv1d`, `scapy`, `dpkt`, `pyarrow`, `sklearn` (for the OAS aggregator).
- Two `pyproject.toml` files exist: root and `Mixed_CFM/`; they are not declared as a uv workspace and resolve independently. Run `uv run ...` from whichever directory owns the entry point.
- `Unified_CFM/` has no `pyproject.toml`; it uses the root venv (`uv run --no-sync python <script.py>`).
- Scripts under `scripts/download/` are pure stdlib — invoke with `python3`.