BattleTag/JANUS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge RESULTS_THRESHOLDED.md into RESULTS.md (section D); rewrite README.md for JANUS

Browse Source
This commit is contained in:
BattleTag
2026-05-07 21:07:08 +08:00
parent e5d96494a5
commit 056b1445db
3 changed files with 190 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

186
README.md
View File

@@ -1,57 +1,151 @@
# mambafortrafficmodeling
# JANUS
Network traffic anomaly detection with continuous flow matching (CFM). Three
sibling model packages over a shared canonical data contract.
**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.
| Task | Shafir 2026 SOTA | **JANUS** | Δ |
|---|---|---|---|
| ISCXTor2016 (NonTor → Tor) | 0.8731 | **0.9908 ± 0.0012** | **+0.118** |
| CICIDS2017 within | 0.9303 | **0.9845 ± 0.0030** | **+0.054** |
| CICDDoS2019 within | 0.93 | **0.9913 ± 0.0009** | **+0.061** |
| CICIDS2017 → CICDDoS2019 cross | 0.89 | **0.9594 ± 0.0046** | **+0.07** |
| CICDDoS2019 → CICIDS2017 reverse cross | 0.93 | **0.9301 ± 0.0122** | matches |
3/3 directly comparable within-dataset benchmarks beat external Shafir 2026 SOTA. CICIDS2017→CICDDoS2019 cross also beats; reverse direction matches. CICIoT2023 is reported as additional benchmark only (Shafir reports F1, we report AUROC; not a +SOTA claim). See `RESULTS.md` for caveats and the full headline table.
## Layout
- `common/data_contract.py` — single source of truth for the canonical
packet schema (9-d) and flow schema (20-d, packet-derived). All three
packages import constants and helpers from here.
- `Packet_CFM/` — packet-sequence OT-CFM with explicit σ-band benign
distribution learning.
- `Flow_CFM/` — flow-level CFM on the workspace-canonical 20-d packet-derived
`flow_features.parquet`. Legacy 61-d CICFlowMeter CSV caches are kept only
for paper reproduction (`--legacy-csv-features` flag).
- `Unified_CFM/` — unified packet+flow token CFM. **Current SOTA model** —
used for all main results (within-dataset SOTA on ISCXTor2016 / CICIDS2017
/ CICDDoS2019, near-SOTA cross-dataset).
- `datasets/<name>/processed/` — canonical artifact bundle:
- `packets.npz` (small/medium) or `full_store/` (large, sharded)
- `flows.parquet` (label + 5-tuple metadata)
- `flow_features.parquet` (20-d packet-derived, row-aligned)
- `scripts/` — workspace-level pcap → artifact extraction, CSV adapters,
cross-package eval tooling. `scripts/download/` is also here.
- `artifacts/` — run outputs (training checkpoints, eval JSONs, reports).
Phase 0 / 1 / 2 / 2.5 experiment summaries live under
`artifacts/phase{0,1,2}*` directories.
- `paper/` — paper PDFs we compare against (Shafir 2026 NF, ConMD 2026,
TIPSO-GAN 2026, Lipman 2210.02747 flow matching).
```
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).
tests/ Data-contract unit tests.
```
The root keeps only workspace-level files. All model/training/eval code
lives under one of the three packages.
The following directories are **gitignored** (live on the dev box, not in the repo):
## Current best results (Unified_CFM, λ=0.3, 3 seeds)
```
artifacts/ All run outputs (checkpoints, eval JSONs, score npzs,
figures). Score-router aggregator at
artifacts/route_comparison/aggregate_score_router.py.
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.
```
Shafir baselines verified from paper PDF tables — see `artifacts/locked_baselines.md`.
## Data contract
| Task | Shafir 2026 SOTA | Our best | Δ |
|---|---|---|---|
| ISCXTor2016 (NonTor → Tor) | 0.8731 (Table VI) | 0.9945 ± 0.0011 (σ=0.1) | **+0.121** |
| CICIDS2017 within (10k/10k Shafir protocol) | 0.9303 (Table VII) | **0.9858 ± 0.0021** (σ=0.6) | **+0.055** |
| CICDDoS2019 within | 0.93 (Table IX) | **0.9958 ± 0.0010** (σ=0.1) | **+0.066** |
| CICIDS2017 → CICDDoS2019 cross (`terminal_norm`) | 0.89 (Table IX, IDS→DDoS row) | **0.9109 ± 0.0032** (σ=0.6) | **+0.021** |
| CICIDS2017 → CICDDoS2019 cross (`terminal_flow`) | 0.89 | **0.9197 ± 0.0036** | **+0.030** |
Every processed dataset under `datasets/<name>/processed/` ships an aligned triple, all with the same row order (`flow_id = arange(N)`):
**4 of 4 reported tasks achieve SOTA**. Cross-dataset baseline was previously misread as 0.93; the IDS→DDoS direction in Shafir Table IX is 0.89.
```
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
```
Plus an architectural contribution: a `flow_consistency` diagnostic score
that lifts from random (~0.6) to discriminative (~0.9) only when the model
is trained with the masked-prediction consistency loss. On SSH-Patator (the
hardest CICIDS2017 class for `terminal_norm` at 0.64) it reaches 0.94.
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.
Authoritative result tables live in `RESULTS.md` (root) and
`artifacts/locked_baselines.md` (Shafir baseline verification trail).
Thresholded F1 / Precision / Recall / TPR@FPR under unsupervised threshold
protocol: `RESULTS_THRESHOLDED.md`.
Per-attack-family multi-seed analysis: `artifacts/phase25_multiseed_2026_04_25/PER_ATTACK_TABLE.md`.
## 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_ac_combo_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>
# Cross-dataset evaluation
uv run --no-sync python eval_cross.py \
--src-model <src_model_dir> \
--tgt-data datasets/<tgt>/processed/ \
--out <result.json>
```
JANUS hyper-parameters (locked in `Mixed_CFM/configs/<dataset>_ac_combo_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: `artifacts/route_comparison/aggregate_score_router.py` (artifacts/ is gitignored; the script lives on the dev box).
## 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, ablations, per-attack analysis, JANUS configuration, thresholded operating-point metrics, what the experiments proved / disproved.
- `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`.

57
RESULTS.md
View File

@@ -159,10 +159,10 @@ This is **+0.31 over our own legacy memory baseline of 0.62**. The "main
attack direction" recorded in `reverse_cross_score_redirection_2026_04_25`
is now substantially solved.
Thresholded F1 / Precision / Recall / TPR@FPR (unsupervised protocol, τ from
benign-val percentile) are reported separately in `RESULTS_THRESHOLDED.md`.
Headline thresholded numbers: CICDDoS2019 within `terminal_norm` F1=0.993 ± 0.001
at τ=P95; cross `terminal_norm` F1=0.632 ± 0.051 at τ=P95 (precision ≈ 0.95, recall ≈ 0.47).
Thresholded F1 / Precision / Recall / TPR@FPR under the unsupervised threshold
protocol are reported in **section D** below. Headline thresholded numbers:
CICDDoS2019 within `terminal_norm` F1=0.993 ± 0.001 at τ=P95; cross `terminal_norm`
F1=0.632 ± 0.051 at τ=P95 (precision ≈ 0.95, recall ≈ 0.47).
> **Note on cross-dataset baseline**: Shafir's Table IX is asymmetric.
> The IDS2017→DDoS2019 direction (which we evaluate) reads **0.89**, not
@@ -175,6 +175,52 @@ at τ=P95; cross `terminal_norm` F1=0.632 ± 0.051 at τ=P95 (precision ≈ 0.95
> Single-policy σ=0.6 also beats Shafir on 4/4. Full 4×2 sensitivity table
> in `artifacts/sigma_validation.md`.
### D. Thresholded operating-point metrics
⚠️ Numbers in this section are from the **Unified_CFM legacy** recipe (σ=0.1
within, σ=0.6 cross, λ=0.3, single fixed score). Equivalent thresholded
numbers for current JANUS + Mahalanobis-OAS have not been recomputed yet;
the AUROC tables (A/B/C above) are the authoritative JANUS comparison.
**Protocol**: τ is set from a benign-val half (A); F1 / Precision / Recall /
FPR are measured on benign-val half B + attack. AUROC / AUPRC use full
benign val + attack. TPR@FPR is measured on the test half. Both percentiles
τ ∈ {P95, P99} are reported because they correspond to different operating
points and F1 is sensitive to that choice.
**CICDDoS2019 within** (σ=0.1, λ=0.3):
| Score | AUROC | AUPRC | F1 (P95) | Prec (P95) | Recall (P95) | FPR (P95) | F1 (P99) | TPR@1%FPR | TPR@5%FPR |
|---|---|---|---|---|---|---|---|---|---|
| `terminal_norm` | 0.9960 ± 0.0011 | 0.9975 ± 0.0008 | 0.9932 ± 0.0012 | 0.9881 ± 0.0015 | 0.9983 ± 0.0008 | 0.0481 ± 0.0061 | 0.9112 ± 0.0402 | 0.9013 ± 0.0540 | 0.9980 ± 0.0014 |
| `terminal_flow` | 0.9885 ± 0.0028 | 0.9918 ± 0.0017 | 0.9788 ± 0.0086 | 0.9868 ± 0.0009 | 0.9710 ± 0.0163 | 0.0517 ± 0.0030 | 0.7752 ± 0.0128 | 0.6052 ± 0.0347 | 0.9697 ± 0.0169 |
**CICIDS2017 → CICDDoS2019 cross** (σ=0.6, λ=0.3):
| Score | AUROC | AUPRC | F1 (P95) | Prec (P95) | Recall (P95) | FPR (P95) | F1 (P99) | TPR@1%FPR | TPR@5%FPR |
|---|---|---|---|---|---|---|---|---|---|
| `terminal_norm` | 0.9109 ± 0.0032 | 0.8974 ± 0.0047 | 0.6321 ± 0.0513 | 0.9545 ± 0.0045 | 0.4745 ± 0.0550 | 0.0441 ± 0.0011 | 0.4202 ± 0.0171 | 0.2685 ± 0.0139 | 0.4940 ± 0.0399 |
| `terminal_flow` | 0.9197 ± 0.0036 | 0.8957 ± 0.0086 | 0.6324 ± 0.0585 | 0.9517 ± 0.0055 | 0.4762 ± 0.0639 | 0.0469 ± 0.0019 | 0.4028 ± 0.0049 | 0.2534 ± 0.0039 | 0.4776 ± 0.0636 |
**Reading**:
- *Within-dataset CICDDoS2019* saturates: at τ=P95 F1 ≈ 0.99 with balanced
precision and recall ≈ 0.99; at τ=P99 (≈1% FPR) F1 ≈ 0.91 with TPR@1%FPR
≈ 0.90. The model is a working detector at fixed thresholds, not just an
AUROC artifact.
- *Cross-dataset CICIDS2017→CICDDoS2019* keeps AUROC ≈ 0.91 but at fixed τ
shows precision ≈ 0.95 / recall ≈ 0.50 at P95 and ≈0.27 at 1% FPR — the
cross-dataset domain shift compresses the score gap, so source-calibrated
thresholds are conservative on target. **AUROC alone overstates
deployability cross-dataset; thresholded numbers are the honest figure.**
**TIPSO-GAN comparability**: TIPSO-GAN's CICDDoS2019 F1 ≈ 0.99 is reported
under a **supervised** protocol (model has seen attack examples). Our F1
≈ 0.99 on CICDDoS2019 within is achieved under the **unsupervised** protocol
(benign-only training, threshold from benign val), which is the strictly
harder setting. Direct F1 numerical equivalence; protocol asymmetry is in
our favor.
## Methodological contribution: `flow_consistency` diagnostic score
Phase 2 masked-prediction consistency loss unlocks a new score that is
@@ -319,9 +365,6 @@ artifacts.
## Source artifacts
- `RESULTS_THRESHOLDED.md` — F1 / Precision / Recall / TPR@FPR under unsupervised
threshold protocol (τ = benign-val P95/P99) for CICDDoS2019 within and
CICIDS2017→CICDDoS2019 cross.
- `artifacts/locked_baselines.md` — verified Shafir baselines (PDF inspection trail).
- `artifacts/sigma_validation.md` — full 4×2 σ-sensitivity table (σ ∈ {0.1, 0.6} ×
4 tasks, 3 seeds each) and per-task σ-selection protocol.

34
RESULTS_THRESHOLDED.md
View File

@@ -1,34 +0,0 @@
# Thresholded metrics — unsupervised AD protocol
3-seed mean ± std. Threshold τ is set on benign-val half A; F1 / Precision / Recall / FPR are measured on benign-val half B + attack. AUROC/AUPRC use full benign val + attack. TPR@FPR is measured on the test half.
Both percentiles are reported because P95 and P99 give different operating points; F1 numbers are sensitive to that choice.
Primary score: `terminal_norm`. `terminal_flow` is reported on cross because RESULTS.md headlines both.
## CICDDoS2019 within (σ=0.1, λ=0.3)
| Score | AUROC | AUPRC | F1 (P95) | Prec (P95) | Recall (P95) | FPR (P95) | F1 (P99) | TPR@1%FPR | TPR@5%FPR |
|---|---|---|---|---|---|---|---|---|---|
| `terminal_norm` | 0.9960 ± 0.0011 | 0.9975 ± 0.0008 | 0.9932 ± 0.0012 | 0.9881 ± 0.0015 | 0.9983 ± 0.0008 | 0.0481 ± 0.0061 | 0.9112 ± 0.0402 | 0.9013 ± 0.0540 | 0.9980 ± 0.0014 |
| `terminal_flow` | 0.9885 ± 0.0028 | 0.9918 ± 0.0017 | 0.9788 ± 0.0086 | 0.9868 ± 0.0009 | 0.9710 ± 0.0163 | 0.0517 ± 0.0030 | 0.7752 ± 0.0128 | 0.6052 ± 0.0347 | 0.9697 ± 0.0169 |
## CICIDS2017 → CICDDoS2019 cross (σ=0.6, λ=0.3)
| Score | AUROC | AUPRC | F1 (P95) | Prec (P95) | Recall (P95) | FPR (P95) | F1 (P99) | TPR@1%FPR | TPR@5%FPR |
|---|---|---|---|---|---|---|---|---|---|
| `terminal_norm` | 0.9109 ± 0.0032 | 0.8974 ± 0.0047 | 0.6321 ± 0.0513 | 0.9545 ± 0.0045 | 0.4745 ± 0.0550 | 0.0441 ± 0.0011 | 0.4202 ± 0.0171 | 0.2685 ± 0.0139 | 0.4940 ± 0.0399 |
| `terminal_flow` | 0.9197 ± 0.0036 | 0.8957 ± 0.0086 | 0.6324 ± 0.0585 | 0.9517 ± 0.0055 | 0.4762 ± 0.0639 | 0.0469 ± 0.0019 | 0.4028 ± 0.0049 | 0.2534 ± 0.0039 | 0.4776 ± 0.0636 |
## Reading
- **Within-dataset (CICDDoS2019)**: at τ=P95, `terminal_norm` reaches F1 ≈ 0.99 with precision ≈ 0.99 and recall ≈ 0.99 — saturation. At τ=P99 (≈1% FPR), F1 ≈ 0.91 / TPR@1%FPR ≈ 0.90. The model is a working detector at fixed thresholds, not just an AUROC artifact.
- **Cross-dataset (CICIDS2017 → CICDDoS2019)**: AUROC stays high (≈ 0.91) but at fixed thresholds Precision is high (≈0.95) and Recall drops to ≈0.50 at P95 / ≈0.27 at 1% FPR. The cross-dataset domain shift compresses the score gap, so a source-calibrated threshold is conservative on target — false positives stay low, but a substantial fraction of target-domain attacks score below the source benign P95. **AUROC alone overstates deployability cross-dataset; thresholded numbers are the honest figure.**
- TIPSO-GAN comparability: TIPSO-GAN's CIC-DDoS2019 F1 ≈ 0.99 is reported under a **supervised** protocol (model has seen attack examples). Our F1 ≈ 0.99 on CICDDoS2019 within is achieved under the **unsupervised** protocol (benign-only training, threshold from benign-val), which is the strictly harder setting. Direct F1 numerical equivalence; protocol asymmetry is in our favor.
## Source artifacts
- `artifacts/verify_2026_04_24/thresholded_metrics.py` — per-file metric tool.
- `artifacts/verify_2026_04_24/aggregate_thresholded.py` — this aggregator.
- Within: `artifacts/phase1_2026_04_25/cicddos2019_lambda0p3_seed*/thresholded_metrics.json` (computed from existing `phase1_scores.npz`).
- Cross: `artifacts/phase25_sigma06_cross_2026_04_25/with_scores/thresholded_seed*.json` (raw scores re-saved by patched `eval_phase2_cross_cicddos2019.py`).