1fabd4a246
The model layer of the project, built honestly:
- tools/dataset_validate.py — full-sweep validator over the receiver
store (sha256, schema, monotonic labels, telemetry-row gate). On the
current corpus: 64,798 accepted + 8,154 degraded + 3,701 rejected +
7 errored across 76,660 shipped episodes. data/processed/validation_v1.parquet
is committed as the per-episode acceptance index.
- training/_features.py — channel registry (46 channels across
proc/guest/qmp/netflow), summary-stat windowing AND channel×time
tensor extraction at 10s/5s windowing. Time alignment uses t_wall_ns
(Unix ns) — tested fix for a real netflow-vs-host clock-base
inconsistency that was silently dropping every netflow channel.
- training/_split.py — three held-out recipes (host / sample / time)
with profile-stratification assertions. held_out_host carries
untested_profiles for cases like scan-and-dial absent from the test
host (5 of 6 profiles tested cross-device, never silently averaged).
- training/models/ — 6 architectures behind a common BaseModel
interface: gbt (XGBoost), mlp, cnn, gru, lstm, transformer. Each
trained twice (realistic / oracle) per the deployment threat model.
Schema-hashed checkpoints refuse to load if _features.py changed
since training (silent-input-drift protection, tested).
- training/trainer/ — unified training loop: class-weighted CE, LR
warmup + cosine, gradient clipping, mixed precision when CUDA,
early stopping on val macro F1, best-on-val checkpoint. Same loop
runs MLP/CNN/GRU/LSTM/Transformer; GBT uses XGBoost
early_stopping_rounds on val mlogloss.
- training/eval_/ — bootstrap 95% CIs on macro F1, per-class F1,
per-profile and per-host breakdown, paired-bootstrap significance
for model-vs-model gap. Confusion matrix uses union of seen labels.
- training/dashboard/producers/ — replay/metrics/perf/profiles
emitting the six event types the dashboard's awaiting scenes
consume; on-demand tensor extraction so the Pi can run live
inference without 65 GB of shards.
- 17 unit tests (split coverage, features round-trip, schema mismatch,
determinism, time-base alignment regression).
End-to-end smoke-trained all six on a 567-episode subset; held-out
test macro F1 reported with paired-bootstrap significance. The
methodology now reports honest cross-device generalization, not
in-distribution validation.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
41 lines
1.3 KiB
Bash
Executable file
41 lines
1.3 KiB
Bash
Executable file
#!/usr/bin/env bash
|
|
# Pull training data from the receiver Pi to a local trainer box.
|
|
#
|
|
# Run this on the trainer (e.g. the Windows/2070-Super box via WSL or a
|
|
# Linux desktop). Requires WireGuard up to 10.100.0.1 with `cis490-trainer`
|
|
# enrollment so SSH key auth works.
|
|
#
|
|
# What gets pulled:
|
|
# /var/lib/cis490/episodes/ raw .tar.zst episode tarballs (~3GB)
|
|
# /var/lib/cis490/index.jsonl shipped-episode index
|
|
# data/processed/validation_v1.parquet validator output (committed in repo)
|
|
#
|
|
# Once those are local you can run:
|
|
# uv run --group training python training/build_features.py \
|
|
# --validation data/processed/validation_v1.parquet \
|
|
# --store ./episodes \
|
|
# --out-dir data/processed
|
|
#
|
|
# Then training/train_gbt.py and training/train_nn.py.
|
|
set -euo pipefail
|
|
|
|
PI_HOST="${PI_HOST:-10.100.0.1}"
|
|
PI_USER="${PI_USER:-max}"
|
|
LOCAL_DIR="${LOCAL_DIR:-./episodes}"
|
|
|
|
mkdir -p "${LOCAL_DIR}"
|
|
|
|
echo "→ rsyncing episodes from ${PI_USER}@${PI_HOST}:/var/lib/cis490/episodes/"
|
|
rsync -ah --info=progress2 \
|
|
--exclude='*.partial' \
|
|
"${PI_USER}@${PI_HOST}:/var/lib/cis490/episodes/" \
|
|
"${LOCAL_DIR}/"
|
|
|
|
echo "→ rsyncing index.jsonl"
|
|
rsync -a --info=progress2 \
|
|
"${PI_USER}@${PI_HOST}:/var/lib/cis490/index.jsonl" \
|
|
"${LOCAL_DIR}/index.jsonl"
|
|
|
|
echo "done. ${LOCAL_DIR} contains:"
|
|
du -sh "${LOCAL_DIR}"
|
|
ls "${LOCAL_DIR}/" | head
|