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>
32 lines
1.2 KiB
Python
32 lines
1.2 KiB
Python
"""1D-CNN over channel × time windows.
|
||
|
||
Three conv blocks + global average pooling. Small enough to fit on the
|
||
Pi for live inference, expressive enough to learn cross-channel patterns
|
||
the GBT baseline can't see.
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
from training.models import register
|
||
from training.models._torch_seq import _SeqBase
|
||
|
||
|
||
@register("cnn")
|
||
class CNN(_SeqBase):
|
||
def _build_module(self, *, n_channels_in: int, n_timesteps: int,
|
||
n_classes: int, ch1: int = 64, ch2: int = 128,
|
||
ch3: int = 128, dropout: float = 0.1):
|
||
from torch import nn
|
||
return nn.Sequential(
|
||
nn.Conv1d(n_channels_in, ch1, kernel_size=5, padding=2),
|
||
nn.BatchNorm1d(ch1), nn.GELU(),
|
||
nn.MaxPool1d(2), # T/2
|
||
nn.Conv1d(ch1, ch2, kernel_size=5, padding=2),
|
||
nn.BatchNorm1d(ch2), nn.GELU(),
|
||
nn.MaxPool1d(2), # T/4
|
||
nn.Conv1d(ch2, ch3, kernel_size=3, padding=1),
|
||
nn.BatchNorm1d(ch3), nn.GELU(),
|
||
nn.AdaptiveAvgPool1d(1), # → (B, ch3, 1)
|
||
nn.Flatten(),
|
||
nn.Dropout(dropout),
|
||
nn.Linear(ch3, n_classes),
|
||
)
|