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
1.7 KiB
Python
41 lines
1.7 KiB
Python
"""Gated Recurrent Unit over channel × time windows.
|
||
|
||
Sees the window one timestep at a time and accumulates state. Cheaper
|
||
than LSTM, often comparable on short sequences. Last-step output → linear.
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
from training.models import register
|
||
from training.models._torch_seq import _SeqBase
|
||
|
||
|
||
@register("gru")
|
||
class GRU(_SeqBase):
|
||
def _build_module(self, *, n_channels_in: int, n_timesteps: int,
|
||
n_classes: int, hidden: int = 128, n_layers: int = 2,
|
||
dropout: float = 0.1, bidirectional: bool = False):
|
||
from torch import nn
|
||
return _GRUClassifier(n_channels_in=n_channels_in, n_classes=n_classes,
|
||
hidden=hidden, n_layers=n_layers,
|
||
dropout=dropout, bidirectional=bidirectional)
|
||
|
||
|
||
from torch import nn # noqa: E402
|
||
|
||
|
||
class _GRUClassifier(nn.Module):
|
||
def __init__(self, *, n_channels_in: int, n_classes: int, hidden: int,
|
||
n_layers: int, dropout: float, bidirectional: bool):
|
||
super().__init__()
|
||
self.gru = nn.GRU(
|
||
input_size=n_channels_in, hidden_size=hidden,
|
||
num_layers=n_layers, dropout=dropout if n_layers > 1 else 0.0,
|
||
batch_first=True, bidirectional=bidirectional,
|
||
)
|
||
d_out = hidden * (2 if bidirectional else 1)
|
||
self.head = nn.Sequential(nn.Dropout(dropout), nn.Linear(d_out, n_classes))
|
||
|
||
def forward(self, x): # x: (B, C, T)
|
||
x = x.transpose(1, 2) # → (B, T, C)
|
||
out, _ = self.gru(x) # (B, T, hidden*dirs)
|
||
return self.head(out[:, -1, :]) # last timestep
|