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>
89 lines
2.7 KiB
Python
89 lines
2.7 KiB
Python
"""Shared scaffolding for sequence models.
|
|
|
|
All four sequence models (CNN, GRU, LSTM, Transformer) follow the same
|
|
input/output contract:
|
|
|
|
Input: (B, n_channels_keep, n_timesteps) float32
|
|
Output: (B, n_classes) float32 logits
|
|
|
|
This module factors out the common BaseModel boilerplate so each
|
|
architecture file only declares its torch.nn.Module.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
from typing import Any
|
|
|
|
import numpy as np
|
|
|
|
from training.models._base import BaseModel, StandardizeStats
|
|
|
|
|
|
class _SeqBase(BaseModel):
|
|
"""Composition wrapper: a torch.nn.Module under self._mod plus the
|
|
BaseModel interface (select, predict, predict_proba, save_sidecar).
|
|
Subclasses override _build_module(self, **cfg) -> nn.Module."""
|
|
|
|
input_kind = "tensor"
|
|
|
|
def __init__(
|
|
self,
|
|
*,
|
|
n_channels_in: int,
|
|
n_timesteps: int,
|
|
n_classes: int,
|
|
keep_mask: np.ndarray,
|
|
standardize: StandardizeStats,
|
|
device: str = "cpu",
|
|
**arch_config,
|
|
) -> None:
|
|
self.n_classes = n_classes
|
|
self.keep_mask = keep_mask.astype(bool)
|
|
self.standardize = standardize
|
|
self.config = {
|
|
"n_channels_in": n_channels_in,
|
|
"n_timesteps": n_timesteps,
|
|
**arch_config,
|
|
}
|
|
self._device = device
|
|
self._mod = self._build_module(
|
|
n_channels_in=n_channels_in,
|
|
n_timesteps=n_timesteps,
|
|
n_classes=n_classes,
|
|
**arch_config,
|
|
).to(device)
|
|
|
|
@property
|
|
def module(self):
|
|
return self._mod
|
|
|
|
def _build_module(self, **cfg):
|
|
raise NotImplementedError
|
|
|
|
def predict_proba(self, X: np.ndarray) -> np.ndarray:
|
|
import torch
|
|
Xk = self.select(X) # (N, C_keep, T) float32
|
|
self._mod.eval()
|
|
with torch.no_grad():
|
|
t = torch.from_numpy(Xk).to(self._device)
|
|
logits = self._mod(t)
|
|
return torch.softmax(logits, dim=-1).cpu().numpy()
|
|
|
|
def state_for_checkpoint(self) -> dict[str, Any]:
|
|
return {"state_dict": self._mod.state_dict(), "config": self.config}
|
|
|
|
@classmethod
|
|
def from_checkpoint(cls, header: dict, payload: dict, *,
|
|
device: str = "cpu") -> "_SeqBase":
|
|
cfg = dict(payload["config"])
|
|
n_ch = cfg.pop("n_channels_in")
|
|
n_t = cfg.pop("n_timesteps")
|
|
m = cls(
|
|
n_channels_in=n_ch, n_timesteps=n_t,
|
|
n_classes=int(header["n_classes"]),
|
|
keep_mask=np.asarray(header["keep_mask"], dtype=bool),
|
|
standardize=StandardizeStats.from_dict(header["standardize"]),
|
|
device=device,
|
|
**cfg,
|
|
)
|
|
m._mod.load_state_dict(payload["state_dict"])
|
|
return m
|