You can deploy this model and start earning money today!

mars2titan/
trace-c-conv-ae-baseline

Name: trace-c-conv-ae-baseline
Availability: InStock

Views

No views yet

TRACE-C convolutional autoencoder baseline

Status: BASELINE. Paper comparison autoencoder — not TRACE-C.

This is not TRACE-C. It is the paper's post-hoc conv-AE comparison baseline (keras-io 1D time-series autoencoder, ported to PyTorch, seed 42, 40 epochs, reconstruction MSE). TRACE-C itself is an untrained rank-calibrated detector: github.com/mars-arch/trace-c.

Evidence dataset: mars2titan/trace-c-neso-benchmark.

runs/trace-c-baselines/ — TensorBoard tfevents (loss curves, weight histograms, comparison scalars). Open the Training metrics tab on this model page, or:

tensorboard --logdir runs/trace-c-baselines --host 127.0.0.1 --port 6006
ae-training-history.json — committed loss curves and weight snapshots used to generate those logs (scripts/log_tensorboard.py in the GitHub repo).

Fit cutoff: Jan–Apr 2019 NESO streams only. Scoring is causal. The baseline protocol is not identical to TRACE-C (no daily budget, global standardization). See baselines-report.json in the dataset repo.

Code MIT (Matthew Faucher). Underlying telemetry: NESO Open Data Licence; Supported by National Energy SO Open Data.

mars2titan/trace-c-conv-ae-baseline

TRACE-C convolutional autoencoder baseline

Contents

mars2titan/
trace-c-conv-ae-baseline