TokEye

TokEye is a open-source Python-based application for automatic classification and localization of fluctuating signals. It is designed to be used in the context of plasma physics, but can be used for any type of fluctuating signal.

Check out this preprint for more information.

Example Demonstration

Expected processing time:

• V100: < 0.5 seconds on any size spectrogram after warmup.
• CPU: ~5-10 seconds.

Quickstart

pip install 'tokeye[app]'   # web app + CLI   (or: uv tool install 'tokeye[app]')
tokeye app                  # opens web app on http://localhost:7860

• The default model downloads automatically from Hugging Face on first use (~30 MB).
• No data on hand? Click "Load Example Signal" in the app, or generate one from the shell with tokeye example.
• pip install requires Python >= 3.13; uvx/uv tool install fetch a compatible Python automatically.

Zero-install trial: uvx 'tokeye[app]' app runs the app without installing anything into your environment.

Install variants

Heavy dependencies are split into extras so you download only what you use:

Install	What you get
pip install tokeye	Python API + CLI (tokeye run, mode-analysis suite). Smallest core install.
pip install 'tokeye[app]'	+ the Gradio web app (tokeye app).
pip install 'tokeye[train]'	+ the training/ablation pipeline.
pip install 'tokeye[eigspec]'	+ tokeye eigspec clustering.

GPU vs CPU PyTorch. A plain install pulls the default PyTorch wheel — the CUDA (GPU) build (~2.5 GB) on Linux. On a machine without a GPU, install the CPU wheel (~200 MB) instead:

uv pip install 'tokeye[app]' --torch-backend=cpu

Python API

To use TokEye inside your own program, import the TokEye class:

import numpy as np
from tokeye import TokEye

eye = TokEye()  # loads the default model (auto-downloads on first use)

mask = eye(signal)             # 1D time series → STFT → inference
mask = eye(spectrogram)        # 2D spectrogram → inference directly
coherent, transient = mask     # (2, H, W) sigmoid scores in [0, 1]

Input is auto-detected by shape: a 1D array is treated as a raw time series (TokEye computes the spectrogram), a 2D array as a ready spectrogram. Standardization happens internally.

If your 2D spectrogram is stored in linear scale (raw STFT magnitude/power), pass log=True so TokEye applies log1p first since the model expects log-scaled input:

mask = eye(linear_spectrogram, log=True)      # per call
eye = TokEye(log=True)                        # or for every call

log is off by default and ignored for 1D inputs (the STFT already log-scales). Everything is configurable through the constructor, but the defaults just work:

eye = TokEye(
    model="big_tf_unet",   # registry name or path to a local .pt/.pt2
    device="auto",         # "cpu", "cuda", or "auto"
    n_fft=1024, hop=256,   # STFT settings (1D inputs only)
    clip_dc=True, clip_low=1.0, clip_high=99.0,
    log=False,             # log1p for linear-scale 2D spectrograms
)

Batch processing (CLI)

For headless / scripted use (no browser needed), run inference directly. For example:

tokeye run "files/*.npy" --output-dir results

INPUT arguments can be files, directories (all *.npy files inside are used), or quoted glob patterns. Each input is interpreted by its shape:

• 1D array — a raw time series. TokEye computes its STFT spectrogram using the flags below before running inference.
• 2D array — a precomputed spectrogram, fed to the model directly.

For each input file, tokeye run writes:

• <stem>_mask.npy — float32 array, shape (2, H, W), sigmoid scores per pixel (channel 0 = coherent, channel 1 = transient).
• <stem>_preview.png — a grayscale spectrogram with the mask overlaid (green = coherent, red = transient), unless --no-png is passed.

The process exit code is the number of files that failed.

Flags:

Flag	Default	Description
--model	big_tf_unet	Registry name or path to a .pt/.pt2 checkpoint.
--output-dir	tokeye_output	Directory for masks and previews.
--n-fft	1024	STFT window size (1D inputs only).
--hop	256	STFT hop size (1D inputs only).
--keep-dc	off	Keep the DC bin (dropped by default).
--clip-low / --clip-high	1.0 / 99.0	Percentile clip bounds applied to the spectrogram.
--log	off	Apply log1p to 2D spectrogram inputs stored in linear scale (1D signals are always log-scaled during the STFT).
--threshold	0.5	Mask threshold used only for the preview PNG overlay.
--no-png	off	Skip preview PNGs; write masks only.
--device	auto	cpu, cuda, or auto.

The released model was trained on spectrograms built with hop=128; for closest match to the training configuration use --hop 128.

On HPC clusters where compute nodes have no internet access, pre-fetch the weights on the login node, then run the batch job on the compute node:

tokeye download big_tf_unet   # on the login node; prints the cached path
tokeye run ... --model big_tf_unet   # on the compute node — model is already cached

Mode-analysis suite

Beyond segmentation, tokeye bundles the analyses DIII-D researchers usually reach for separate tools to get. Each is a subcommand; --help on any of them shows the full flags.

Command	What it does
tokeye modespec <config.yaml>	Classic Mirnov mode analysis (vendored pymodespec, the Python port of the IDL modespec tool): power spectrograms, matched-filter toroidal mode-number fits, per-shot mode CSVs. Data fetch needs MDSplus (GA cluster / conda-forge) or a local cache; an example config ships at src/tokeye/modespec/classic/modes.yaml.
tokeye elmspec INPUTS...	ELM detection from the segmentation model's transient channel: per-event time intervals plus per-shot count, ELM frequency (with --fs), and duty cycle, written to elm_events.csv / elm_summary.csv.
tokeye alfvenspec INPUTS...	Alfvén-eigenmode detection with the ae_tf_maskrcnn instance model: per-detection boxes/scores (ae_detections.csv) and instance masks. Wide spectrograms are processed in training-width windows automatically.
tokeye eigspec [SCRIPT]	Interactive modal identification and spectral analysis (vendored eigspec, the Python port of the MATLAB toolbox): stochastic subspace ID, AR/PCA, random-projection spectral analysis, clustering (clustering needs pip install tokeye[eigspec]).
tokeye modesearch	Design stage — prints the plan for a searchable database of detected modes.

The suite roadmap (including the next-generation modespec --engine deep) lives in docs/ROADMAP.md.

Web app guide

tokeye app (or python -m tokeye.app) launches a Gradio interface with three tabs:

• Analyze — load a signal, compute its spectrogram, run a model, and visualize the result. Guided for first-time use: the model dropdown defaults to the bundled big_tf_unet model, the STFT transform has working defaults, and "Load Example Signal" generates a synthetic demo signal so a brand-new user needs zero files. "Analyze" runs the whole load-model → infer → visualize pipeline in one click. View modes: Original, Enhanced (percentile-clipped amplitude), Mask (thresholded model output), Amplitude.
• Annotate — manually draw and save mask annotations over a read-only backdrop image.
• Utilities — audio-format conversion and .npy file inspection.

Flags: tokeye app [--port 7860] [--share] [--open] — --share creates a public Gradio link, --open opens a browser tab on launch.

If you're on a remote server (e.g. an HPC login node), forward the port over SSH instead of using --share:

ssh -L 7860:localhost:7860 user@remote

Then open http://localhost:7860 in your local browser.

Verified Datatypes

• DIII-D Fast Magnetics (cite)
• DIII-D CO2 Interferometer (cite)
• DIII-D Electron Cyclotron Emission (cite)
• DIII-D Beam Emission Spectroscopy (cite)

Evaluation

Recall Scores:

• TJII2021: 0.8254
• DCLDE2011 (Delphinus capensis): 0.7708
• DCLDE2011 (Delphinus delphis): 0.7953

With more data, comes better models. Please contribute to the project!

Installation (from source / development)

uv is the dev tool for this repo:

git clone git@github.com:PlasmaControl/TokEye.git
cd TokEye
uv sync                # core deps (default GPU/CUDA PyTorch on Linux)
uv sync --extra app    # + Gradio web app (`tokeye app`)
uv sync --dev          # + pytest, ruff, etc.
uv sync --group train  # + training deps (lightning, h5py, etc.)

Additionally, run these the first time

uv run pre-commit install
uv run ruff check .
uv run pytest

This creates a .venv/; activate it with source .venv/bin/activate, or prefix commands with uv run.

Models

Registry name	HF repo	HF file	Description
big_tf_unet	nc1/big_tf_unet	big_tf_unet_251210.pt	Transformer U-Net trained on multiscale (multiwindow, multihop) spectrograms.
ae_tf_maskrcnn	nc1/ae_tf_maskrcnn	ae_tf_maskrcnn_251223.pt	Mask R-CNN instance detector for Alfvén-eigenmode activity (used by tokeye alfvenspec).

Weights download automatically the first time a registry name is used (cached in ~/.cache/huggingface). Override the default repo with the TOKEYE_HF_REPO environment variable (per-model repos are fixed in the registry).

To use a local checkpoint instead, put .pt/.pt2 files in a model/ directory (picked up by the app's model dropdown) or pass a path directly via --model PATH.

Input should be a tensor that has shape (B, 1, H, W) where B, H, and W can vary Output will be a tensor of shape (B, 2, H, W)

Best performance when spectrograms are oriented so that when they are plotted with matplotlib, the lowest frequency bin is oriented with the bottom when origin='lower'. Spectrograms should be standardized (mean = 0, std = 1). If baseline activity is very strong, clipping the input may help, but is generally not needed.

The first channel of the output will return preferential measurements of coherent activity (useful for most tasks) The second channel of the output will return preferential measurements of transient activity

Data

Keep signals as 1D numpy float arrays (raw time series). No need to normalize or preprocess them. The CLI also accepts 2D arrays (precomputed spectrograms) directly. The app scans a signal directory for .npy files (default data/input, configurable in the Analyze tab).

Bringing your own data takes two lines:

import numpy as np

signal = ...  # any 1D float array: tokamak diagnostic, hydrophone, etc.
np.save("shots/myshot.npy", signal)

tokeye run shots/myshot.npy --output-dir results

No data yet? tokeye example writes a synthetic demo signal you can run immediately, and the web app has a matching "Load Example Signal" button.

Citation

If you use this code in your research, please cite:

@article{chen_TokEye_2026,
  title={TokEye: Fast Signal Extraction for Fluctuating Time Series via Offline Self-Supervised Learning From Fusion Diagnostics to Bioacoustics},
  author={Chen, Nathaniel},
  year={2026},
  publisher={ArXiv},
  doi={10.48550/arXiv.2602.20317},
  url={https://www.arxiv.org/abs/2602.20317}
}

Contact

Nathaniel Chen — nathaniel [at] princeton [dot] edu — https://nathanielchen.net