tb_hydrograph

A TensorBoard plugin for interactive hydrograph visualization. Log observed vs. simulated time series with per-step metrics and browse them as an interactive Plotly chart inside TensorBoard.

Install

From PyPI-style wheel (once built):

uv pip install tensorboard_plugin_hydrograph

From source:

uv pip install .

Editable install for development:

uv pip install -e .

Build

This project uses uv with the hatchling build backend declared in pyproject.toml.

uv build

This produces both an sdist and a wheel under dist/. The plugin's static assets (tensorboard_plugin_hydrograph/static/*) are bundled into the wheel.

Usage

Log a hydrograph from your training loop:

from torch.utils.tensorboard import SummaryWriter
from tensorboard_plugin_hydrograph import add_hydrograph

writer = SummaryWriter("runs/exp1")
add_hydrograph(
    writer,
    tag="Hydrographs/Comparison",
    dates=dates,           # iterable of datetime-like values
    observed=observed,     # array-like
    simulated=simulated,   # array-like
    step=epoch,
    metrics={"NSE": 0.81}, # optional
)

Then launch TensorBoard against the same log directory:

tensorboard --logdir runs/

The Hydrograph tab will appear once the plugin detects logged summaries.

Development

uv venv
uv pip install -e .