kt-masterviz 0.1.0

Live Streamlit dashboard for kt-masterlog master CSV files. Companion to kt-masterlog.

kt-masterviz

Live Streamlit dashboard for kt-masterlog master CSV files. Companion package — separate install so you don't pay the dashboard's dependency footprint when you only want the logger.

What it does

Point kt-masterviz at any master CSV produced by kt-masterlog (or any CSV that follows the schema: trial_id, epoch, hyperparameter columns, metric columns) and get a live Streamlit dashboard showing:

• Trial summary — one row per trial, sorted by the objective metric, with hyperparameter values inline.
• Training curves — per-trial loss/accuracy lines, switchable across any detected metric.
• Raw data — the full DataFrame, for ad-hoc filtering.

The reader is file-safe: tolerant of partial last rows from a writer mid-append, so you can launch the dashboard against a CSV that's still being written by a running tuner.

Install

With pip:

pip install kt-masterviz

With uv:

uv add kt-masterviz

Usage

kt-masterviz path/to/master_log.csv

Opens the dashboard at http://localhost:8501. The page auto-refreshes every few seconds (configurable in the sidebar) so trials show up as the tuner writes them.

The CSV doesn't need to exist when you launch — the dashboard polls and renders a waiting state until the writer produces the first row. Useful for starting the viewer before kicking off a long sweep.

Programmatic use

If you'd rather build your own viewer, the loader is the reusable bit:

from kt_masterviz import load_master_csv, detect_hyperparameter_columns

df = load_master_csv("runs/sweep_master_log.csv")
hp_cols = detect_hyperparameter_columns(df)
# ... your own pandas / plotly / altair code

CSV schema

kt-masterviz works with any CSV that has:

Column	Required	Purpose
trial_id	yes	Identifies a single tuning trial
epoch	yes	Epoch number within the trial
loss / val_loss / accuracy / etc.	no (auto-detected)	Training/validation metrics
anything else	no	Treated as a hyperparameter or extra field

Detection of "metric vs hyperparameter" is heuristic — a column counts as a metric if its lowercased name contains any of: loss, accuracy, acc, auc, mae, mse, rmse, f1, precision, recall, iou, dice. Everything else is treated as a hyperparameter or static field (and shown in the trial summary).

Requirements

• Python 3.12 (tested; 3.13+ may work but is not yet verified)
• pandas ≥ 2.0
• streamlit ≥ 1.28

No dependency on kt-masterlog itself — kt-masterviz reads its file format but doesn't import its code.

License

MIT.