trackit-train 0.1.0

A lightweight local-first ML experiment tracker

TrackIt

A lightweight, local-first ML experiment tracker. No servers, no accounts, no configuration.

TrackIt is a simpler alternative to MLflow and Weights & Biases for tracking experiments during local development.

---

Features

• Zero configuration — install and start tracking in 30 seconds
• Local-first — all data stored in a single SQLite file
• No accounts — no remote server, no sign-up
• Experiment SDK — log params and metrics with a clean Python API
• CLI — list experiments, find best runs, launch a dashboard
• Web dashboard — interactive HTMX + Chart.js UI for exploring results

---

Installation

pip install trackit

For development:

pip install trackit[dev]

---

Quick Start

Log an experiment

from trackit import Experiment

exp = Experiment()

exp.log_param("lr", 0.001)
exp.log_param("batch_size", 32)

exp.log_metric("accuracy", 0.94)
exp.log_metric("loss", 0.12)

exp.finish()

Use as a context manager

from trackit import Experiment

with Experiment(name="resnet50-augmented") as exp:
    exp.log_param("lr", 0.001)
    for epoch in range(10):
        exp.log_metric("accuracy", evaluate(epoch))
        exp.log_metric("loss", compute_loss(epoch))

---

CLI

List experiments

trackit list

Output:

┌─────────────────────────────────────────────────────────────────┐
│                        Experiments                              │
├────┬──────────────────┬──────────┬───────────────────┬────────┬─────────┤
│ ID │ Name             │ Status   │ Created           │ Params │ Metrics │
├────┼──────────────────┼──────────┼───────────────────┼────────┼─────────┤
│  1 │ baseline         │ finished │ 2025-01-15 10:30  │      2 │       4 │
│  2 │ tuned            │ finished │ 2025-01-15 11:00  │      2 │       4 │
└────┴──────────────────┴──────────┴───────────────────┴────────┴─────────┘

Find the best run

trackit best accuracy

To minimize (e.g. loss):

trackit best loss --minimize

Launch the web dashboard

trackit ui

Open http://localhost:8042 in your browser.

---

Storage

All experiment data is stored in a single SQLite database at:

.trackit/experiments.db

This file lives in your project root. Add it to .gitignore if you don't want to commit experiment history.

---

Dashboard

The web dashboard provides:

• Experiment list with status, param count, and metric count
• Click-to-detail view showing all params and metrics
• Chart.js visualizations for accuracy and loss curves
• "Find Best Run" lookup for any metric

---

API Reference

Experiment

Experiment(name: str | None = None, db_path: str | None = None)

Method	Description
log_param(key, value)	Log a named parameter (stored as string)
log_metric(key, value, step=None)	Log a named metric; step auto-increments per key
finish()	Mark the experiment as finished

Experiment also supports with statement for automatic finish on exit.

---

Development

git clone <repo>
cd trackit
pip install -e ".[dev]"
pytest

---

License

MIT