p95 0.2.0

Python SDK for p95 ML experiment tracking

p95 Python SDK

Track ML experiments locally. No server setup required.

Installing the Python package includes the binary for the TUI and the web server.

Install

pip install p95

Usage

from p95 import Run

with Run(project="my-project", name="experiment-1") as run:
    run.log_config({"learning_rate": 0.001, "epochs": 10})

    for epoch in range(10):
        loss = train_one_epoch()
        run.log_metrics({"loss": loss}, step=epoch)

View Results

# Opens a dashboard at http://localhost:6767
pnf serve --logdir ~/.p95/logs

# Opens the TUI
pnf tui --logdir ~/.p95/logs

API

run.log_metrics({"loss": 0.5, "accuracy": 0.85}, step=epoch)  # Log metrics
run.log_config({"lr": 0.001})                                  # Log config
run.add_tags(["baseline"])                                     # Add tags

Environment Variables

Variable	Description	Default
P95_LOGDIR	Where to save logs	~/.p95/logs

License

MIT