nnlogging 0.2.0a1

A powerful and elegant logging library designed specifically for neural network and machine learning experiments.

nnlogging

nnlogging is a logging extension designed for machine learning experiments and competitions.

nnlogging is mostly built on Rich, DuckDB and DVC.

Features

• Modern Logging Record: With pre-configured but customizable rich settings, nnlogging handles logs as modern and structured records.

• Extendable Tracking History: nnlogging leverages duckdb to store experiment metadata and data (tags, artifacts, ...).

• Run Commit Solution: nnlogging treats every run of any experiments as a git commit point.

Installation

pip install nnlogging  # Python >= 3.10

Quick Start

The new nnlogging API is designed for simplicity. You can get a pre-configured, global logger instance and start logging right away.

Basic Logging

Add a branch to the root logger and start logging messages.

import nnlogging

# 0. Configure logger levels
nnlogging.configure_logger([None], level="DEBUG")

# 1. Add a console branch to root logger
nnlogging.add_branch(("console", "stdout"))

# 2. Log messages!
nnlogging.info(__name__, "Starting training...")
nnlogging.warning(__name__, "Learning rate seems high.")
nnlogging.debug(__name__, "This is a detailed debug message.")
nnlogging.error(__name__, "CUDA out of memory.")

Experiment Tracking

Configure experiment tracking to store metrics, hyperparameters, and more in DuckDB tables.

import uuid
import nnlogging

# 1. Configure the experiment run
nnlogging.configure_run(experiment="resnet_training", uuid=uuid.uuid4(), run="run_1")

# 2. Log hyperparameters
nnlogging.add_hparams({"lr": 0.001, "batch_size": 32, "model": "ResNet50"})

# 3. Track metrics in your training loop
for epoch in range(10):
    train_loss = 1.0 / (epoch + 1)
    nnlogging.track(step=epoch, metrics={"train_loss": train_loss})
    nnlogging.info(None, f"Epoch {epoch}: Loss={train_loss:.4f}")

# 4. Close run to tell when the run is finished
nnlogging.close_run()

Progress Tracking

Add tasks to display and update rich progress bars.

import time
import nnlogging

# 0. Add at least 1 branch to show progress
nnlogging.add_branch(("stderr", "stderr"))

# 1. Add a task to the progress display
nnlogging.add_task("training", total=100)

# 2. Update the task during your training loop
for step in range(100):
    time.sleep(0.01)
    nnlogging.advance("training", 1)

# Tasks will be recycled when finished

Contributing

Contributions are welcome! Please feel free to open an issue or submit a pull request.

License

This project is licensed under the MIT License.