lycokie / swanlab

🧐SwanLab: track and visualize all the pieces of your machine learning pipeline. 跟踪与可视化你的机器学习全流程

SwanLab

Track and visualize all the pieces of your machine learning pipeline

👀 View Document | 👋 Join our WeChat

English | 中文

Table of Contents

• Changelog
• Key Function
• Installation
• Getting Started
• Tutorial
• Use Case
• LICENSE

Changelog

[24/03/19] 🔧 We fixed some issues, and prepare for the upcoming cloud version.（v0.2.4）

[24/03/12] 👽 We supported the line chart smoothing feature, supporting 3 different smoothing algorithms; we support comparing Images and Audio charts in Charts; we have improved swanlab.Image, supporting tensor as inputs.（v0.2.3）

[24/03/03] 🔧 We fixed some issues in v0.2.1 and added support for switching multimedia chart content by pressing keys.（v0.2.2）

[24/03/01] 🚀 Still Very Big Upgrade！We supported Text Chartto meet the needs of tasks in NLP, LLM, Agent and other scenarios; We have made a lot of optimizations to line chart's UI, legend, and rendering speed. We improved the rendering performance of Logs —— now viewing terminal printing information with 200k lines is also not lagging.（v0.2.1）

[24/02/08] 🔥 Very Big Upgrade! We supported Image Chart、Audio Chart、Multi-Experiments Chart and a series of comprehensive optimizations and improvements! Please be sure to upgrade to the latest version via pip install -U swanlab.（v0.2.0）

Full Changelog

Upgrade to the latest version: pip install -U swanlab。

Key Function

• 📊 Visualize Training: visualize your entire training process.

• 🚀 Multimedia Charts: log Image/Audio/Video/Text/object3D...

• 🧪 Experiments GridView: compare your key metrics for inspiration faster.

• ⚡️ Auto Save Environment Info: Hyperparameters, Config, Metric, Terminal logging, Environment Information auto save.

• 🥔 Offline Support: SwanLab can run entirely offile, ithout requiring any access to the Internet. For instance, this may be on your local machine, behind a corporate firewall, or in a datacenter.

Installation

with pip

Requirement：Python 3.8+.

Use pip to install our stable release version. The installation command is as follows:

pip install -U swanlab

with source code

If you need the bleeding edge of the code and can't wait for a new release, you must install the library from source。

Getting Started

1. First, install the SwanLab SDK with pip:

pip install -U swanlab

2. Second, Use the example code snippet below as a template to integrate SwanLab to your Python script:

import swanlab

# Start a SwanLab Run with swanlab.init
swanlab.init(
  # Save model inputs and hyperparameters in a swanlab.config object
  config={'epochs': 20, 'learning_rate': 0.01}, 
  # Specify the save path of the log file
  logdir="./logs",
)

# Model training code here...

# Log metrics over time for visualizing performance with swanlab.log
for epoch in range(1, swanlab.config.epochs):
    swanlab.log({"loss": loss})

For example, We wrote a simulation experiment script:

import swanlab
import random

offset = random.random() / 5

run = swanlab.init(
    experiment_name="Example",
    description="It's a ML experiment.",
    config={
        "learning_rate": 0.01,
        "epochs": 20,
    },
    logdir="./logs"
)

# Simulate the ML training process
for epoch in range(2, run.config.epochs):
    acc = 1 - 2**-epoch - random.random() / epoch - offset
    loss = 2**-epoch + random.random() / epoch + offset
    swanlab.log({"loss": loss, "accuracy": acc})

3. Third, open the terminal and use the following command to start a SwanLab dashboard:

$ swanlab watch -l ./logs

SwanLab will give you a URL link (default is http://127.0.0.1:5092). Check the link to see the visualization results of your first experiment in the browser.

Tutorial

Get Started

• Installation
• Quick Start
• Run Dashboard

Python API

• init
• log
• config
• Image
• Audio
• Text

CLI API

• watch

技巧

• Remotely view Dashboard
• Pass argparse into swanlab.config

Use Case

Learn how to use SwanLab more effectively by following these use cases:

Code Cases	Description
Hello World	Getting Started
MNIST	Handwriting recognition based on a plain net and MNIST dataset with pytroch, swanlab.
Image Classification	Cat and dog classification based on ResNet50 with pytorch, swanlab and gradio. Tutorial.
Text Generation	Text generation based on Word_language_model (RNN/LSTM/GRU/Transformer)
UIE-Finetune	how to use personal data to finetune UIE model and monitor training process through swanlab

LICENSE

This project is currently licensed under Apache 2.0 License.