SwanLab is the next-generation machine learning experiment management and visualization tool released by the SwanHub, designed to facilitate effective collaboration in machine learning training.
SwanLab provides streamlined APIs that make it easy to track machine learning metrics and record configurations. Additionally, SwanLab provides a visual dashboard for the most intuitive way to monitor, analyze, and compare your training.
For specific guidance on SwanLab's features, please refer to the User Guide.
Currently, SwanLab is undergoing rapid iterations and will continue to add new features.
This repository is tested on Python 3.8+.
SwanLab can be installed using pip as follows:
pip install swanlab
Let's simulate a simple machine learning training process, using swanlab.init
to initialize the experiment and record configuration information, and using swanlab.log
to track key metrics (in this example, it's loss
and acc
):
import swanlab
import time
import random
lr = 0.01
epochs = 20
offset = random.random() / 5
# Initialize the experiment and record configuration information
swanlab.init(
description="This is a sample experiment for machine learning training.",
config={
"learning_rate": lr,
"epochs": epochs,
},
)
# Simulate a machine learning training process
for epoch in range(2, epochs):
acc = 1 - 2**-epoch - random.random() / epoch - offset
loss = 2**-epoch + random.random() / epoch + offset
print(f"epoch={epoch}, accuracy={acc}, loss={loss}")
# Track key metrics
swanlab.log({"loss": loss, "accuracy": acc})
time.sleep(1)
During the program running, a swanlog
folder will be generated in the directory to record your training data.
If you want to visualize your experiment, open the terminal and enter the root directory (no need to enter the swanlog
folder), and run the following command:
swanlab watch
If you see the following output, it means that the experimental board is running successfully:
[SwanLab-INFO]: SwanLab Experiment Dashboard ready in 375ms
ā Local: http://127.0.0.1:5092
Accesshttp://127.0.0.1:5092
at this time to enter the experiment dashboard and browse your experimental resultsļ¼