Create Experiment with Configuration File

This section will introduce how to create SwanLab experiments using configuration files in json or yaml format.

Load Configuration File with swanlab.config

The config parameter of swanlab.init supports passing the path of a configuration file in json or yaml format, and parses the configuration file into a dictionary for experiment creation.

Using json File

Below is an example of a configuration file in json format:

{
    "epochs": 20,
    "learning-rate": 0.001
}

Pass the path of the configuration file to the config parameter, it will parse the configuration file into a dictionary:

swanlab.init(config="swanlab-init-config.json")
# Equivalent to swanlab.init(config={"epochs": 20, "learning-rate": 0.001})

Using yaml File

Below is an example of a configuration file in yaml format:

epochs: 20
learning-rate: 0.001

Pass the path of the configuration file to the config parameter, it will parse the configuration file into a dictionary:

swanlab.init(config="swanlab-init-config.yaml")
# Equivalent to swanlab.init(config={"epochs": 20, "learning-rate": 0.001})

Load Configuration File with swanlab.init

The load parameter of swanlab.init supports passing the path of a configuration file in json or yaml format, and parses the configuration file for experiment creation.

Using json File

Below is an example of a configuration file in json format:

{
    "project": "cat-dog-classification",
    "experiment_name": "Resnet50",
    "description": "My first AI experiment",
    "config": {
        "epochs": 20,
        "learning-rate": 0.001
    }
}

Pass the path of the configuration file to the load parameter, it will parse the configuration file to initialize the experiment:

swanlab.init(load="swanlab-config.json")
# Equivalent to
# swanlab.init(
#     project="cat-dog-classification",
#     experiment_name="Resnet50",
#     description="My first AI experiment",
#     config={
#         "epochs": 20,
#         "learning-rate": 0.001
#     }
# )

Using yaml File

Below is an example of a configuration file in yaml format:

project: cat-dog-classification
experiment_name: Resnet50
description: My first AI experiment
config:
  epochs: 20
  learning-rate: 0.001

Pass the path of the configuration file to the load parameter, it will parse the configuration file to initialize the experiment:

swanlab.init(load="swanlab-config.yaml")
# Equivalent to
# swanlab.init(
#     project="cat-dog-classification",
#     experiment_name="Resnet50",
#     description="My first AI experiment",
#     config={
#         "epochs": 20,
#         "learning-rate": 0.001
#     }
# )

FAQ

1. Is the configuration file naming fixed?

The naming of the configuration file is free, but it is recommended to use swanlab-init and swanlab-init-config as the configuration names.

2. What is the relationship between the parameters in the configuration file and the script?

The priority of the parameters in the script is higher than that in the configuration file, that is, the parameters in the script will override the parameters in the configuration file.

For example, there is a yaml configuration file and a code snippet below:

project: cat-dog-classification
experiment_name: Resnet50
description: My first AI experiment
config:
  epochs: 20
  learning-rate: 0.001

swanlab.init(
    experiment_name="resnet101",
    config={"epochs": 30},
    load="swanlab-init.yaml"
)

The final experiment_name is resnet101, and config is {"epochs": 30}.