cleanUrl: "comet-usage"
description: "MLOps 도구인 Comet의 사용법을 정리합니다."

Comet

https://www.comet.ml/

무료로 딥러닝 실험 결과 및 training 상황을 저장하고, 비교하고 관리할 수 있게 해 주는 툴.

Installation

mamba install -c comet_ml comet_ml

Define experiment

from comet_ml import Experiment
experiment = Experiment(
	api_key='XAHbm......',
	project_name='my_project',
	workspace='dohlee',
)

log_parameters

hyperparameters = {
	'learning_rate': 0.05,
	'batch_size': 128,
}
experiment.log_parameters(hyperparameters)

log_metric

experiment.log_metric('train_loss', running_loss / show_every, step=step)

log_histogram_3d

experiment.log_histogram_3d(net.state_dict()['fc.9.weight'].cpu(), step=step)