mlflow

mlflow

MLflow api client for Node.js

Install

npm install mlflow

Usage

const MLflow = require('mlflow')
const {ViewType, RunState, LifecysleStage} = require('mlflow/enum')

const client = new MLflow({endpoint: 'http://localhost:5000'})

const {Experiments, Runs, Metrics, Artifacts} = client

;(async () => {
	
	// Get list of all experiments
	const {experiments} = await Experiments.list()
	const experiment = experiments[0]
	
	// Get list of all active runs in specific experiment
	const {runs} = await Runs.search({
		experiment_ids: [experiment.experiment_id],
		run_view_type: ViewType.ACTIVE_ONLY
	})
	const {info, data} = runs[0]
	const {run_id} = info
	
	// Get metric history
	const auc_history = await Metrics.getHistory({run_id, metric_key: 'auc_score'})
	console.log(auc_history)
	
})()

API

MLflow

Constructor

• endpoint:string (required)

  • MLflow server's endpoint url
  • ex.) 'http://localhost:5000'

• heders:object (option)

  • Key-Value pairs of RequestHeaders which will be passed for fetch method.
  • ex.) {Authorization: 'Bearer abcdefgh'}

Props

• Experiments
• Runs
• Metrics
• Artifacts

Experiments

• create()
• restore()
• get()
• list()
• update()
• delete()
• setExperimentTag()

Runs

• create()
• restore()
• get()
• search()
• update()
• delete()
• logMetric()
• logParam()
• setTag()
• deleteTag()
• logBatch()

Metrics

• getHistory()

Artifacts

• list()

For more details about each method's args, also see MLflow REST API Official Docs.
API's are basically implemented just as they are.