ML Experiment Cockpit: read your MLflow, Optuna and W&B runs without leaving your IDE
Reading an ML experiment usually means leaving your editor. You open a browser tab for the MLflow UI, a second one for the W&B dashboard, maybe a notebook cell to call mlflow.search_runs() and print a metric curve. Each of those is a context switch: a different window, a different login, a different rendering of the same numbers. ML Experiment Cockpit is a free, Apache-2.0 JetBrains plugin that…
ML Experiment Cockpit is a JetBrains plugin that allows you to read your MLflow, Optuna and W&B runs without leaving your IDE. It installs as a free, Apache-2.0 plugin for PyCharm, IntelliJ IDEA, WebStorm, GoLand and CLion. The plugin connects to servers you already run and renders their answers, never writing anything back.
There are four tabs in the plugin: Experiments, Optuna, Models, and Sweeps. The Experiments tab shows the MLflow experiment tree, including experiments, runs, their status, name and parameters. Selecting a run opens a detail pane where you can pick a metric and view its history as a curve, including the last value and the step it was recorded at.
The Optuna tab displays studies and trials, retrieved from a running optuna-dashboard. It supports multi-objective studies, but only displays the first objective. The Models tab shows the MLflow Model Registry, with registered models and their versions, including the stage each version is in. The Sweeps tab lists W&B sweeps by state, with their run count.
To use the plugin, install it in your JetBrains IDE or grab it from the Marketplace page. Then, open Settings | Tools | ML Experiment Cockpit and fill in the required information, such as MLflow tracking server URL, Optuna dashboard URL, and W&B API key. The plugin does not start any servers on its own.
When you click Refresh in the Experiments tab, the tree fills with your experiments. Selecting a run opens a detailed pane with parameters and metrics. You can pick a metric from the drop-down, and the plugin fetches its full history and displays the curve, labeled with the last value and its step.
The plugin's code for getting run metrics, MlflowClient.getRunMetrics(runId): Map<String, Double>, demonstrates its read-only nature. It uses a strict JSON parser to avoid silently turning diverged runs into flat 0.0 values on the chart. The plugin does not include any dependencies like mlflow-client-java, instead using JDK's HttpClient and a small hand-rolled JSON parser.
ML Experiment Cockpit is a read-only plugin, designed to show you the status of your runs, metrics, models, and sweeps without allowing any changes. It connects to a tracking server you already run, such as MLflow, Optuna or W&B, and does not require local installations or additional dependencies.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
