What you will be able to do
- Find an experiment and its runs from the Experiments page, the workspace, or a notebook's Experiment Runs sidebar
- Name what the run page shows: run ID, parameters, metrics, details with the source notebook, and tags
- Find logged artifacts and models, and the load-and-predict code snippets Databricks generates for them
Key concept
MLflow run — One execution of model code. Nearly everything the MLflow UI shows is attached to a run: its parameters, metrics, tags, artifacts and source notebook. Experiments are just containers that group related runs.
1.From the Experiments page down to a single run
The MLflow UI follows how tracking data is organised. An experiment is a collection of related runs, and each run is one execution of your training code. In the UI you start broad and narrow down: the Experiments page lists every experiment you can access, each experiment's details page lists its runs, and each run has its own page.
To open the Experiments page, go to the workspace sidebar and click Experiments under AI/ML. You can narrow the list with the Only my experiments checkbox or the Filter experiments search box, which matches text in the Name or Location columns. Click an experiment name to open its experiment details page. That page lists all of the experiment's runs in a table. Click a Run Name to open the run, or use the Source column to open the notebook version that created the run.
On the experiment details page, the information icon next to the experiment name opens a pop-up with the experiment's path, its experiment ID and its artifact location.
There are two kinds of experiment. A workspace experiment is not tied to any notebook, and any notebook can log to it by experiment ID or name. A notebook experiment belongs to one notebook and is created automatically when needed. Every run goes to the active experiment, and if none is set, it goes to the notebook experiment.
This gives notebook users a shortcut. Click the Experiment icon in the notebook's right sidebar to open the Experiment Runs sidebar. It shows a summary of each run, including its parameters and metrics. From the sidebar you can open a run directly or go to the full experiment details page.
Checkpoint 1 of 5· Check yourself
Which information does the Experiment Runs sidebar in a notebook show for each run?
The sidebar summarises each run in the notebook experiment, including its parameters and metrics. You can then click through to the full run page.
“The Experiment Runs sidebar appears and shows a summary of each run associated with the notebook experiment, including run parameters and metrics.”Source: docs.databricks.com
2.What the run page shows
The run page is the core of this objective. A run records the notebook that launched it, any models it created, parameters and metrics saved as key-value pairs, tags for run metadata, and any artifacts (output files). The run page shows that record: the run ID, the parameters used, the metrics produced, and a Details table with run information, including a link to the source notebook. Artifacts are on their own Artifacts tab.
The source link records exactly which code produced a result. Click Source on the run page, or the link in the Source column of the experiment details page, and the notebook version associated with the run opens. A highlight bar shows the date and time of the run, so you see the code as it was when the run executed, not as it is today.
| Information | What it is | Where you see it |
|---|---|---|
| Parameters | Key-value inputs to the run, such as max_depth | Run page; Experiment Runs sidebar |
| Metrics | Key-value results, such as rmse or auc | Run page; Experiment Runs sidebar |
| Tags | Key-value metadata you can search on later | Details table on the run page |
| Source | The notebook version that launched the run | Link next to Source on the run page; Source column of the experiment details page |
| Artifacts | Output files, including logged models | Artifacts tab of the run page |
You can add, edit and delete tags from the run page. In the Details table, click Add tags next to Tags. Enter a key and a value, click the plus sign to save the pair, repeat for more tags, then click Save tags. The same dialog lets you edit a tag's value or delete a tag with its X. Tags matter because you can search for runs by them later. You can also rename a run from the kebab menu at the upper right of the run page.
The run page describes one run. To look at several runs together, go to the experiment details page, check the boxes next to two or more runs and click Compare. The Comparing <N> Runs screen presents the selected runs in tabular format, with Parameters and Metrics tables showing the values from all selected runs. You can hide parameters and metrics that are identical in every run.
For charts, click the Chart view icon on the experiment details page, then Add chart. A parallel coordinates plot lets you pick parameters and metrics to investigate, which helps you see how parameter settings affect model performance, for example whether lower max_depth goes with a higher auc.
Checkpoint 2 of 5· Check yourself
A reviewer opens a run page and wants the exact code that produced a metric. What does the Source link open?
Source opens the notebook version tied to the run, not the latest edit. The highlight bar marks when the run happened.
“appears in the main window with a highlight bar showing the date and time of the run”Source: docs.databricks.com
Checkpoint 3 of 5· Exam question
A data scientist ran a Hyperopt search with `SparkTrials`, producing 50 child runs logged to one MLflow experiment. Each run has `learning_rate`, `max_depth`, and a validation metric `val_auc` logged. The data scientist wants to see, in one view, how the two hyperparameters and the metric move together across all 50 runs at once. What should they do in the MLflow UI?
Correct answer: A — Open the experiment's Compare Runs view, select the child runs, and choose the Parallel Coordinates Plot, mapping `learning_rate`, `max_depth`, and `val_auc` as axes so the relationship across the runs is visible in one chart.
- A. The Parallel Coordinates Plot in the Compare Runs view is built exactly for this: each vertical axis is a selected parameter or metric, and each run draws one line across the axes, so patterns between `learning_rate`, `max_depth`, and `val_auc` across all 50 runs are visible at once.
- B. The per-run Metrics tab charts a single metric's value against logging step for that one run only; it has no mechanism to overlay parameters or metrics from the other 49 runs, so it cannot show the requested cross-run relationship.
- C. The `MLmodel` file describes the model's flavor, signature, and loading instructions for a single logged model artifact; it does not contain hyperparameter or metric values from any run, let alone the full set of child runs.
- D. Sorting by creation time only reorders the rows of the run table; it does not add any chart or visual encoding that relates hyperparameters to the validation metric across runs.
3.Artifacts, logged models and generated code snippets
Every file a run saves (plots, data samples, model files) is on the run's Artifacts tab. A model logged from a run appears there too, along with code you can copy. When you log a model in a Databricks notebook, Databricks automatically generates snippets for loading the model and running predictions.
To see them, open the run page, scroll to the Artifacts section, and click the logged model's name. A panel opens on the right with code for loading the model and making predictions on Spark or pandas DataFrames.
The model appears in the run's Artifacts tab. Databricks generates code snippets that show how to load the logged model and make predictions on Spark and pandas DataFrames. Click the model name in the Artifacts section and a panel opens on the right with them.
MLflow 3 changes where models live. Models are now first-class objects in their own right, not just files inside a run. In the UI, open an experiment and select its Models tab. That page lists all of the experiment's Logged Models with their metrics, parameters and artifacts, and you can search and filter them by attributes, parameters, tags and metrics. Click a model name to open its model page, which shows its parameters and metrics plus details such as its source run and related datasets.
Checkpoint 4 of 5· Check yourself
In an MLflow 3 workspace, where do you see all of an experiment's logged models side by side, with their metrics, parameters and artifacts?
MLflow 3 tracks models as separate entities. The experiment's Models tab collects them in one place along with their metrics, parameters and artifacts.
“This page contains all the Logged Models associated with the experiment, along with their metrics, parameters, and artifacts.”Source: docs.databricks.com
Checkpoint 5 of 5· Exam question
A teammate trained a model with `mlflow.sklearn.log_model()` inside a run and now wants to confirm, from the MLflow UI alone, the exact input feature schema and output type the model expects before wiring it into a batch scoring job. Where in the UI should they look?
Correct answer: A — Open the run's Artifacts tab, expand the logged model directory, and read the `MLmodel` file's signature section, which records the input columns, their types, and the output type inferred when the model was logged.
- A. When a model is logged, MLflow writes an `MLmodel` file into the model artifact directory, and if a signature was inferred or supplied it records the expected input columns and types plus the output type, making the Artifacts tab the correct place to check the schema.
- B. Tags are free-form key-value labels a user or job sets on a run for organization and search; MLflow does not automatically populate a `schema` tag with input or output column information.
- C. The Parallel Coordinates Plot only visualizes the logged parameters and metrics selected for comparison across runs; it has no relationship to the input feature schema of a logged model artifact.
- D. Scikit-learn does not automatically log an `input_columns` parameter, and the Parameters tab reflects only what was explicitly logged with `mlflow.log_param`, so this is not where a model signature would appear.
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.If you don't set an experiment, the run is lost or fails to log.Why is that wrong?
MLflow logs it to the notebook experiment, which Databricks creates automatically. You'll find it in that notebook's Experiment Runs sidebar.
2.In MLflow 3, a logged model is still only a file under the run's Artifacts tab.Why is that wrong?
MLflow 3 makes models first-class objects, listed on the experiment's Models tab with their own metrics and parameters.
Covered in Artifacts, logged models and generated code snippets
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“The experiment details page lists all runs associated with the experiment.”
↩︎ From the Experiments page down to a single run“A pop-up note appears that shows the path to the experiment, the experiment ID, and the artifact location.”
↩︎ From the Experiments page down to a single run“Databricks automatically creates a notebook experiment if there is no active experiment when you start a run using mlflow.start_run().”
↩︎ Prediction“The Experiment Runs sidebar appears and shows a summary of each run associated with the notebook experiment, including run parameters and metrics.”
↩︎ Checkpoint - 2.
“You can select the Only my experiments checkbox or use the Filter experiments search box to filter the list of experiments.”
↩︎ From the Experiments page down to a single run“the model page, which contains information like the model's parameters and metrics, as well as details such as its source run”
↩︎ Artifacts, logged models and generated code snippets - 3.https://docs.databricks.com/aws/en/mlflow/runsOfficial docs
“The run screen shows the run ID, the parameters used for the run, the metrics resulting from the run”
↩︎ What the run page shows“model parameters and metrics saved as key-value pairs, tags for run metadata, and any artifacts, or output files, created by the run”
↩︎ What the run page shows“Tags are key-value pairs that you can create and use later to search for runs.”
↩︎ What the run page shows“Click Compare. The Comparing <N> Runs screen appears.”
↩︎ What the run page shows“If you log a model from a run, the model appears in the Artifacts tab”
↩︎ Artifacts, logged models and generated code snippets“An MLflow run corresponds to a single execution of model code.”
↩︎ Key concept“In MLflow 3, models are now their distinct first-class object rather than being logged as run artifact.”
↩︎ Exam trap 2“appears in the main window with a highlight bar showing the date and time of the run”
↩︎ Checkpoint - 4.
“A parallel coordinates plot is useful in understanding the effect of parameter settings on model performance”
↩︎ What the run page shows - 5.https://docs.databricks.com/aws/en/mlflow/modelsOfficial docs
“Databricks automatically generates code snippets that you can copy and use to load and run the model.”
↩︎ Artifacts, logged models and generated code snippets
Also cited
- https://docs.databricks.com/aws/en/mlflow/trackingOfficial docs
“If no active experiment is set, runs are logged to the notebook experiment.”
↩︎ Exam trap 1 “This page contains all the Logged Models associated with the experiment, along with their metrics, parameters, and artifacts.”
↩︎ Checkpoint