Skip to main content

How to See an Evaluation's Metrics and Plots in the MLflow UI

· 6 min read
Jagdish Kumawat
Founder @ Dewiride

mlflow.models.evaluate puts its results in two places in the MLflow UI: on the run it logged to, with the plots, and on the page of the model it evaluated. This guide finds both for a tumour classifier, and explains the one oddity you'll see, a +1 next to each metric's model, on MLflow 3.18.0.

How an Evaluation Shows Up​

An evaluation writes to a run: its metrics, and for a classifier five plots on the run's Artifacts tab. Each metric is also linked to the evaluated model, so the model's page lists it with the dataset it was measured on and the run that measured it.

Prerequisites​

  • An MLflow tracking server at http://127.0.0.1:8080. MLflow's tracking server guide shows how to start one.
  • mlflow and scikit-learn in your project, for example with uv add mlflow scikit-learn pandas.

Step 1: Run an Evaluation​

The example logs a logistic regression named tumour-model and evaluates it, in a run named tumour-classifier:

The ui_evaluation.py script in a macOS-style code window: the breast cancer classifier logged as tumour-model in a run named tumour-classifier, then evaluated on the test rows with targets malignant and model_type classifier.

ui_evaluation.py
import mlflow
from sklearn.datasets import load_breast_cancer
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler

mlflow.set_tracking_uri("http://127.0.0.1:8080")
mlflow.set_experiment("evaluation-ui-demo")

X, y = load_breast_cancer(return_X_y=True, as_frame=True)
y = y == 0 # True when the tumour is malignant
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=0)
model = make_pipeline(StandardScaler(), LogisticRegression()).fit(X_train, y_train)

with mlflow.start_run(run_name="tumour-classifier"):
info = mlflow.sklearn.log_model(model, name="tumour-model", input_example=X_train.head(3))
mlflow.models.evaluate(
info.model_uri,
X_test.assign(malignant=y_test),
targets="malignant",
model_type="classifier",
)
Terminal
uv run python ui_evaluation.py
Output (last lines)
2026/11/21 17:55:31 INFO mlflow.models.evaluation.default_evaluator: Testing metrics on first row...
2026/11/21 17:55:32 WARNING mlflow.models.evaluation.evaluators.shap: SHAP or matplotlib package is not installed, so model explainability insights will not be logged.
🏃 View run tumour-classifier at: http://127.0.0.1:8080/#/experiments/13/runs/09fd9ccabc6a490a8298c5bca12d0754
🧪 View experiment at: http://127.0.0.1:8080/#/experiments/13

How to Evaluate a Classification Model with mlflow.models.evaluate explains the call and its output.

Step 2: See the Metrics on the Run​

Open the View run link, or the experiment evaluation-ui-demo and the run tumour-classifier. Its Overview lists the 13 metrics:

The run's Metrics (13) table: score, true_negatives 87, false_positives 3, false_negatives 3, true_positives 50, example_count 143, accuracy_score, recall_score, precision_score, f1_score, log_loss, roc_auc 0.9914 and precision_recall_auc, each with tumour-model +1 in the Models column.

The Models column links each metric to tumour-model, and shows +1 after it. Clicking +1 lists tumour-model a second time: the same model. MLflow 3.18.0's evaluate logs every metric twice with the model's ID, once from the evaluator and once more after it, and the UI counts both. The values are the same. I reported it to MLflow as issue #28013.

Step 3: Open the Plots​

Click the Artifacts tab. It lists the five plots; click one to open it, here roc_curve_plot.png:

The run's roc_curve_plot.png on the Artifacts tab: its path, then the ROC curve rising almost straight to a true positive rate of 1.0 at a low false positive rate, with AUC=0.991 in the legend and a dashed diagonal for chance.

The other four are the normalized confusion matrix, the precision-recall curve, the lift curve and the calibration curve. A plot taller than the viewer scrolls inside it; the download icon at the top right saves the file.

Step 4: See the Metrics on the Model's Page​

Go back to the experiment, open Models in the left menu, and click tumour-model. Its Metrics table lists the evaluation's metrics with the dataset and the run:

The model's Metrics (26) table: score, true_negatives, false_positives, false_negatives, true_positives, example_count, accuracy_score and recall_score, each with the dataset dataset (#9d2b530…) and the source run tumour-classifier.

  • Dataset: dataset, the name MLflow gives a plain DataFrame, with a short form of its digest. How to Name the Dataset of an MLflow Evaluation gives it a real name.
  • Source run: the evaluation's run, one click away.
  • Metrics (26): the 13 metrics, each twice, for the reason in Step 2.

Common Mistakes​

MistakeWhat happensDo this instead
Reading +1 as a second modelIt's the same model, linked twiceClick +1 to see it
Looking for a regressor's plots on the Artifacts tabA regressor gets plots only from SHAPInstall shap, or add your own plot

Troubleshooting​

SymptomLikely causeFix
The model's page lists each metric twiceMLflow 3.18.0 logs evaluation metrics twiceNothing to fix: the values are the same
The Artifacts tab is empty after an evaluationA regressor without SHAP, or the evaluation failedCheck the run's status and the script's output

FAQ​

Can I compare two evaluations side by side?​

Yes: select both runs on the experiment's Runs page and click Compare: How to Compare Runs in the MLflow UI.

Where are the plots stored?​

On the tracking server, under the run's artifacts folder: the path at the top of each file.

Conclusion​

The evaluation's 13 metrics are on the run's Overview, its five plots on the Artifacts tab, and the same metrics on the model's page with the dataset and the run that measured them, each counted twice in MLflow 3.18.0.

Additional Resources​