Hey Martin, it's really weird, but running this same code, you added here, still leads to these empty figures in the "Plots" section of the clearML experiment.

We decided now to use the "report_image"-option which at least somehow displays the plots in the "debug samples"-section of the experiment.

Thank you very much for your help 🙏

"plt" comes from matplotlib.pyplot and as I understand the clearML documentation, matplotlib plots are logged automatically. In other scripts, this works just fine but not with these SHAP-plots that are just displayed as empty plots:

another though, see what happens if you remove the .save and .close and stay with the .show, maybe the close call somehow interfere's with it ?
Otherwise, if you can test one of the shaps examples and see maybe they fail in your setup that's another avenue to take for reproducing the issue

AgitatedDove14 can you please tell me which SHAP-, matplotlib-, and clearML-versions you are using? And would it be possible to add a title to that plot you are tracking or would plt.title() again destroy something there?

Hey, good day and thank you for your quick replies! So this is the code snippet I was using to create the plots (see appended image). I also tried removing the plt.savefig() part or the plt.show() part or manually adding the report_matplotlib_figure-part for the task but nothing seems to make a difference.

Hmm Could you check if it makes a difference importing ClearML before shap ?
If this changes nothing, could you put a standalone script to reproduce the issue ?

This should work:

from clearml import Task

task = Task.init(project_name="examples", task_name="shap example")
import xgboost
import shap

# train an XGBoost model
X, y = shap.datasets.california()
model = xgboost.XGBRegressor().fit(X, y)

# explain the model's predictions using SHAP
# (same syntax works for LightGBM, CatBoost, scikit-learn, transformers, Spark, etc.)
explainer = shap.Explainer(model)
shap_values = explainer(X)

# visualize the first prediction's explanation
shap.plots.waterfall(shap_values[0])

Notice that in your example you have

plt.figure()

This actually clears the matplotlib figure, this is why we are getting a first white image then the actual plot,
once I removed it I got a single plot (no need for the manual reporting)

X, y = make_regression(
    n_samples=100,     # Number of samples
    n_features=10,     # Number of features
    noise=0.1,   # Number of informative features
    random_state=42    # For reproducibility
)

# Convert to DataFrame for better feature naming
feature_names = [f"Feature {i+1}" for i in range(X.shape[1])]
X = pd.DataFrame(X, columns=feature_names)

# Step 2: Split Data into Train/Test Sets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Step 3: Train a Simple Model
model = RandomForestRegressor(random_state=42)
model.fit(X_train, y_train)

# Step 4: Use SHAP to Explain Model Predictions
explainer = shap.Explainer(model, X_train)
shap_values = explainer(X_test)  # Returns a SHAP Explanation object in new SHAP versions

shap_plot = shap.summary_plot(shap_values, X_test)  # Show top 5 features

plt.show()

Hey Martin, unfortunately the adapted import order did not change anything. Thank you anyways 🙂 I will try to provide some standalone code so the issue can be reproduced

I think latest:

clearml==1.17.0
matplotlib==3.6.2
shap==0.46.0

Python 3.10

Hey AgitatedDove14 , thank you for your help. Unfortunately it did not help to remove the "close". I also created a Dummy-Notebook to reproduce the issue. But it does not seem to work. My plot in the clearML dashboard is empty, unfortunately. I also tried, not using plt at all and uploading the shap-plot as is:

# Create the Beeswarm Plot
shap_plot = shap.summary_plot(shap_values, X_test)  # Show top 5 features

clearml_task.get_logger().report_matplotlib_figure(title="Shap beeswarm plot", series="My SHAP Plots", iteration=0, figure=shap_plot, report_interactive=True)

But this just leads to an empty plot even without a title (see image).

Could it be connected to the version of clearML I use? So our team, right now, works with the Community version - might it only work in the Pro version?

Hi BlandCormorant75 , how are you logging those plots? Can you provide a stand alone snippet that reproduces your behaviour?