-
Notifications
You must be signed in to change notification settings - Fork 0
FAQ
Q: What Python versions are supported? A: Python 3.10, 3.11, 3.12, and 3.13.
Q: Does it work on Windows/Mac/Linux? A: Yes, all platforms are supported. CI runs on Ubuntu.
Q: How do I silence all output?
A: scomp_link.set_verbosity("silent") or --silent flag in CLI.
Q: How does model selection work? A: A decision tree based on data size, feature types, and task type selects the optimal algorithm. See Model Selection.
Q: Can I use my own model?
A: Yes. ScompArtifact works with any sklearn-compatible estimator. Just call artifact.set_model(your_model).
Q: Does it support GPU? A: PyTorch and TensorFlow are included as dependencies. They'll use GPU if available and configured.
Q: What file formats are supported? A: CSV, TSV, and Parquet for input. CSV, Parquet, JSON, and HTML for output.
Q: Does it handle missing values?
A: The Preprocessor handles basic cleaning. For detailed profiling, use DataQualityReport.
Q: Can I use polars DataFrames?
A: Yes. The Preprocessor accepts both pandas and polars DataFrames.
Q: How do I deploy a model to production?
A: Save with ScompArtifact.save(), ship the .scomp file, load with ScompArtifact.load(), predict with artifact.predict(X).
Q: How do I monitor a model in production?
A: Use DriftDetector with the artifact's sample data as reference. Compare against incoming production data regularly.
Q: Can I compare model versions?
A: Yes. scomp-link compare --artifacts v1.scomp v2.scomp shows side-by-side metrics.
Q: ModuleNotFoundError: No module named 'tf_keras'
A: Upgrade to scomp-link ≥ 1.1.1 which includes tf-keras as a dependency for Python 3.12+.
Q: Tests fail with SHAP on Python 3.12+
A: Same fix — ensure tf-keras is installed: pip install tf-keras.
Q: The CLI command is not found after install
A: Make sure pip install scomp-link was done in the active environment. Check with which scomp-link or scomp-link --version.