Multi-AutoML Interface v5.3.0

Multi-AutoML Interface 5.3.0

Added

• ONNX export and SHAP explanations now ship in the installers. They were never in
  requirements.txt, so the desktop app - which installs exactly that file - could not run the
  🧠 Explain Prediction or 📦 Export to ONNX buttons at all. The lock now carries
  onnx, onnxruntime, skl2onnx, onnxconverter-common and shap (plus numba, llvmlite,
  slicer, tqdm, flatbuffers, ml-dtypes), with shap excluded on Intel macOS, where the
  numba version it allows cannot take the numpy==2.5.0 pin. Windows and Linux resolve; the
  packaging workflow verifies the macOS build.

Fixed

• export_to_onnx never wrote a model. It called to_onnx(model, input_sample[:1], ...),
  which makes skl2onnx treat every column as a separate input, so even a plain
  RandomForestClassifier raised InvalidInputLengthException; the export now names one
  FloatTensor of the sample's width and the artifact loads and predicts through onnxruntime.
  The Experiments button also handed over FLAML's AutoML wrapper where the engine had passed the
  inner estimator - the wrapper is unwrapped now. Boosted-tree learners (lgbm, xgboost,
  catboost) genuinely have no converter in skl2onnx, so that case raises a message naming the
  estimator instead of a warning logged inside a training thread nobody reads; FLAML's default
  learner is lgbm, which is why the feature looked like it worked and did nothing.
• The tabular SHAP path could not be imported without OpenCV. src/xai_utils.py had
  import cv2 at module scope while only the saliency-map function uses it (and imports it
  there), so Explain Prediction failed on any interpreter without opencv-python.

What is inside

Each installer bundles a standalone CPython 3.12 with everything in
requirements.txt already installed, so no Python setup is needed on
the target machine. Runs, models and the data lake are written to the app's
per-user workspace (the Electron userData directory):

Operating system	Location
Windows	%APPDATA%\multi-automl-desktop\workspace\
macOS	~/Library/Application Support/multi-automl-desktop/workspace/
Linux	~/.config/multi-automl-desktop/workspace/

The heavy AutoML backends (AutoGluon, PyCaret, TPOT, Lale, H2O, AutoKeras)
stay optional and are lazy-imported; the bundled runtime
contains the core stack (Streamlit, MLflow, FLAML, scikit-learn, XGBoost,
LightGBM, ONNX export with skl2onnx, SHAP explanations), so those features work
in the desktop app out of the box. The heavy engines stay optional: the
installers offer FLAML until you install whichever
engines you need into it. H2O additionally requires Java 11+.

Signing

These builds are not code-signed or notarized, so SmartScreen and Gatekeeper
will warn on first launch.