Objective 2 - placeholder
Model runs
Metadata for predictive models trained on FoodProt plant-protein data. Training itself happens outside the app - in Python notebooks, GCP Vertex AI, RStudio, Molegro Data Modeler, or wherever the team is most productive. Each completed run gets recorded back here for reproducibility and side-by-side comparison.
| Model family / name | Target | Dataset / split | Encoding / missing | Optimizer | RMSE | PCC | R² | Status |
|---|---|---|---|---|---|---|---|---|
| Loading... | ||||||||
Proposal model families
Modelling approaches the proposal explicitly calls out. Use these as values in the model_family column when recording a run.
Regression Transformer
Transformer adapted for tabular regression
XLNet adapted model
Permutation language model adapted for tabular features
Traditional regression
Linear / ridge / lasso baselines
PLSR
Partial least-squares regression
SVM
Support vector machine regression
Random Forest
Tree ensemble baseline
Gradient Boosting
XGBoost, LightGBM, or scikit-learn GBR
Neural Network
MLP or domain-specific architectures
RStudio regression model
R workflow regression export
Molegro Data Modeler
Molegro export imported into FoodProt
External training is the design. FoodProt stores metadata, predictions, metrics, and visualizations - not training infrastructure. See model results for predicted vs. actual values per run, and model-ready datasets to export training data.