Compare Every Model Candidate, Export Any of Them
Choosing between Random Forest, XGBoost, and several Neural Network search strategies by hand means training each one separately and comparing notes. Data Sensors' Model Competition trains all of them concurrently on the same measurement data, ranks them by held-out validation accuracy, and lets you export any entry from the leaderboard - not just the winner - to C, C++, Python, or MATLAB.
Train several model types concurrently, not one at a time
Pick any combination of Random Forest, XGBoost, and Neural Network competitors, each with its own search strategy - random search, grid search, Bayesian (Gaussian process) optimization, or a Random Forest/XGBoost surrogate search - and they all train at once against the same input/output channels. Auto-tuning isn't limited to Neural Networks: Random Forest and XGBoost competitors can each auto-tune their own hyperparameters too, so a fair, tuned comparison doesn't mean manually searching three different hyperparameter spaces by hand. There's no need to re-open the training dialog for each candidate, remember which configuration you tried last, or keep a spreadsheet of results on the side.
Ranked by held-out validation accuracy, not guesswork
Every competitor is scored on data it never trained on, so the leaderboard reflects how each model actually generalizes, not just how well it memorized the training set. The leaderboard shows validation RMSE and R² side by side for every entry, along with training time and a fit diagnosis, so you can see at a glance whether a candidate is underfitting, overfitting, or genuinely competitive.
A genuine, untouched test split for larger datasets
A leaderboard's own ranking has a small amount of optimism baked in, since it's used to pick a winner. For datasets large enough to afford it, Data Sensors automatically carves out a real, held-out test split before any ranking happens - never touched during training or model selection - so you get a second, more trustworthy set of accuracy numbers for the model you actually plan to ship.
Export the winner - or any other entry
The top-ranked model isn't always the right choice - maybe a slightly less accurate Random Forest is faster to run on your target hardware, or a smaller Neural Network fits your flash budget better. Every entry on the leaderboard, not just the winner, can be exported to C, C++, Python, or MATLAB, in full precision or int8-quantized, with a live size and accuracy preview before you save. See Virtual Sensor Training for training a single model directly, without running a full competition.
