Limits & FAQ

Current limits and frequently asked questions

Current Limits

FeatureLimit
CSV upload size10,000 rows max
Training hardwareCPU only (GPU coming soon)
Model outputSingle output (binary classification or scalar regression)
OptimizersAdam, AdamW, SGD, RMSprop
Loss functionsCross-Entropy, MSE, MAE, Huber
Experiments saved50 per workspace
Canvas blocksNo hard limit (performance may degrade at 50+)
Browser storage~5 MB localStorage per project

Frequently Asked Questions

Can I use my own GPU?

Not yet. Training currently runs on your browser's CPU via the backend server. GPU support is on the roadmap.

What happens to my data?

CSV data is sent to the backend for training only. It is not stored permanently. Sessions and experiments are saved in your browser's localStorage.

Can I export trained models?

Yes. After training, you can export the generated PyTorch code, including model definition and trained weights. Use the "Export Code" button in the training results.

What file formats are supported for import?

CSV is the primary format. You can also import datasets from Kaggle, HuggingFace, UCI Machine Learning Repository, and direct URLs. See Data Sources for details.

Why did my formula fall back to logistic regression?

If the formula compiler can't parse your LaTeX, it falls back to a default logistic regression model. Check that your notation follows the conventions described in the Formulas reference.

Can I use MathExec for multi-class classification?

Yes, with the softmax classifier formula. Set the output layer to softmax and the loss to cross-entropy. The system auto-detects the number of classes from your target column.

Is there an API?

Not currently. MathExec is designed as an interactive workspace. An API for programmatic training is being considered for a future release.

How do I report a bug?

Use the chat widget in the app (press / to open) or email support.