⚗️ DoE interactive Bayesian optimizer

New DoE session

Define the mixture ingredients (with concentration ranges and prices) and the responses you measure per experiment. Everything is saved to the session file, so you can close the browser and resume anytime. Looking for earlier work? Open the session manager.

Ingredients (factors)

namemin conc.max conc.price / unit

Responses (measured criteria)

Goal & costs

Next experiments

Manual entry — an experiment you ran on your own

Best so far

GP prediction with 95% band, sliced through the best experiment (dotted line) one factor at a time. Dots are all measured experiments — their other factors vary, so they need not lie on the curve. ◆ marks the model's predicted optimum.

Predicted response over every ingredient pair, the other ingredients held at the best experiment (★). Lower maps: GP mean — same color scale as the measured dots in the upper scatters, so a dot that clashes with the map around it disagrees with the model. Diagonal: the 1D slices with 95% band. ◆ predicted optimum, ✕ open proposals.

Factor influence (ARD lengthscales)

Response correlation (learned task covariance)

Response vs. cost per experiment

Experiment history

⬇ CSV