GaMPEN (Galaxy Morphology Posterior Estimation Network) estimates robust posteriors (i.e., values + uncertainties) for structural parameters of galaxies using a Bayesian machine learning framework. The code also automatically crops input images to an optimal size before structural parameter estimation. The package produces extremely well-calibrated (less than 5% deviation) predicted posteriors; these have been shown to be up to 60% more accurate compared to the uncertainties predicted by many light-profile fitting algorithms. Once trained, it takes GaMPEN less than a millisecond to perform a single model evaluation on a CPU. Thus, GaMPEN’s posterior prediction capabilities are ready for large galaxy samples expected from upcoming large imaging surveys, such as Rubin-LSST, Euclid, and NGRST.

GaMorNet classifies galaxies morphologically using a Convolutional Neural Network. The code does not need a large amount of training data, as it is trained on simulations and then transfer-learned on a small portion of real data, and can be applied on multiple datasets. The software has a misclassification rate of less than 5%. GaMorNet is written in Python and uses the Keras and TFLearn deep learning libraries to perform all of the machine learning operations. Both these aforementioned libraries in turn use TensorFlow for their underlying tensor operations.