ATAT (Astronomical Transformer for time series And Tabular data) implements Transformer-based models for classifying astronomical sources from light-curve time series and tabular metadata. It provides scripts to download, process, and partition survey data, extracting and normalizing light curves and assembling metadata into training, validation, and test sets. Specialized layers for time modulation and multi-head attention encode light curves and tabular features, while training drivers construct models, run optimization and validation loops, and save trained configurations. ATAT also provides utilities to compute classification metrics at multiple observation durations and generate visualizations such as performance curves and confusion matrices.​

The ALeRCE anomaly detector cross-validates six anomaly detection algorithms for three classes (transient, periodic, and stochastic) of anomalous sources within the Zwicky Transient Facility (ZTF) data stream using the ALeRCE light curve features. A machine and deep learning-based framework is used for anomaly detection. For each class, a distinct anomaly detection model is constructed using only information about the known objects (i.e., inliers) for training. An anomaly score is computed using the probabilities to determine whether the light curve corresponds to a transient, stochastic, or periodic nature.