This is the online repository of *Predictive Comment Updating with Heuristics and AST-Path-Based Neural Learning: A Two-Phase Approach*, a research paper under review by TSE. We release the source code and relevant data of Toper, the data used in our evaluation, as well as the experiment results. Dataset Basically, the dataset is from Liu et al.'s ASE20 paper (i.e., Automating Just-In-Time Comment Updating), and then cleaned by Lin et al.'s ICPC21 paper (i.e., Automated Comment Update: How Far are We?). We classify the dataset into code-indicative and non-code-indicative items and store them in Data directory, which is named by the format of [data catagory]_Items_[Dataset].jsonl. For example, All_Item_Test.jsonl means this file includes all (i.e., including code-indicative and non-code-indicative) items in the test set. Similarly, NCIU_Items_Test.json means this file only covers non-code-indicative items in the test set. The Code-Indicative Update Classifier We design a classifier to differentiate the Code-Indicative and Non-Code-Indicative updates. The replication package is available at Code/TypeClassifier.py. To obtain the result of the classifier, please run the following command: python3 TypeClassifier.py -training/FilePath FeaturesForClassifier/featuresForTrain.csv -testFilePath FeaturesForClassifier/featuresForTest.csv Operation Path Extractor The customized tool for extracting operation path from the dataset is provided by previous studies. To obtain the preprocessed data, run the following command: java -cp OperationPathExtractor.jar Extractor.App --data_dir path/to/data --input_name semi-finished/data/path --output_name path/to/store/data --num_threads 1 The preprocessed data are stored in Data/Preprocessed. The Non-Code-Indicative Comment Updater Our replication code is available at Code, and the detail instructions of command are at comment_update.py. Or you can simply execute the following command: python3 comment_update.py -data_path path/to/data -gpu -use_features